saved

daemons/node/common/common_pubsub.go

@@ -166,8 +166,6 @@ func SubscribeEvents[T interface{}](s *LongLivedPubSubService,
 }
 
 func waitResults[T interface{}](s *LongLivedPubSubService, ctx context.Context, sub *pubsub.Subscription, proto string, timeout int, f func(context.Context, T, string)) {
-	fmt.Println("waitResults", proto)
-
 	defer ctx.Done()
 	for {
 		s.PubsubMu.Lock()                     // check safely if cache is actually notified subscribed to topic

daemons/node/common/common_stream.go

@@ -35,6 +35,10 @@ type LongLivedStreamRecordedService[T interface{}] struct {
 	AfterDelete func(pid pp.ID, name string, did string)
 }
 
+func (ix *LongLivedStreamRecordedService[T]) MaxNodesConn() int {
+	return ix.maxNodesConn
+}
+
 func NewStreamRecordedService[T interface{}](h host.Host, maxNodesConn int) *LongLivedStreamRecordedService[T] {
 	service := &LongLivedStreamRecordedService[T]{
 		LongLivedPubSubService: NewLongLivedPubSubService(h),
@@ -160,25 +164,26 @@ func (ix *LongLivedStreamRecordedService[T]) HandleHeartbeat(s network.Stream) {
 		// if record already seen update last seen
 		if rec, ok := streams[*pid]; ok {
 			rec.DID = hb.DID
-			if rec.HeartbeatStream == nil {
+			// Preserve the existing UptimeTracker so TotalOnline accumulates correctly.
+			// hb.Stream is a fresh Stream with no UptimeTracker; carry the old one over.
+			oldTracker := rec.GetUptimeTracker()
 			rec.HeartbeatStream = hb.Stream
+			if oldTracker != nil {
+				rec.HeartbeatStream.UptimeTracker = oldTracker
+			} else {
+				rec.HeartbeatStream.UptimeTracker = &UptimeTracker{FirstSeen: time.Now().UTC()}
 			}
-			rec.HeartbeatStream = hb.Stream
-			if rec.HeartbeatStream.UptimeTracker == nil {
-				rec.HeartbeatStream.UptimeTracker = &UptimeTracker{
-					FirstSeen: time.Now().UTC(),
-					LastSeen:  time.Now().UTC(),
-				}
-			}
+			rec.HeartbeatStream.UptimeTracker.RecordHeartbeat()
+			rec.LastScore = hb.Score
 			logger.Info().Msg("A new node is updated : " + pid.String())
 		} else {
-			hb.Stream.UptimeTracker = &UptimeTracker{
-				FirstSeen: time.Now().UTC(),
-				LastSeen:  time.Now().UTC(),
-			}
+			tracker := &UptimeTracker{FirstSeen: time.Now().UTC()}
+			tracker.RecordHeartbeat()
+			hb.Stream.UptimeTracker = tracker
 			streams[*pid] = &StreamRecord[T]{
 				DID:             hb.DID,
 				HeartbeatStream: hb.Stream,
+				LastScore:       hb.Score,
 			}
 			logger.Info().Msg("A new node is subscribed : " + pid.String())
 		}
@@ -215,30 +220,33 @@ func CheckHeartbeat(h host.Host, s network.Stream, dec *json.Decoder, streams ma
 	if err := dec.Decode(&hb); err != nil {
 		return nil, nil, err
 	}
-	_, bpms, _ := getBandwidthChallengeRate(h, s.Conn().RemotePeer(), MinPayloadChallenge+int(rand.Float64()*(MaxPayloadChallenge-MinPayloadChallenge)))
+	_, bpms, latencyScore, _ := getBandwidthChallengeRate(h, s.Conn().RemotePeer(), MinPayloadChallenge+int(rand.Float64()*(MaxPayloadChallenge-MinPayloadChallenge)))
 	{
 		pid, err := pp.Decode(hb.PeerID)
 		if err != nil {
 			return nil, nil, err
 		}
-		upTime := float64(0)
-		isFirstHeartbeat := true
+		uptimeRatio := float64(0)
+		age := time.Duration(0)
 		lock.Lock()
 		if rec, ok := streams[pid]; ok && rec.GetUptimeTracker() != nil {
-			upTime = rec.GetUptimeTracker().Uptime().Hours() / float64(time.Since(TimeWatcher).Hours())
-			isFirstHeartbeat = false
+			uptimeRatio = rec.GetUptimeTracker().UptimeRatio()
+			age = rec.GetUptimeTracker().Uptime()
 		}
 		lock.Unlock()
-		diversity := getDiversityRate(h, hb.IndexersBinded)
-		hb.ComputeIndexerScore(upTime, bpms, diversity)
-		// First heartbeat: uptime is always 0 so the score ceiling is 60, below the
-		// steady-state threshold of 75. Use a lower admission threshold so new peers
-		// can enter and start accumulating uptime. Subsequent heartbeats must meet
-		// the full threshold once uptime is tracked.
-		minScore := float64(40)
-		if isFirstHeartbeat {
-			minScore = 40
+		// E: measure the indexer's own subnet diversity, not the node's view.
+		diversity := getOwnDiversityRate(h)
+		// fillRate: fraction of indexer capacity used — higher = more peers trust this indexer.
+		fillRate := 0.0
+		if maxNodes > 0 {
+			fillRate = float64(len(h.Network().Peers())) / float64(maxNodes)
+			if fillRate > 1 {
+				fillRate = 1
 			}
+		}
+		hb.ComputeIndexerScore(uptimeRatio, bpms, diversity, latencyScore, fillRate)
+		// B: dynamic minScore — starts at 20% for brand-new peers, ramps to 80% at 24h.
+		minScore := dynamicMinScore(age)
 		if hb.Score < minScore {
 			return nil, nil, errors.New("not enough trusting value")
 		}
@@ -247,7 +255,7 @@ func CheckHeartbeat(h host.Host, s network.Stream, dec *json.Decoder, streams ma
 			DID:    hb.DID,
 			Stream: s,
 			Expiry: time.Now().UTC().Add(2 * time.Minute),
-		} // here is the long-lived bidirectionnal heart bit.
+		} // here is the long-lived bidirectional heartbeat.
 		return &pid, &hb, err
 	}
 }
@@ -268,7 +276,40 @@ func getDiversityRate(h host.Host, peers []string) float64 {
 	if len(diverse) == 0 || len(peers) == 0 {
 		return 1
 	}
-	return float64(len(diverse) / len(peers))
+	return float64(len(diverse)) / float64(len(peers))
+}
+
+// getOwnDiversityRate measures subnet /24 diversity of the indexer's own connected peers.
+// This evaluates the indexer's network position rather than the connecting node's topology.
+func getOwnDiversityRate(h host.Host) float64 {
+	diverse := map[string]struct{}{}
+	total := 0
+	for _, pid := range h.Network().Peers() {
+		for _, maddr := range h.Peerstore().Addrs(pid) {
+			total++
+			ip, err := ExtractIP(maddr.String())
+			if err != nil {
+				continue
+			}
+			diverse[ip.Mask(net.CIDRMask(24, 32)).String()] = struct{}{}
+		}
+	}
+	if total == 0 {
+		return 1
+	}
+	return float64(len(diverse)) / float64(total)
+}
+
+// dynamicMinScore returns the minimum acceptable score for a peer, starting
+// permissive (20%) for brand-new peers and hardening linearly to 80% over 24h.
+// This prevents ejecting newcomers in fresh networks while filtering parasites.
+func dynamicMinScore(age time.Duration) float64 {
+	hours := age.Hours()
+	score := 20.0 + 60.0*(hours/24.0)
+	if score > 80.0 {
+		score = 80.0
+	}
+	return score
 }
 
 func checkPeers(h host.Host, peers []string) ([]string, []string) {
@@ -295,53 +336,95 @@ const MaxPayloadChallenge = 2048
 const BaseRoundTrip = 400 * time.Millisecond
 
 // getBandwidthChallengeRate opens a dedicated ProtocolBandwidthProbe stream to
-// remotePeer, sends a random payload, reads the echo, and computes throughput.
+// remotePeer, sends a random payload, reads the echo, and computes throughput
+// and a latency score. Returns (ok, bpms, latencyScore, error).
+// latencyScore is 1.0 when RTT is very fast and 0.0 when at or beyond maxRoundTrip.
 // Using a separate stream avoids mixing binary data on the JSON heartbeat stream
 // and ensures the echo handler is actually running on the remote side.
-func getBandwidthChallengeRate(h host.Host, remotePeer pp.ID, payloadSize int) (bool, float64, error) {
+func getBandwidthChallengeRate(h host.Host, remotePeer pp.ID, payloadSize int) (bool, float64, float64, error) {
 	payload := make([]byte, payloadSize)
 	if _, err := cr.Read(payload); err != nil {
-		return false, 0, err
+		return false, 0, 0, err
 	}
 
 	ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
 	defer cancel()
 	s, err := h.NewStream(ctx, remotePeer, ProtocolBandwidthProbe)
 	if err != nil {
-		return false, 0, err
+		return false, 0, 0, err
 	}
 	defer s.Reset()
 	s.SetDeadline(time.Now().Add(10 * time.Second))
 	start := time.Now()
 	if _, err = s.Write(payload); err != nil {
-		return false, 0, err
+		return false, 0, 0, err
 	}
 	s.CloseWrite()
 	// Half-close the write side so the handler's io.Copy sees EOF and stops.
 	// Read the echo.
 	response := make([]byte, payloadSize)
 	if _, err = io.ReadFull(s, response); err != nil {
-		return false, 0, err
+		return false, 0, 0, err
 	}
 
 	duration := time.Since(start)
 	maxRoundTrip := BaseRoundTrip + (time.Duration(payloadSize) * (100 * time.Millisecond))
 	mbps := float64(payloadSize*8) / duration.Seconds() / 1e6
-	if duration > maxRoundTrip || mbps < 5.0 {
-		return false, float64(mbps / MaxExpectedMbps), nil
+
+	// latencyScore: 1.0 = instant, 0.0 = at maxRoundTrip or beyond.
+	latencyScore := 1.0 - float64(duration)/float64(maxRoundTrip)
+	if latencyScore < 0 {
+		latencyScore = 0
 	}
-	return true, float64(mbps / MaxExpectedMbps), nil
+	if latencyScore > 1 {
+		latencyScore = 1
+	}
+
+	if duration > maxRoundTrip || mbps < 5.0 {
+		return false, float64(mbps / MaxExpectedMbps), latencyScore, nil
+	}
+	return true, float64(mbps / MaxExpectedMbps), latencyScore, nil
 }
 
 type UptimeTracker struct {
 	FirstSeen   time.Time
 	LastSeen    time.Time
+	TotalOnline time.Duration
+}
+
+// RecordHeartbeat accumulates online time gap-aware: only counts the interval if
+// the gap since the last heartbeat is within 2× the recommended interval (i.e. no
+// extended outage). Call this each time a heartbeat is successfully processed.
+func (u *UptimeTracker) RecordHeartbeat() {
+	now := time.Now().UTC()
+	if !u.LastSeen.IsZero() {
+		gap := now.Sub(u.LastSeen)
+		if gap <= 2*RecommendedHeartbeatInterval {
+			u.TotalOnline += gap
+		}
+	}
+	u.LastSeen = now
 }
 
 func (u *UptimeTracker) Uptime() time.Duration {
 	return time.Since(u.FirstSeen)
 }
 
+// UptimeRatio returns the fraction of tracked lifetime during which the peer was
+// continuously online (gap ≤ 2×RecommendedHeartbeatInterval). Returns 0 before
+// the first heartbeat interval has elapsed.
+func (u *UptimeTracker) UptimeRatio() float64 {
+	total := time.Since(u.FirstSeen)
+	if total <= 0 {
+		return 0
+	}
+	ratio := float64(u.TotalOnline) / float64(total)
+	if ratio > 1 {
+		ratio = 1
+	}
+	return ratio
+}
+
 func (u *UptimeTracker) IsEligible(min time.Duration) bool {
 	return u.Uptime() >= min
 }
@@ -350,6 +433,7 @@ type StreamRecord[T interface{}] struct {
 	DID             string
 	HeartbeatStream *Stream
 	Record          T
+	LastScore       float64
 }
 
 func (s *StreamRecord[T]) GetUptimeTracker() *UptimeTracker {
@@ -426,7 +510,24 @@ const (
 
 var TimeWatcher time.Time
 
+// IndexerRecord holds admission metadata for an indexer in the pool.
+// AdmittedAt is zero for seed entries (IndexerAddresses) never validated by a native.
+// It is set to the admission time when a native confirms the indexer via consensus.
+type IndexerRecord struct {
+	AdmittedAt time.Time
+}
+
+// IsStableVoter returns true when this indexer has been admitted by a native
+// long enough ago to participate as a voter in Phase 2 liveness voting.
+func (r *IndexerRecord) IsStableVoter() bool {
+	return !r.AdmittedAt.IsZero() && time.Since(r.AdmittedAt) >= MinStableAge
+}
+
 var StaticIndexers map[string]*pp.AddrInfo = map[string]*pp.AddrInfo{}
+
+// StaticIndexerMeta mirrors StaticIndexers with admission metadata.
+// Both maps are always updated together under StreamMuIndexes.
+var StaticIndexerMeta map[string]*IndexerRecord = map[string]*IndexerRecord{}
 var StreamMuIndexes sync.RWMutex
 var StreamIndexers ProtocolStream = ProtocolStream{}
 
@@ -462,28 +563,64 @@ func ConnectToIndexers(h host.Host, minIndexer int, maxIndexer int, myPID pp.ID,
 		return nil
 	}
 
+	// No native configured: bootstrap from IndexerAddresses seed set.
 	addresses := strings.Split(conf.GetConfig().IndexerAddresses, ",")
-
 	if len(addresses) > maxIndexer {
 		addresses = addresses[0:maxIndexer]
 	}
 
 	StreamMuIndexes.Lock()
 	for _, indexerAddr := range addresses {
+		indexerAddr = strings.TrimSpace(indexerAddr)
+		if indexerAddr == "" {
+			continue
+		}
 		ad, err := pp.AddrInfoFromString(indexerAddr)
 		if err != nil {
 			logger.Err(err)
 			continue
 		}
+		// AdmittedAt zero = seed, not yet validated by a native.
 		StaticIndexers[indexerAddr] = ad
+		StaticIndexerMeta[indexerAddr] = &IndexerRecord{}
 	}
 	indexerCount := len(StaticIndexers)
 	StreamMuIndexes.Unlock()
 
-	SendHeartbeat(context.Background(), ProtocolHeartbeat, conf.GetConfig().Name, h, StreamIndexers, StaticIndexers, &StreamMuIndexes, 20*time.Second, recordFn...) // your indexer is just like a node for the next indexer.
 	if indexerCount < minIndexer {
 		return errors.New("you run a node without indexers... your gonna be isolated.")
 	}
+
+	// Start long-lived heartbeat to seed indexers. The single goroutine follows
+	// all subsequent StaticIndexers changes (including after native discovery).
+	SendHeartbeat(context.Background(), ProtocolHeartbeat, conf.GetConfig().Name,
+		h, StreamIndexers, StaticIndexers, &StreamMuIndexes, 20*time.Second, recordFn...)
+
+	// Async: ask seed indexers whether they know a native — same logic as
+	// replenishNativesFromPeers. Runs after a short delay to let h.Connect warm up.
+	go func() {
+		time.Sleep(2 * time.Second)
+		logger.Info().Msg("[startup] no native configured — asking seed indexers for native addresses")
+		newAddr := fetchNativeFromIndexers(h, nil)
+		if newAddr == "" {
+			logger.Info().Msg("[startup] no native found from seed indexers — pure indexer mode")
+			return
+		}
+		ad, err := pp.AddrInfoFromString(newAddr)
+		if err != nil {
+			return
+		}
+		logger.Info().Str("addr", newAddr).Msg("[startup] native discovered via seed indexers — bootstrapping")
+		StreamNativeMu.Lock()
+		StaticNatives[newAddr] = ad
+		StreamNativeMu.Unlock()
+		// Full native bootstrap: fetch pool, run consensus, replace StaticIndexers
+		// with properly admitted records (AdmittedAt set).
+		if err := ConnectToNatives(h, minIndexer, maxIndexer, myPID); err != nil {
+			logger.Warn().Err(err).Msg("[startup] native bootstrap failed after discovery")
+		}
+	}()
+
 	return nil
 }
 
@@ -536,10 +673,19 @@ type Heartbeat struct {
 	Record json.RawMessage `json:"record,omitempty"`
 }
 
-func (hb *Heartbeat) ComputeIndexerScore(uptimeHours float64, bpms float64, diversity float64) {
-	hb.Score = ((0.3 * uptimeHours) +
-		(0.3 * bpms) +
-		(0.4 * diversity)) * 100
+// ComputeIndexerScore computes a composite quality score [0, 100] for the connecting peer.
+//   - uptimeRatio:   fraction of tracked lifetime online (gap-aware) — peer reliability
+//   - bpms:         bandwidth normalized to MaxExpectedMbps — link capacity
+//   - diversity:    indexer's own /24 subnet diversity — network topology quality
+//   - latencyScore: 1 - RTT/maxRoundTrip — link responsiveness
+//   - fillRate:     fraction of indexer slots used (0=empty, 1=full) — collective trust signal:
+//     a fuller indexer has been chosen and retained by many peers, which is evidence of quality.
+func (hb *Heartbeat) ComputeIndexerScore(uptimeRatio float64, bpms float64, diversity float64, latencyScore float64, fillRate float64) {
+	hb.Score = ((0.20 * uptimeRatio) +
+		(0.20 * bpms) +
+		(0.20 * diversity) +
+		(0.15 * latencyScore) +
+		(0.25 * fillRate)) * 100
 }
 
 type HeartbeatInfo []struct {
@@ -616,6 +762,9 @@ func SendHeartbeat(ctx context.Context, proto protocol.ID, name string, h host.H
 
 			for _, ix := range snapshot {
 				wasConnected := h.Network().Connectedness(ix.ID) == network.Connected
+				StreamNativeMu.RLock()
+				hasNative := len(StaticNatives) > 0
+				StreamNativeMu.RUnlock()
 				if err := sendHeartbeat(ctx, h, proto, ix, hb, ps, interval*time.Second); err != nil {
 					// Step 3: heartbeat failed — remove from pool and trigger replenish.
 					logger.Info().Str("peer", ix.ID.String()).Str("proto", string(proto)).Msg("[native] step 3 — heartbeat failed, removing peer from pool")
@@ -639,6 +788,9 @@ func SendHeartbeat(ctx context.Context, proto protocol.ID, name string, h host.H
 						if ad.ID == ix.ID {
 							lostAddr = addr
 							delete(peers, addr)
+							if isIndexerHB {
+								delete(StaticIndexerMeta, addr)
+							}
 							break
 						}
 					}
@@ -650,7 +802,8 @@ func SendHeartbeat(ctx context.Context, proto protocol.ID, name string, h host.H
 					logger.Info().Int("remaining", remaining).Int("min", conf.GetConfig().MinIndexer).Int("need", need).Msg("[native] step 3 — pool state after removal")
 
 					// Step 4: ask the native for the missing indexer count.
-					if isIndexerHB && conf.GetConfig().NativeIndexerAddresses != "" {
+					// hasNative computed above (used in both err and success branches).
+					if isIndexerHB && hasNative {
 						if need < 1 {
 							need = 1
 						}
@@ -663,7 +816,7 @@ func SendHeartbeat(ctx context.Context, proto protocol.ID, name string, h host.H
 					// from StaticIndexers immediately without waiting for the indexer HB tick.
 					if isNativeHB {
 						logger.Info().Str("addr", lostAddr).Msg("[native] step 3 — native heartbeat failed, triggering native replenish")
-						if lostAddr != "" && conf.GetConfig().NativeIndexerAddresses != "" {
+						if lostAddr != "" && hasNative {
 							StreamMuIndexes.Lock()
 							if _, wasIndexer := StaticIndexers[lostAddr]; wasIndexer {
 								delete(StaticIndexers, lostAddr)
@@ -695,7 +848,7 @@ func SendHeartbeat(ctx context.Context, proto protocol.ID, name string, h host.H
 					// blank state (responsiblePeers empty). Evict it from StaticIndexers and
 					// re-request an assignment so the native re-tracks us properly and
 					// runOffloadLoop can eventually migrate us to real indexers.
-					if !wasConnected && isIndexerHB && conf.GetConfig().NativeIndexerAddresses != "" {
+					if !wasConnected && isIndexerHB && hasNative {
 						StreamNativeMu.RLock()
 						isNativeIndexer := false
 						for _, ad := range StaticNatives {

daemons/node/common/native_stream.go

@@ -35,8 +35,27 @@ const (
 	consensusQueryTimeout = 3 * time.Second
 	// consensusCollectTimeout is the total wait for all native responses.
 	consensusCollectTimeout = 4 * time.Second
+
+	// ProtocolIndexerConsensus is the Phase 2 liveness-voting protocol.
+	// Each stable indexer is asked which candidates it considers reachable.
+	ProtocolIndexerConsensus = "/opencloud/indexer/consensus/1.0"
+
+	// MinStableAge is the minimum time since native admission before an indexer
+	// may participate as a voter in Phase 2 liveness voting.
+	MinStableAge = 2 * time.Minute
 )
 
+// IndexerConsensusRequest is sent to stable indexers during Phase 2 liveness voting.
+// Each voter replies with which candidates from the list it can currently reach.
+type IndexerConsensusRequest struct {
+	Candidates []string `json:"candidates"`
+}
+
+// IndexerConsensusResponse is the reply from a Phase 2 voter.
+type IndexerConsensusResponse struct {
+	Alive []string `json:"alive"`
+}
+
 // ConsensusRequest is sent by a node/indexer to a native to validate a candidate
 // indexer list. The native replies with what it trusts and what it suggests instead.
 type ConsensusRequest struct {
@@ -61,6 +80,7 @@ type IndexerRegistration struct {
 	Timestamp int64   `json:"ts,omitempty"`          // Unix nanoseconds (anti-replay)
 	PubKey    []byte  `json:"pub_key,omitempty"`     // marshaled libp2p public key
 	Signature []byte  `json:"sig,omitempty"`         // Sign(signaturePayload())
+	FillRate  float64 `json:"fill_rate,omitempty"`  // connected_nodes / max_nodes (0=empty, 1=full)
 }
 
 // SignaturePayload returns the canonical byte slice that is signed/verified.
@@ -106,9 +126,12 @@ type GetIndexersRequest struct {
 }
 
 // GetIndexersResponse is returned by the native with live indexer multiaddrs.
+// FillRates maps each indexer address to its last reported fill rate (0=empty, 1=full).
+// Nodes use fill rates to prefer indexers with available capacity.
 type GetIndexersResponse struct {
 	Indexers       []string           `json:"indexers"`
 	IsSelfFallback bool               `json:"is_self_fallback,omitempty"`
+	FillRates      map[string]float64 `json:"fill_rates,omitempty"`
 }
 
 var StaticNatives = map[string]*pp.AddrInfo{}
@@ -177,8 +200,8 @@ func ConnectToNatives(h host.Host, minIndexer int, maxIndexer int, myPID pp.ID)
 	logger.Info().Int("candidates", len(candidates)).Bool("fallback", isFallback).Msg("[native] step 1 — pool received")
 
 	// Step 2: populate StaticIndexers — consensus for real indexers, direct for fallback.
-	pool := resolvePool(h, candidates, isFallback, maxIndexer)
-	replaceStaticIndexers(pool)
+	pool, admittedAt := resolvePool(h, candidates, isFallback, maxIndexer)
+	replaceStaticIndexers(pool, admittedAt)
 
 	StreamMuIndexes.RLock()
 	indexerCount := len(StaticIndexers)
@@ -216,7 +239,7 @@ func replenishIndexersFromNative(h host.Host, need int) {
 	}
 	logger.Info().Int("candidates", len(candidates)).Bool("fallback", isFallback).Msg("[native] step 4 — candidates received")
 
-	pool := resolvePool(h, candidates, isFallback, need)
+	pool, admittedAt := resolvePool(h, candidates, isFallback, need)
 	if len(pool) == 0 {
 		logger.Warn().Msg("[native] step 4 — consensus yielded no confirmed indexers")
 		return
@@ -226,9 +249,11 @@ func replenishIndexersFromNative(h host.Host, need int) {
 	StreamMuIndexes.Lock()
 	for addr, ad := range pool {
 		StaticIndexers[addr] = ad
+		if StaticIndexerMeta[addr] == nil {
+			StaticIndexerMeta[addr] = &IndexerRecord{AdmittedAt: admittedAt}
+		}
 	}
 	total := len(StaticIndexers)
-
 	StreamMuIndexes.Unlock()
 	logger.Info().Int("added", len(pool)).Int("total", total).Msg("[native] step 4 — pool replenished")
 
@@ -335,9 +360,9 @@ collect:
 }
 
 // resolvePool converts a candidate list to a validated addr→AddrInfo map.
-// When isFallback is true the native itself is the indexer — no consensus needed.
-// When isFallback is false, consensus is run before accepting the candidates.
-func resolvePool(h host.Host, candidates []string, isFallback bool, maxIndexer int) map[string]*pp.AddrInfo {
+// When isFallback is true the native itself is the indexer — no Phase 1 consensus needed.
+// Returns the pool and the admission timestamp (zero for fallback/seed entries).
+func resolvePool(h host.Host, candidates []string, isFallback bool, maxIndexer int) (map[string]*pp.AddrInfo, time.Time) {
 	logger := oclib.GetLogger()
 	if isFallback {
 		logger.Info().Strs("addrs", candidates).Msg("[native] resolve — fallback mode, skipping consensus")
@@ -349,9 +374,10 @@ func resolvePool(h host.Host, candidates []string, isFallback bool, maxIndexer i
 			}
 			pool[addr] = ad
 		}
-		return pool
+		return pool, time.Time{}
 	}
 
+	// Phase 1 — native admission.
 	// Round 1.
 	logger.Info().Int("candidates", len(candidates)).Msg("[native] resolve — consensus round 1")
 	confirmed, suggestions := clientSideConsensus(h, candidates)
@@ -372,6 +398,7 @@ func resolvePool(h host.Host, candidates []string, isFallback bool, maxIndexer i
 		logger.Info().Int("confirmed", len(confirmed)).Msg("[native] resolve — consensus round 2 done")
 	}
 
+	admittedAt := time.Now().UTC()
 	pool := make(map[string]*pp.AddrInfo, len(confirmed))
 	for _, addr := range confirmed {
 		ad, err := pp.AddrInfoFromString(addr)
@@ -380,18 +407,130 @@ func resolvePool(h host.Host, candidates []string, isFallback bool, maxIndexer i
 		}
 		pool[addr] = ad
 	}
-	logger.Info().Int("pool_size", len(pool)).Msg("[native] resolve — pool ready")
-	return pool
+
+	// Phase 2 — indexer liveness vote.
+	logger.Info().Int("pool_size", len(pool)).Msg("[native] resolve — Phase 1 done, running Phase 2 liveness vote")
+	pool = indexerLivenessVote(h, pool)
+	logger.Info().Int("pool_size", len(pool)).Msg("[native] resolve — Phase 2 done, pool ready")
+	return pool, admittedAt
+}
+
+// indexerLivenessVote runs Phase 2 of the hybrid consensus: it queries every
+// stable indexer in StaticIndexers (AdmittedAt non-zero, age >= MinStableAge)
+// for their view of the candidate list and returns only the candidates confirmed
+// by quorum. When no stable voter exists the full admitted set is returned
+// unchanged — this is correct on first boot before any indexer is old enough.
+func indexerLivenessVote(h host.Host, admitted map[string]*pp.AddrInfo) map[string]*pp.AddrInfo {
+	logger := oclib.GetLogger()
+
+	StreamMuIndexes.RLock()
+	voters := make([]*pp.AddrInfo, 0, len(StaticIndexers))
+	for addr, ad := range StaticIndexers {
+		if meta, ok := StaticIndexerMeta[addr]; ok && meta.IsStableVoter() {
+			voters = append(voters, ad)
+		}
+	}
+	StreamMuIndexes.RUnlock()
+
+	if len(voters) == 0 {
+		logger.Info().Msg("[phase2] no stable voters yet — trusting Phase 1 result")
+		return admitted
+	}
+
+	candidates := make([]string, 0, len(admitted))
+	for addr := range admitted {
+		candidates = append(candidates, addr)
+	}
+
+	type result struct {
+		alive map[string]struct{}
+		ok    bool
+	}
+	ch := make(chan result, len(voters))
+
+	for _, voter := range voters {
+		go func(v *pp.AddrInfo) {
+			ctx, cancel := context.WithTimeout(context.Background(), consensusQueryTimeout)
+			defer cancel()
+			if err := h.Connect(ctx, *v); err != nil {
+				ch <- result{}
+				return
+			}
+			s, err := h.NewStream(ctx, v.ID, ProtocolIndexerConsensus)
+			if err != nil {
+				ch <- result{}
+				return
+			}
+			s.SetDeadline(time.Now().Add(consensusQueryTimeout))
+			defer s.Close()
+			if err := json.NewEncoder(s).Encode(IndexerConsensusRequest{Candidates: candidates}); err != nil {
+				ch <- result{}
+				return
+			}
+			var resp IndexerConsensusResponse
+			if err := json.NewDecoder(s).Decode(&resp); err != nil {
+				ch <- result{}
+				return
+			}
+			alive := make(map[string]struct{}, len(resp.Alive))
+			for _, a := range resp.Alive {
+				alive[a] = struct{}{}
+			}
+			ch <- result{alive: alive, ok: true}
+		}(voter)
+	}
+
+	timer := time.NewTimer(consensusCollectTimeout)
+	defer timer.Stop()
+
+	aliveCounts := map[string]int{}
+	total, collected := 0, 0
+collect:
+	for collected < len(voters) {
+		select {
+		case r := <-ch:
+			collected++
+			if !r.ok {
+				continue
+			}
+			total++
+			for addr := range r.alive {
+				aliveCounts[addr]++
+			}
+		case <-timer.C:
+			break collect
+		}
+	}
+
+	if total == 0 {
+		logger.Info().Msg("[phase2] no voter responded — trusting Phase 1 result")
+		return admitted
+	}
+
+	quorum := conf.GetConfig().ConsensusQuorum
+	if quorum <= 0 {
+		quorum = 0.5
+	}
+	confirmed := make(map[string]*pp.AddrInfo, len(admitted))
+	for addr, ad := range admitted {
+		if float64(aliveCounts[addr]) > float64(total)*quorum {
+			confirmed[addr] = ad
+		}
+	}
+	logger.Info().Int("admitted", len(admitted)).Int("confirmed", len(confirmed)).Int("voters", total).Msg("[phase2] liveness vote complete")
+	return confirmed
 }
 
 // replaceStaticIndexers atomically replaces the active indexer pool.
-// Peers no longer in next have their heartbeat streams closed so the SendHeartbeat
-// goroutine stops sending to them on the next tick.
-func replaceStaticIndexers(next map[string]*pp.AddrInfo) {
+// admittedAt is the time of native admission (zero for fallback/seed entries).
+func replaceStaticIndexers(next map[string]*pp.AddrInfo, admittedAt time.Time) {
 	StreamMuIndexes.Lock()
 	defer StreamMuIndexes.Unlock()
 	for addr, ad := range next {
 		StaticIndexers[addr] = ad
+		if StaticIndexerMeta[addr] == nil {
+			StaticIndexerMeta[addr] = &IndexerRecord{AdmittedAt: admittedAt}
+		}
 	}
 }
 
@@ -508,8 +647,10 @@ collect:
 }
 
 // RegisterWithNative sends a one-shot registration to each configured native indexer.
+// fillRateFn, when non-nil, is called to obtain the current fill rate (0=empty, 1=full)
+// which the native uses to route new nodes toward less-loaded indexers.
 // Should be called periodically every RecommendedHeartbeatInterval.
-func RegisterWithNative(h host.Host, nativeAddressesStr string) {
+func RegisterWithNative(h host.Host, nativeAddressesStr string, fillRateFn func() float64) {
 	logger := oclib.GetLogger()
 	myAddr := ""
 	if !strings.Contains(h.Addrs()[len(h.Addrs())-1].String(), "127.0.0.1") {
@@ -524,6 +665,9 @@ func RegisterWithNative(h host.Host, nativeAddressesStr string) {
 		Addr:      myAddr,
 		Timestamp: time.Now().UnixNano(),
 	}
+	if fillRateFn != nil {
+		reg.FillRate = fillRateFn()
+	}
 	reg.Sign(h)
 	for _, addr := range strings.Split(nativeAddressesStr, ",") {
 		addr = strings.TrimSpace(addr)
@@ -619,7 +763,10 @@ func EnsureNativePeers(h host.Host) {
 	})
 }
 
-func StartNativeRegistration(h host.Host, nativeAddressesStr string) {
+// StartNativeRegistration starts a goroutine that periodically registers this
+// indexer with all configured native indexers (every RecommendedHeartbeatInterval).
+// fillRateFn is called on each registration tick to report current capacity usage.
+func StartNativeRegistration(h host.Host, nativeAddressesStr string, fillRateFn func() float64) {
 	go func() {
 		// Poll until a routable (non-loopback) address is available before the first
 		// registration attempt. libp2p may not have discovered external addresses yet
@@ -636,11 +783,11 @@ func StartNativeRegistration(h host.Host, nativeAddressesStr string) {
 			}
 			time.Sleep(5 * time.Second)
 		}
-		RegisterWithNative(h, nativeAddressesStr)
+		RegisterWithNative(h, nativeAddressesStr, fillRateFn)
 		t := time.NewTicker(RecommendedHeartbeatInterval)
 		defer t.Stop()
 		for range t.C {
-			RegisterWithNative(h, nativeAddressesStr)
+			RegisterWithNative(h, nativeAddressesStr, fillRateFn)
 		}
 	}()
 }
@@ -917,7 +1064,7 @@ func retryLostNative(ctx context.Context, h host.Host, addr string, nativeProto
 		NudgeNativeHeartbeat()
 		replenishIndexersIfNeeded(h)
 		if nativeProto == ProtocolNativeGetIndexers {
-			StartNativeRegistration(h, addr) // register back
+			StartNativeRegistration(h, addr, nil) // register back (fill rate unknown in this context)
 		}
 		return
 	}

daemons/node/indexer/handler.go

@@ -178,6 +178,43 @@ func (ix *IndexerService) initNodeHandler() {
 	ix.Host.SetStreamHandler(common.ProtocolPublish, ix.handleNodePublish)
 	ix.Host.SetStreamHandler(common.ProtocolGet, ix.handleNodeGet)
 	ix.Host.SetStreamHandler(common.ProtocolIndexerGetNatives, ix.handleGetNatives)
+	ix.Host.SetStreamHandler(common.ProtocolIndexerConsensus, ix.handleIndexerConsensus)
+}
+
+// handleIndexerConsensus implements Phase 2 liveness voting (ProtocolIndexerConsensus).
+// The caller sends a list of candidate multiaddrs; this indexer replies with the
+// subset it considers currently alive (recent heartbeat in StreamRecords).
+func (ix *IndexerService) handleIndexerConsensus(stream network.Stream) {
+	defer stream.Reset()
+
+	var req common.IndexerConsensusRequest
+	if err := json.NewDecoder(stream).Decode(&req); err != nil {
+		return
+	}
+
+	ix.StreamMU.RLock()
+	streams := ix.StreamRecords[common.ProtocolHeartbeat]
+	ix.StreamMU.RUnlock()
+
+	alive := make([]string, 0, len(req.Candidates))
+	for _, addr := range req.Candidates {
+		ad, err := peer.AddrInfoFromString(addr)
+		if err != nil {
+			continue
+		}
+		ix.StreamMU.RLock()
+		rec, ok := streams[ad.ID]
+		ix.StreamMU.RUnlock()
+		if !ok || rec.HeartbeatStream == nil || rec.HeartbeatStream.UptimeTracker == nil {
+			continue
+		}
+		// D: consider alive only if recent heartbeat AND score above minimum quality bar.
+		if time.Since(rec.HeartbeatStream.UptimeTracker.LastSeen) <= 2*common.RecommendedHeartbeatInterval &&
+			rec.LastScore >= 30.0 {
+			alive = append(alive, addr)
+		}
+	}
+	json.NewEncoder(stream).Encode(common.IndexerConsensusResponse{Alive: alive})
 }
 
 func (ix *IndexerService) handleNodePublish(s network.Stream) {

daemons/node/indexer/native.go

@@ -40,6 +40,7 @@ const (
 // liveIndexerEntry tracks a registered indexer in the native's in-memory cache and DHT.
 // PubKey and Signature are forwarded from the IndexerRegistration so the DHT validator
 // can verify that the entry was produced by the peer owning the declared PeerID.
+// FillRate is the fraction of capacity used (0=empty, 1=full) at last registration.
 type liveIndexerEntry struct {
 	PeerID       string    `json:"peer_id"`
 	Addr         string    `json:"addr"`
@@ -47,6 +48,7 @@ type liveIndexerEntry struct {
 	RegTimestamp int64     `json:"reg_ts,omitempty"` // Timestamp from the original IndexerRegistration
 	PubKey       []byte    `json:"pub_key,omitempty"`
 	Signature    []byte    `json:"sig,omitempty"`
+	FillRate     float64   `json:"fill_rate,omitempty"`
 }
 
 // NativeState holds runtime state specific to native indexer operation.
@@ -265,6 +267,7 @@ func (ix *IndexerService) handleNativeSubscription(s network.Stream) {
 			RegTimestamp: reg.Timestamp,
 			PubKey:       reg.PubKey,
 			Signature:    reg.Signature,
+			FillRate:     reg.FillRate,
 		}
 
 		// Verify that the declared address is actually reachable before admitting
@@ -428,11 +431,40 @@ func (ix *IndexerService) handleNativeGetIndexers(s network.Stream) {
 					"native: fallback pool saturated, refusing self-delegation")
 			}
 		} else {
-			rand.Shuffle(len(reachable), func(i, j int) { reachable[i], reachable[j] = reachable[j], reachable[i] })
+			// Sort by fill rate ascending so less-full indexers are preferred for routing.
+			ix.Native.liveIndexersMu.RLock()
+			fillRates := make(map[string]float64, len(reachable))
+			for _, addr := range reachable {
+				ad, err := pp.AddrInfoFromString(addr)
+				if err != nil {
+					continue
+				}
+				for _, e := range ix.Native.liveIndexers {
+					if e.PeerID == ad.ID.String() {
+						fillRates[addr] = e.FillRate
+						break
+					}
+				}
+			}
+			ix.Native.liveIndexersMu.RUnlock()
+
+			// Sort by routing weight descending: weight = fillRate × (1 − fillRate).
+			// This prefers indexers in the "trust sweet spot" — proven popular (fillRate > 0)
+			// but not saturated (fillRate < 1). Peak at fillRate ≈ 0.5.
+			routingWeight := func(addr string) float64 {
+				f := fillRates[addr]
+				return f * (1 - f)
+			}
+			for i := 1; i < len(reachable); i++ {
+				for j := i; j > 0 && routingWeight(reachable[j]) > routingWeight(reachable[j-1]); j-- {
+					reachable[j], reachable[j-1] = reachable[j-1], reachable[j]
+				}
+			}
 			if req.Count > len(reachable) {
 				req.Count = len(reachable)
 			}
 			resp.Indexers = reachable[:req.Count]
+			resp.FillRates = fillRates
 		}
 
 		if err := json.NewEncoder(s).Encode(resp); err != nil {

daemons/node/indexer/service.go

@@ -96,9 +96,24 @@ func NewIndexerService(h host.Host, ps *pubsub.PubSub, maxNode int, isNative boo
 		ix.InitNative()
 	} else {
 		ix.initNodeHandler()
-		// Register with configured natives so this indexer appears in their cache
+		// Register with configured natives so this indexer appears in their cache.
+		// Pass a fill rate provider so the native can route new nodes to less-loaded indexers.
 		if nativeAddrs := conf.GetConfig().NativeIndexerAddresses; nativeAddrs != "" {
-			common.StartNativeRegistration(ix.Host, nativeAddrs)
+			fillRateFn := func() float64 {
+				ix.StreamMU.RLock()
+				n := len(ix.StreamRecords[common.ProtocolHeartbeat])
+				ix.StreamMU.RUnlock()
+				maxN := ix.MaxNodesConn()
+				if maxN <= 0 {
+					return 0
+				}
+				rate := float64(n) / float64(maxN)
+				if rate > 1 {
+					rate = 1
+				}
+				return rate
+			}
+			common.StartNativeRegistration(ix.Host, nativeAddrs, fillRateFn)
 		}
 	}
 	return ix

daemons/node/node.go

@@ -123,7 +123,6 @@ func InitNode(isNode bool, isIndexer bool, isNativeIndexer bool) (*Node, error)
 			m := map[string]interface{}{}
 			err := json.Unmarshal(evt.Payload, &m)
 			if err != nil || evt.From == node.PeerID.String() {
-				fmt.Println(evt.From, node.PeerID.String(), err)
 				return
 			}
 			if p, err := node.GetPeerRecord(ctx, evt.From, false); err == nil && len(p) > 0 && m["search"] != nil {

docs/FUTURE_DHT_ARCHITECTURE.txt

@@ -0,0 +1,362 @@
+================================================================================
+  OC-DISCOVERY : ARCHITECTURE CIBLE — RÉSEAU DHT SANS NATIFS
+  Vision d'évolution long terme, issue d'une analyse comparative
+================================================================================
+
+Rédigé à partir de l'analyse de l'architecture actuelle et de la discussion
+comparative avec Tapestry, Kademlia, EigenTrust et les systèmes de réputation
+distribués.
+
+Référence : DECENTRALIZED_SYSTEMS_COMPARISON.txt §9
+
+
+================================================================================
+1. MOTIVATION
+================================================================================
+
+L'architecture actuelle (node → indexer → native indexer) est robuste et bien
+adaptée à une phase précoce du réseau. Ses limites à l'échelle sont :
+
+  - Pool de natives statique au démarrage → dépendance à la configuration
+  - Cache local des natives = point de défaillance unique (perte = pool vide)
+  - Consensus inter-natives bloquant (~7s) déclenché à chaque bootstrap node
+  - État O(N indexers) par native → croît linéairement avec le réseau
+  - Nœuds privilégiés structurellement → SPOFs relatifs
+
+La cible décrite ici supprime la notion de native indexer en tant que tier
+architectural. Le réseau devient plat : indexers et nodes sont des acteurs
+de même nature, différenciés uniquement par leur rôle volontaire.
+
+
+================================================================================
+2. PRINCIPES FONDAMENTAUX
+================================================================================
+
+  P1. Aucun nœud n'est structurellement privilégié.
+  P2. La confiance est un produit du temps et de la vérification, pas d'un arbitre.
+  P3. Les claims d'un acteur sont vérifiables indépendamment par tout pair.
+  P4. La réputation émerge du comportement collectif, pas d'un signalement central.
+  P5. La DHT est une infrastructure neutre — elle stocke des faits, pas des jugements.
+  P6. La configuration statique n'existe plus au runtime — seulement au bootstrap.
+
+
+================================================================================
+3. RÔLES
+================================================================================
+
+3.1 Node
+--------
+Consommateur du réseau. Démarre, sélectionne un pool d'indexers via DHT,
+heartbeat ses indexers, accumule des scores localement. Ne publie rien en
+routine. Participe aux challenges de consensus à la demande.
+
+3.2 Indexer
+-----------
+Acteur volontaire. S'inscrit dans la DHT à la naissance, maintient son record,
+sert le trafic des nodes (heartbeat, Publish, Get). Déclare ses métriques dans
+chaque réponse heartbeat. Maintient un score agrégé depuis ses nodes connectés.
+
+  Différence avec l'actuel : l'indexer n'a plus de lien avec une native.
+  Il est autonome. Son existence dans le réseau est prouvée par son record DHT
+  et par les nodes qui le contactent directement.
+
+3.3 Nœud DHT infrastructure (ex-native)
+----------------------------------------
+N'importe quel nœud suffisamment stable peut maintenir la DHT sans être un
+indexer. C'est une configuration, pas un type architectural : `dht_mode: server`.
+Ces nœuds maintiennent les k-buckets Kademlia et stockent les records des
+indexers. Ils ne connaissent pas le trafic node↔indexer et ne l'orchestrent pas.
+
+
+================================================================================
+4. BOOTSTRAP D'UN NODE
+================================================================================
+
+4.1 Entrée dans le réseau
+-------------------------
+Le node démarre avec 1 à 3 adresses de nœuds DHT connus (bootstrap peers).
+Ce sont les seules informations statiques nécessaires. Ces peers n'ont pas de
+rôle sémantique — ils servent uniquement à entrer dans l'overlay DHT.
+
+4.2 Découverte du pool d'indexers
+----------------------------------
+
+  Node → DHT.FindProviders(hash("/opencloud/indexers"))
+       → reçoit une liste de N candidats avec leurs records
+
+Sélection du pool initial :
+
+  1. Filtre latence   : ping < seuil → proximité réseau réelle
+  2. Filtre fill rate : préférer les indexers moins chargés
+  3. Tirage pondéré   : probabilité ∝ (1 - fill_rate), courbe w(F) = F×(1-F)
+                        indexer à 20% charge → très probable
+                        indexer à 80% charge → peu probable
+  4. Filtre diversité : subnet /24 différent pour chaque entrée du pool
+
+Aucun consensus nécessaire à cette étape. Le node démarre avec une tolérance
+basse (voir §7) — il accepte des indexers imparfaits et les évalue au fil du temps.
+
+
+================================================================================
+5. REGISTRATION D'UN INDEXER DANS LA DHT
+================================================================================
+
+À la naissance, l'indexer publie son record DHT :
+
+  clé   : hash("/opencloud/indexers")          ← clé fixe, connue de tous
+  valeur: {
+    multiaddr   : <adresse réseau>,
+    region      : <subnet /24>,
+    capacity    : <maxNodesConn>,
+    fill_rate   : <float 0-1>,                 ← auto-déclaré, vérifiable
+    peer_count  : <int>,                       ← auto-déclaré, vérifiable
+    peers       : [hash(nodeID1), ...],        ← liste hashée des nodes connectés
+    born_at     : <timestamp>,
+    sig         : <signature clé indexer>,     ← non-forgeable (PSK context)
+  }
+
+Le record est rafraîchi toutes les ~60s (avant expiration du TTL).
+Si l'indexer tombe : TTL expire → disparaît de la DHT automatiquement.
+
+La peer list est hashée pour la confidentialité mais reste vérifiable :
+un challenger peut demander directement à un node s'il est connecté à cet indexer.
+
+
+================================================================================
+6. PROTOCOLE HEARTBEAT — QUESTION ET RÉPONSE
+================================================================================
+
+Le heartbeat devient bidirectionnel : le node pose des questions, l'indexer
+répond avec ses déclarations courantes.
+
+6.1 Structure
+-------------
+
+  Node → Indexer :
+  {
+    ts        : now,
+    challenge : <optionnel, voir §8>
+  }
+
+  Indexer → Node :
+  {
+    ts           : now,
+    fill_rate    : 0.42,
+    peer_count   : 87,
+    cached_score : 0.74,        ← score agrégé depuis tous ses nodes connectés
+    challenge_response : {...}  ← si challenge présent dans la requête
+  }
+
+Le heartbeat normal (sans challenge) est quasi-identique à l'actuel en poids.
+Le cached_score indexer est mis à jour progressivement par les feedbacks reçus.
+
+6.2 Le cached_score de l'indexer
+---------------------------------
+L'indexer agrège les scores que ses nodes connectés lui communiquent
+(implicitement via le fait qu'ils restent connectés, ou explicitement lors
+d'un consensus). Ce score lui donne une vision de sa propre qualité réseau.
+
+Un node peut comparer son score local de l'indexer avec le cached_score déclaré.
+Une forte divergence est un signal d'alerte.
+
+  Score local node : 0.40   ← cet indexer est médiocre pour moi
+  Cached score     : 0.91   ← il se prétend excellent globalement
+  → déclenche un challenge de vérification
+
+
+================================================================================
+7. MODÈLE DE CONFIANCE PROGRESSIVE
+================================================================================
+
+7.1 Cycle de vie d'un node
+---------------------------
+
+  Naissance
+    → tolérance basse : accepte presque n'importe quel indexer du DHT
+    → switching cost faible : peu de contexte accumulé
+    → minScore ≈ 20% (dynamicMinScore existant, conservé)
+
+  Quelques heures
+    → uptime s'accumule sur chaque indexer connu
+    → scores se stabilisent
+    → seuil de remplacement qui monte progressivement
+
+  Long terme (jours)
+    → pool stable, confiance élevée sur les indexers connus
+    → switching coûteux mais déclenché sur déception franche
+    → minScore ≈ 80% (maturité)
+
+7.2 Modèle sous-jacent : beta distribution implicite
+------------------------------------------------------
+
+  α = succès cumulés  (heartbeats OK, probes OK, challenges réussis)
+  β = échecs cumulés  (timeouts, probes échoués, challenges ratés)
+
+  confiance = α / (α + β)
+
+  Nouveau indexer   : α=0, β=0   → prior neutre, tolérance basse
+  Après 10 jours    : α élevé    → confiance stable, seuil de switch élevé
+  Déception franche : β monte    → confiance chute → switch déclenché
+
+7.3 Ce que "décevoir" signifie
+--------------------------------
+
+  Heartbeat rate     → trop de timeouts → fiabilité en baisse
+  Bandwidth probe    → chute sous déclaré → dégradation ou mensonge
+  Fill rate réel     → supérieur au déclaré → indexer surchargé ou malhonnête
+  Challenge échoué   → peer déclaré absent du réseau → claim invalide
+  Latence            → dérive progressive → qualité réseau dégradée
+  Cached_score gonflé → divergence forte avec score local → suspicion
+
+
+================================================================================
+8. VÉRIFICATION DES CLAIMS — TROIS COUCHES
+================================================================================
+
+8.1 Couche 1 : passive (chaque heartbeat, 60s)
+-----------------------------------------------
+Mesures automatiques, zéro coût supplémentaire.
+
+  - RTT du heartbeat          → latence directe
+  - fill_rate déclaré         → tiny payload dans la réponse
+  - peer_count déclaré        → tiny payload
+  - cached_score indexer      → comparé au score local
+
+8.2 Couche 2 : sampling actif (1 heartbeat sur N)
+--------------------------------------------------
+Vérifications périodiques, asynchrones, légères.
+
+  Tous les 5 HB (~5min)  : spot-check 1 peer aléatoire (voir §8.4)
+  Tous les 10 HB (~10min): vérification diversité subnet (lookups DHT légers)
+  Tous les 15 HB (~15min): bandwidth probe (transfert réel, protocole dédié)
+
+8.3 Couche 3 : consensus (événementiel)
+-----------------------------------------
+Déclenché sur : admission d'un nouvel indexer dans le pool, ou suspicion détectée.
+
+  Node sélectionne une claim vérifiable de l'indexer cible X
+  Node vérifie lui-même
+  Node demande à ses indexers de confiance : "vérifiez cette claim sur X"
+  Chaque indexer vérifie indépendamment
+  Convergence des résultats → X est honnête → admission
+  Divergence               → X est suspect → rejet ou probation
+
+  Le consensus est léger : quelques contacts out-of-band, pas de round bloquant.
+  Il n'est pas continu — il est événementiel.
+
+8.4 Vérification out-of-band (pas de DHT writes par les nodes)
+----------------------------------------------------------------
+Les nodes ne publient PAS de contact records continus dans la DHT.
+Cela éviterait N×M records à rafraîchir (coût DHT élevé à l'échelle).
+
+À la place, lors d'un challenge :
+
+  Challenger sélectionne 2-3 peers dans la peer list déclarée par X
+  → contacte ces peers directement : "es-tu connecté à indexer X ?"
+  → réponse directe (out-of-band, pas via DHT)
+  → vérification sans écriture DHT
+
+L'indexer ne peut pas faire répondre "oui" à des peers qui ne lui sont pas
+connectés. La vérification est non-falsifiable et sans coût DHT.
+
+8.5 Pourquoi X ne peut pas tricher
+------------------------------------
+X ne peut pas coordonner des réponses différentes vers des challengers
+simultanés. Chaque challenger contacte indépendamment les mêmes peers.
+Si X ment sur sa peer list :
+
+  - Challenger A contacte peer P → "non, pas connecté à X"
+  - Challenger B contacte peer P → "non, pas connecté à X"
+  - Consensus : X ment → score chute chez tous les challengers
+  - Effet réseau : progressivement, X perd ses connections
+  - Peer list DHT se vide → claims futures encore moins crédibles
+
+
+================================================================================
+9. EFFET RÉSEAU SANS SIGNALEMENT CENTRAL
+================================================================================
+
+Un node qui pénalise un indexer n'envoie aucun "rapport" à quiconque.
+Ses actions locales produisent l'effet réseau par agrégation :
+
+  Node baisse le score de X      → X reçoit moins de trafic de ce node
+  Node switche vers Y            → X perd un client
+  Node refuse les challenges X   → X ne peut plus participer aux consensus
+
+Si 200 nodes font pareil :
+
+  X perd la majorité de ses connections
+  Sa peer list DHT se vide (peers contactés directement disent "non")
+  Son cached_score s'effondre (peu de nodes restent)
+  Les nouveaux nodes qui voient X dans la DHT obtiennent des challenges échoués
+  X est naturellement exclu sans aucune décision centrale
+
+Inversement, un indexer honnête voit ses scores monter sur tous ses nodes
+connectés, sa peer list se densifier, ses challenges réussis systématiquement.
+Sa réputation est un produit observable et vérifiable.
+
+
+================================================================================
+10. RÉSUMÉ DE L'ARCHITECTURE
+================================================================================
+
+  DHT          → annuaire neutre, vérité des records indexers
+                  maintenu par tout nœud stable (dht_mode: server)
+
+  Indexer      → acteur volontaire, s'inscrit, maintient ses claims,
+                  sert le trafic, accumule son propre score agrégé
+
+  Node         → consommateur, score passif + sampling + consensus léger,
+                  confiance progressive, switching adaptatif
+
+  Heartbeat    → métronome 60s + vecteur de déclarations légères + challenge optionnel
+
+  Consensus    → événementiel, multi-challengers indépendants,
+                  vérification out-of-band sur claims DHT
+
+  Confiance    → beta implicite, progressive, switching cost croissant avec l'âge
+
+  Réputation   → émerge du comportement collectif, aucun arbitre central
+
+  Bootstrap    → 1-3 peers DHT connus → seule configuration statique nécessaire
+
+
+================================================================================
+11. TRAJECTOIRE DE MIGRATION
+================================================================================
+
+  Phase 1 (actuel)
+    Natives statiques, pool indexers dynamique, consensus inter-natives
+    → robuste, adapté à la phase précoce
+
+  Phase 2 (intermédiaire)
+    Pool de natives dynamique via DHT (bootstrap + gossip)
+    Même protocole natif, juste la découverte devient dynamique
+    → supprime la dépendance à la configuration statique des natives
+    → voir DECENTRALIZED_SYSTEMS_COMPARISON.txt §9.2
+
+  Phase 3 (cible)
+    Architecture décrite dans ce document
+    Natives disparaissent en tant que tier architectural
+    DHT = infrastructure, indexers = acteurs autonomes
+    Scoring et consensus entièrement côté node
+    → aucun nœud privilégié, scalabilité O(log N)
+
+  La migration Phase 2 → Phase 3 est une refonte du plan de contrôle.
+  Le plan de données (heartbeat node↔indexer, Publish, Get) est inchangé.
+  Les primitives libp2p (Kademlia DHT, GossipSub) sont déjà présentes.
+
+
+================================================================================
+12. PROPRIÉTÉS DU SYSTÈME CIBLE
+================================================================================
+
+  Scalabilité        O(log N) — routage DHT Kademlia
+  Résilience         Pas de SPOF structurel, TTL = seule source de vérité
+  Confiance          Progressive, vérifiable, émergente
+  Sybil resistance   PSK — seuls les nœuds avec la clé peuvent publier
+  Cold start         Tolérance basse initiale, montée progressive (existant)
+  Honnêteté          Claims vérifiables out-of-band, non-falsifiables
+  Décentralisation   Aucun nœud ne connaît l'état global complet
+
+================================================================================

docs/diagrams/01_node_init.mmd

@@ -1,56 +0,0 @@
-sequenceDiagram
-    title Node Initialization — Pair A (InitNode)
-
-    participant MainA as main (Pair A)
-    participant NodeA as Node A
-    participant libp2pA as libp2p (Pair A)
-    participant DBA as DB Pair A (oc-lib)
-    participant NATSA as NATS A
-    participant IndexerA as Indexer (partagé)
-    participant StreamA as StreamService A
-    participant PubSubA as PubSubService A
-
-    MainA->>NodeA: InitNode(isNode, isIndexer, isNativeIndexer)
-
-    NodeA->>NodeA: LoadKeyFromFilePrivate() → priv
-    NodeA->>NodeA: LoadPSKFromFile() → psk
-
-    NodeA->>libp2pA: New(PrivateNetwork(psk), Identity(priv), ListenAddr:4001)
-    libp2pA-->>NodeA: host A (PeerID_A)
-
-    Note over NodeA: isNode == true
-
-    NodeA->>libp2pA: NewGossipSub(ctx, host)
-    libp2pA-->>NodeA: ps (GossipSub)
-
-    NodeA->>IndexerA: ConnectToIndexers → SendHeartbeat /opencloud/heartbeat/1.0
-    Note over IndexerA: Heartbeat long-lived établi<br/>Score qualité calculé (bw + uptime + diversité)
-    IndexerA-->>NodeA: OK
-
-    NodeA->>NodeA: claimInfo(name, hostname)
-    NodeA->>IndexerA: TempStream /opencloud/record/publish/1.0
-    NodeA->>IndexerA: json.Encode(PeerRecord A signé)
-    IndexerA->>IndexerA: DHT.PutValue("/node/"+DID_A, record)
-
-    NodeA->>DBA: NewRequestAdmin(PEER).Search(SELF)
-    DBA-->>NodeA: peer A local (ou UUID généré)
-
-    NodeA->>NodeA: StartGC(30s) — GC sur StreamRecords
-
-    NodeA->>StreamA: InitStream(ctx, host, PeerID_A, 1000, nodeA)
-    StreamA->>StreamA: SetStreamHandler(heartbeat/partner, search, planner, ...)
-    StreamA->>DBA: Search(PEER, PARTNER) → liste partenaires
-    DBA-->>StreamA: [] (aucun partenaire au démarrage)
-    StreamA-->>NodeA: StreamService A
-
-    NodeA->>PubSubA: InitPubSub(ctx, host, ps, nodeA, streamA)
-    PubSubA->>PubSubA: subscribeEvents(PB_SEARCH, timeout=-1)
-    PubSubA-->>NodeA: PubSubService A
-
-    NodeA->>NodeA: SubscribeToSearch(ps, callback)
-    Note over NodeA: callback: GetPeerRecord(evt.From)<br/>→ StreamService.SendResponse
-
-    NodeA->>NATSA: ListenNATS(nodeA)
-    Note over NATSA: Enregistre handlers:<br/>CREATE_RESOURCE, PROPALGATION_EVENT
-
-    NodeA-->>MainA: *Node A prêt

docs/diagrams/01_node_init.puml

@@ -1,10 +1,10 @@
 @startuml
-title Node Initialization — Pair A (InitNode)
+title Node Initialization — Peer A (InitNode)
 
-participant "main (Pair A)" as MainA
+participant "main (Peer A)" as MainA
 participant "Node A" as NodeA
-participant "libp2p (Pair A)" as libp2pA
-participant "DB Pair A (oc-lib)" as DBA
+participant "libp2p (Peer A)" as libp2pA
+participant "DB Peer A (oc-lib)" as DBA
 participant "NATS A" as NATSA
 participant "Indexer (partagé)" as IndexerA
 participant "StreamService A" as StreamA
@@ -24,35 +24,35 @@ NodeA -> libp2pA: NewGossipSub(ctx, host)
 libp2pA --> NodeA: ps (GossipSub)
 
 NodeA -> IndexerA: ConnectToIndexers → SendHeartbeat /opencloud/heartbeat/1.0
-note over IndexerA: Heartbeat long-lived établi\nScore qualité calculé (bw + uptime + diversité)
+note over IndexerA: Heartbeat long-lived established\nQuality Score evaluated (bw + uptime + diversity)
 IndexerA --> NodeA: OK
 
 NodeA -> NodeA: claimInfo(name, hostname)
 NodeA -> IndexerA: TempStream /opencloud/record/publish/1.0
-NodeA -> IndexerA: json.Encode(PeerRecord A signé)
+NodeA -> IndexerA: stream.Encode(PeerRecord A signé)
 IndexerA -> IndexerA: DHT.PutValue("/node/"+DID_A, record)
 
-NodeA -> DBA: NewRequestAdmin(PEER).Search(SELF)
-DBA --> NodeA: peer A local (ou UUID généré)
+NodeA -> DBA: DB(PEER).Search(SELF)
+DBA --> NodeA: local peer A (or new generated UUID)
 
-NodeA -> NodeA: StartGC(30s) — GC sur StreamRecords
+NodeA -> NodeA: StartGC(30s) — GarbageCollector on StreamRecords
 
 NodeA -> StreamA: InitStream(ctx, host, PeerID_A, 1000, nodeA)
 StreamA -> StreamA: SetStreamHandler(heartbeat/partner, search, planner, ...)
-StreamA -> DBA: Search(PEER, PARTNER) → liste partenaires
-DBA --> StreamA: [] (aucun partenaire au démarrage)
+StreamA -> DBA: Search(PEER, PARTNER) → partner list
+DBA --> StreamA: Heartbeat long-lived established to partners
 StreamA --> NodeA: StreamService A
 
 NodeA -> PubSubA: InitPubSub(ctx, host, ps, nodeA, streamA)
 PubSubA -> PubSubA: subscribeEvents(PB_SEARCH, timeout=-1)
 PubSubA --> NodeA: PubSubService A
 
-NodeA -> NodeA: SubscribeToSearch(ps, callback)
+NodeA -> NodeA: SubscribeToSearch(ps, callback) (search global topic for resources)
 note over NodeA: callback: GetPeerRecord(evt.From)\n→ StreamService.SendResponse
 
 NodeA -> NATSA: ListenNATS(nodeA)
-note over NATSA: Enregistre handlers:\nCREATE_RESOURCE, PROPALGATION_EVENT
+note over NATSA: Subscribes handlers:\nCREATE_RESOURCE, PROPALGATION_EVENT
 
-NodeA --> MainA: *Node A prêt
+NodeA --> MainA: *Node A is ready
 
 @enduml

docs/diagrams/02_node_claim.mmd

@@ -1,38 +0,0 @@
-sequenceDiagram
-    title Node Claim — Pair A publie son PeerRecord (claimInfo + publishPeerRecord)
-
-    participant DBA as DB Pair A (oc-lib)
-    participant NodeA as Node A
-    participant IndexerA as Indexer (partagé)
-    participant DHT as DHT Kademlia
-    participant NATSA as NATS A
-
-    NodeA->>DBA: NewRequestAdmin(PEER).Search(SELF)
-    DBA-->>NodeA: existing peer (DID_A) ou nouveau UUID
-
-    NodeA->>NodeA: LoadKeyFromFilePrivate() → priv A
-    NodeA->>NodeA: LoadKeyFromFilePublic() → pub A
-    NodeA->>NodeA: crypto.MarshalPublicKey(pub A) → pubBytes
-
-    NodeA->>NodeA: Build PeerRecord A {<br/>  Name, DID, PubKey,<br/>  PeerID: PeerID_A,<br/>  APIUrl: hostname,<br/>  StreamAddress: /ip4/.../tcp/4001/p2p/PeerID_A,<br/>  NATSAddress, WalletAddress<br/>}
-
-    NodeA->>NodeA: sha256(json(rec)) → hash
-    NodeA->>NodeA: priv.Sign(hash) → signature
-    NodeA->>NodeA: rec.ExpiryDate = now + 150s
-
-    loop Pour chaque StaticIndexer (Indexer A, B, …)
-        NodeA->>IndexerA: TempStream /opencloud/record/publish/1.0
-        NodeA->>IndexerA: json.Encode(PeerRecord A signé)
-
-        IndexerA->>IndexerA: Verify signature
-        IndexerA->>IndexerA: Check heartbeat stream actif pour PeerID_A
-        IndexerA->>DHT: PutValue("/node/"+DID_A, PeerRecord A)
-        DHT-->>IndexerA: ok
-    end
-
-    NodeA->>NodeA: rec.ExtractPeer(DID_A, DID_A, pub A)
-    NodeA->>NATSA: SetNATSPub(CREATE_RESOURCE, {PEER, Peer A JSON})
-    NATSA->>DBA: Upsert Peer A (SearchAttr: peer_id)
-    DBA-->>NATSA: ok
-
-    NodeA-->>NodeA: *peer.Peer A (SELF)

docs/diagrams/02_node_claim.puml

@@ -1,31 +1,29 @@
 @startuml
-title Node Claim — Pair A publie son PeerRecord (claimInfo + publishPeerRecord)
+title Node Claim — Peer A publish its PeerRecord (claimInfo + publishPeerRecord)
 
-participant "DB Pair A (oc-lib)" as DBA
+participant "DB Peer A (oc-lib)" as DBA
 participant "Node A" as NodeA
-participant "Indexer (partagé)" as IndexerA
+participant "Indexer (shared)" as IndexerA
 participant "DHT Kademlia" as DHT
 participant "NATS A" as NATSA
 
-NodeA -> DBA: NewRequestAdmin(PEER).Search(SELF)
-DBA --> NodeA: existing peer (DID_A) ou nouveau UUID
+NodeA -> DBA: DB(PEER).Search(SELF)
+DBA --> NodeA: existing peer (DID_A) or new UUID
 
 NodeA -> NodeA: LoadKeyFromFilePrivate() → priv A
 NodeA -> NodeA: LoadKeyFromFilePublic() → pub A
-NodeA -> NodeA: crypto.MarshalPublicKey(pub A) → pubBytes
 
 NodeA -> NodeA: Build PeerRecord A {\n  Name, DID, PubKey,\n  PeerID: PeerID_A,\n  APIUrl: hostname,\n  StreamAddress: /ip4/.../tcp/4001/p2p/PeerID_A,\n  NATSAddress, WalletAddress\n}
 
-NodeA -> NodeA: sha256(json(rec)) → hash
-NodeA -> NodeA: priv.Sign(hash) → signature
+NodeA -> NodeA: priv.Sign(rec) → signature
 NodeA -> NodeA: rec.ExpiryDate = now + 150s
 
-loop Pour chaque StaticIndexer (Indexer A, B, ...)
+loop For every Node Binded Indexer (Indexer A, B, ...)
     NodeA -> IndexerA: TempStream /opencloud/record/publish/1.0
-    NodeA -> IndexerA: json.Encode(PeerRecord A signé)
+    NodeA -> IndexerA: strea!.Encode(Signed PeerRecord A)
 
     IndexerA -> IndexerA: Verify signature
-    IndexerA -> IndexerA: Check heartbeat stream actif pour PeerID_A
+    IndexerA -> IndexerA: Check PeerID_A heartbeat stream 
     IndexerA -> DHT: PutValue("/node/"+DID_A, PeerRecord A)
     DHT --> IndexerA: ok
 end

docs/diagrams/03_indexer_heartbeat.mmd

@@ -1,47 +0,0 @@
-sequenceDiagram
-    title Indexer — Heartbeat double (Pair A + Pair B → Indexer partagé)
-
-    participant NodeA as Node A
-    participant NodeB as Node B
-    participant Indexer as IndexerService (partagé)
-
-    Note over NodeA,NodeB: Chaque pair tick toutes les 20s
-
-    par Pair A heartbeat
-        NodeA->>Indexer: NewStream /opencloud/heartbeat/1.0
-        NodeA->>Indexer: json.Encode(Heartbeat A {Name, DID_A, PeerID_A, IndexersBinded})
-
-        Indexer->>Indexer: CheckHeartbeat(host, stream, streams, mu, maxNodes)
-        Note over Indexer: len(peers) < maxNodes ?
-
-        Indexer->>Indexer: getBandwidthChallenge(512–2048 bytes, stream)
-        Indexer->>NodeA: Write(random payload)
-        NodeA->>Indexer: Echo(same payload)
-        Indexer->>Indexer: Mesure round-trip → Mbps A
-
-        Indexer->>Indexer: getDiversityRate(host, IndexersBinded_A)
-        Note over Indexer: /24 subnet diversity des indexeurs liés
-
-        Indexer->>Indexer: ComputeIndexerScore(uptimeA%, MbpsA%, diversityA%)
-        Note over Indexer: Score = 0.4×uptime + 0.4×bpms + 0.2×diversity
-
-        alt Score A < 75
-            Indexer->>NodeA: (close stream)
-        else Score A ≥ 75
-            Indexer->>Indexer: StreamRecord[PeerID_A] = {DID_A, Heartbeat, UptimeTracker}
-        end
-    and Pair B heartbeat
-        NodeB->>Indexer: NewStream /opencloud/heartbeat/1.0
-        NodeB->>Indexer: json.Encode(Heartbeat B {Name, DID_B, PeerID_B, IndexersBinded})
-
-        Indexer->>Indexer: CheckHeartbeat → getBandwidthChallenge
-        Indexer->>NodeB: Write(random payload)
-        NodeB->>Indexer: Echo(same payload)
-        Indexer->>Indexer: ComputeIndexerScore(uptimeB%, MbpsB%, diversityB%)
-
-        alt Score B ≥ 75
-            Indexer->>Indexer: StreamRecord[PeerID_B] = {DID_B, Heartbeat, UptimeTracker}
-        end
-    end
-
-    Note over Indexer: Les deux pairs sont désormais<br/>enregistrés avec leurs streams actifs

docs/diagrams/03_indexer_heartbeat.puml

@@ -1,49 +1,49 @@
-@startuml
-title Indexer — Heartbeat double (Pair A + Pair B → Indexer partagé)
+@startuml indexer_heartbeat
+title Indexer — Heartbeat node → indexer (score on 5 metrics)
 
 participant "Node A" as NodeA
 participant "Node B" as NodeB
-participant "IndexerService (partagé)" as Indexer
+participant "IndexerService" as Indexer
 
-note over NodeA,NodeB: Chaque pair tick toutes les 20s
+note over NodeA,NodeB: Every node tick every 20s (SendHeartbeat)
 
-par Pair A heartbeat
+par Node A heartbeat
     NodeA -> Indexer: NewStream /opencloud/heartbeat/1.0
-    NodeA -> Indexer: json.Encode(Heartbeat A {Name, DID_A, PeerID_A, IndexersBinded})
+    NodeA -> Indexer: stream.Encode(Heartbeat{Name, PeerID_A, IndexersBinded, Record})
 
-    Indexer -> Indexer: CheckHeartbeat(host, stream, streams, mu, maxNodes)
-    note over Indexer: len(peers) < maxNodes ?
+    Indexer -> Indexer: CheckHeartbeat(host, stream, dec, streams, mu, maxNodes)
+    note over Indexer: len(h.Network().Peers()) >= maxNodes → reject
 
-    Indexer -> Indexer: getBandwidthChallenge(512-2048 bytes, stream)
-    Indexer -> NodeA: Write(random payload)
-    NodeA -> Indexer: Echo(same payload)
-    Indexer -> Indexer: Mesure round-trip → Mbps A
+    Indexer -> Indexer: getBandwidthChallengeRate(host, remotePeer, 512-2048B)
 
-    Indexer -> Indexer: getDiversityRate(host, IndexersBinded_A)
-    note over Indexer: /24 subnet diversity des indexeurs liés
+    Indexer -> Indexer: getOwnDiversityRate(host)\\nh.Network().Peers() + Peerstore.Addrs()\\n→ ratio /24 subnets distincts
 
-    Indexer -> Indexer: ComputeIndexerScore(uptimeA%, MbpsA%, diversityA%)
-    note over Indexer: Score = 0.4×uptime + 0.4×bpms + 0.2×diversity
+    Indexer -> Indexer: fillRate = len(h.Network().Peers()) / maxNodes
 
-    alt Score A < 75
-        Indexer -> NodeA: (close stream)
-    else Score A >= 75
-        Indexer -> Indexer: StreamRecord[PeerID_A] = {DID_A, Heartbeat, UptimeTracker}
+    Indexer -> Indexer: Retrieve existing UptimeTracker\\noldTracker.RecordHeartbeat()\\n→ TotalOnline += gap si gap ≤ 120s\\nuptimeRatio = TotalOnline / time.Since(FirstSeen)
+
+    Indexer -> Indexer: ComputeIndexerScore(\\n  uptimeRatio, bpms, diversity,\\n  latencyScore, fillRate\\n)\\nScore = (0.20×U + 0.20×B + 0.20×D + 0.15×L + 0.25×F) × 100
+
+    Indexer -> Indexer: dynamicMinScore(age)\\n= 20 + 60×(hours/24), max 80
+
+    alt Score A < dynamicMinScore(age)
+        Indexer -> NodeA: (close stream — "not enough trusting value")
+    else Score A >= dynamicMinScore(age)
+        Indexer -> Indexer: streams[PeerID_A].HeartbeatStream = hb.Stream\\nstreams[PeerID_A].HeartbeatStream.UptimeTracker = oldTracker\\nstreams[PeerID_A].LastScore = hb.Score
+        note over Indexer: AfterHeartbeat → republish PeerRecord on DHT
     end
-else Pair B heartbeat
+
+else Node B heartbeat
     NodeB -> Indexer: NewStream /opencloud/heartbeat/1.0
-    NodeB -> Indexer: json.Encode(Heartbeat B {Name, DID_B, PeerID_B, IndexersBinded})
+    NodeB -> Indexer: stream.Encode(Heartbeat{Name, PeerID_B, IndexersBinded, Record})
 
-    Indexer -> Indexer: CheckHeartbeat → getBandwidthChallenge
-    Indexer -> NodeB: Write(random payload)
-    NodeB -> Indexer: Echo(same payload)
-    Indexer -> Indexer: ComputeIndexerScore(uptimeB%, MbpsB%, diversityB%)
+    Indexer -> Indexer: CheckHeartbeat → getBandwidthChallengeRate\\n→ getOwnDiversityRate → ComputeIndexerScore(5 composants)
 
-    alt Score B >= 75
-        Indexer -> Indexer: StreamRecord[PeerID_B] = {DID_B, Heartbeat, UptimeTracker}
+    alt Score B >= dynamicMinScore(age)
+        Indexer -> Indexer: streams[PeerID_B] subscribed + LastScore updated
     end
 end par
 
-note over Indexer: Les deux pairs sont désormais\nenregistrés avec leurs streams actifs
+note over Indexer: GC ticker 30s — gc()\\nnow.After(Expiry) où Expiry = lastHBTime + 2min\\n→ AfterDelete(pid, name, did)
 
 @enduml

docs/diagrams/04_indexer_publish.mmd

@@ -1,41 +0,0 @@
-sequenceDiagram
-    title Indexer — Pair A publie, Pair B publie (handleNodePublish → DHT)
-
-    participant NodeA as Node A
-    participant NodeB as Node B
-    participant Indexer as IndexerService (partagé)
-    participant DHT as DHT Kademlia
-
-    Note over NodeA: Après claimInfo ou refresh TTL
-
-    par Pair A publie son PeerRecord
-        NodeA->>Indexer: TempStream /opencloud/record/publish/1.0
-        NodeA->>Indexer: json.Encode(PeerRecord A {DID_A, PeerID_A, PubKey_A, Expiry, Sig_A})
-
-        Indexer->>Indexer: Verify sig_A (reconstruit rec minimal, pubKey_A.Verify)
-        Indexer->>Indexer: Check StreamRecords[Heartbeat][PeerID_A] existe
-
-        alt Heartbeat actif pour A
-            Indexer->>Indexer: StreamRecord A → DID_A, Record=PeerRecord A, LastSeen=now
-            Indexer->>DHT: PutValue("/node/"+DID_A, PeerRecord A JSON)
-            DHT-->>Indexer: ok
-        else Pas de heartbeat
-            Indexer->>NodeA: (erreur "no heartbeat", stream close)
-        end
-    and Pair B publie son PeerRecord
-        NodeB->>Indexer: TempStream /opencloud/record/publish/1.0
-        NodeB->>Indexer: json.Encode(PeerRecord B {DID_B, PeerID_B, PubKey_B, Expiry, Sig_B})
-
-        Indexer->>Indexer: Verify sig_B
-        Indexer->>Indexer: Check StreamRecords[Heartbeat][PeerID_B] existe
-
-        alt Heartbeat actif pour B
-            Indexer->>Indexer: StreamRecord B → DID_B, Record=PeerRecord B, LastSeen=now
-            Indexer->>DHT: PutValue("/node/"+DID_B, PeerRecord B JSON)
-            DHT-->>Indexer: ok
-        else Pas de heartbeat
-            Indexer->>NodeB: (erreur "no heartbeat", stream close)
-        end
-    end
-
-    Note over DHT: DHT contient maintenant<br/>"/node/DID_A" et "/node/DID_B"

docs/diagrams/04_indexer_publish.puml

@@ -1,43 +1,47 @@
 @startuml
-title Indexer — Pair A publie, Pair B publie (handleNodePublish → DHT)
+title Indexer — Peer A publishing, Peer B publishing (handleNodePublish → DHT)
 
 participant "Node A" as NodeA
 participant "Node B" as NodeB
-participant "IndexerService (partagé)" as Indexer
+participant "IndexerService (shared)" as Indexer
 participant "DHT Kademlia" as DHT
 
-note over NodeA: Après claimInfo ou refresh TTL
+note over NodeA: Start after claimInfo or refresh TTL
 
-par Pair A publie son PeerRecord
+par Peer A publish its PeerRecord
     NodeA -> Indexer: TempStream /opencloud/record/publish/1.0
-    NodeA -> Indexer: json.Encode(PeerRecord A {DID_A, PeerID_A, PubKey_A, Expiry, Sig_A})
+    NodeA -> Indexer: stream.Encode(PeerRecord A {DID_A, PeerID_A, PubKey_A, Expiry, Sig_A})
 
     Indexer -> Indexer: Verify sig_A (reconstruit rec minimal, pubKey_A.Verify)
     Indexer -> Indexer: Check StreamRecords[Heartbeat][PeerID_A] existe
 
-    alt Heartbeat actif pour A
+    alt A active Heartbeat
         Indexer -> Indexer: StreamRecord A → DID_A, Record=PeerRecord A, LastSeen=now
         Indexer -> DHT: PutValue("/node/"+DID_A, PeerRecord A JSON)
+        Indexer -> DHT: PutValue("/name/"+name_A, DID_A)
+        Indexer -> DHT: PutValue("/peer/"+peer_id_A, DID_A)
         DHT --> Indexer: ok
     else Pas de heartbeat
         Indexer -> NodeA: (erreur "no heartbeat", stream close)
     end
-else Pair B publie son PeerRecord
+else Peer B publish its PeerRecord
     NodeB -> Indexer: TempStream /opencloud/record/publish/1.0
-    NodeB -> Indexer: json.Encode(PeerRecord B {DID_B, PeerID_B, PubKey_B, Expiry, Sig_B})
+    NodeB -> Indexer: stream.Encode(PeerRecord B {DID_B, PeerID_B, PubKey_B, Expiry, Sig_B})
 
     Indexer -> Indexer: Verify sig_B
     Indexer -> Indexer: Check StreamRecords[Heartbeat][PeerID_B] existe
 
-    alt Heartbeat actif pour B
+    alt B Active Heartbeat
         Indexer -> Indexer: StreamRecord B → DID_B, Record=PeerRecord B, LastSeen=now
         Indexer -> DHT: PutValue("/node/"+DID_B, PeerRecord B JSON)
+        Indexer -> DHT: PutValue("/name/"+name_B, DID_B)
+        Indexer -> DHT: PutValue("/peer/"+peer_id_B, DID_B)
         DHT --> Indexer: ok
     else Pas de heartbeat
         Indexer -> NodeB: (erreur "no heartbeat", stream close)
     end
 end par
 
-note over DHT: DHT contient maintenant\n"/node/DID_A" et "/node/DID_B"
+note over DHT: DHT got \n"/node/DID_A" et "/node/DID_B"
 
 @enduml

docs/diagrams/05_indexer_get.mmd

@@ -1,49 +0,0 @@
-sequenceDiagram
-    title Indexer — Pair A résout Pair B (GetPeerRecord + handleNodeGet)
-
-    participant NATSA as NATS A
-    participant DBA as DB Pair A (oc-lib)
-    participant NodeA as Node A
-    participant Indexer as IndexerService (partagé)
-    participant DHT as DHT Kademlia
-    participant NATSA2 as NATS A (retour)
-
-    Note over NodeA: Déclenché par : NATS PB_SEARCH PEER<br/>ou callback SubscribeToSearch
-
-    NodeA->>DBA: NewRequestAdmin(PEER).Search(DID_B ou PeerID_B)
-    DBA-->>NodeA: Peer B local (si connu) → résout DID_B + PeerID_B<br/>sinon utilise la valeur brute
-
-    loop Pour chaque StaticIndexer
-        NodeA->>Indexer: TempStream /opencloud/record/get/1.0
-        NodeA->>Indexer: json.Encode(GetValue{Key: DID_B, PeerID: PeerID_B})
-
-        Indexer->>Indexer: key = "/node/" + DID_B
-        Indexer->>DHT: SearchValue(ctx 10s, "/node/"+DID_B)
-        DHT-->>Indexer: channel de bytes (PeerRecord B)
-
-        loop Pour chaque résultat DHT
-            Indexer->>Indexer: Unmarshal → PeerRecord B
-            alt PeerRecord.PeerID == PeerID_B
-                Indexer->>Indexer: resp.Found=true, resp.Records[PeerID_B]=PeerRecord B
-                Indexer->>Indexer: StreamRecord B.LastSeen = now (si heartbeat actif)
-            end
-        end
-
-        Indexer->>NodeA: json.Encode(GetResponse{Found:true, Records:{PeerID_B: PeerRecord B}})
-    end
-
-    loop Pour chaque PeerRecord retourné
-        NodeA->>NodeA: rec.Verify() → valide signature de B
-        NodeA->>NodeA: rec.ExtractPeer(ourDID_A, DID_B, pubKey_B)
-
-        alt ourDID_A == DID_B (c'est notre propre entrée)
-            Note over NodeA: Republier pour rafraîchir le TTL
-            NodeA->>Indexer: publishPeerRecord(rec) [refresh 2 min]
-        end
-
-        NodeA->>NATSA2: SetNATSPub(CREATE_RESOURCE, {PEER, Peer B JSON,<br/>SearchAttr:"peer_id"})
-        NATSA2->>DBA: Upsert Peer B dans DB A
-        DBA-->>NATSA2: ok
-    end
-
-    NodeA-->>NodeA: []*peer.Peer → [Peer B]

docs/diagrams/05_indexer_get.puml

@@ -1,5 +1,5 @@
 @startuml
-title Indexer — Pair A résout Pair B (GetPeerRecord + handleNodeGet)
+title Indexer — Peer A discover Peer B (GetPeerRecord + handleNodeGet)
 
 participant "NATS A" as NATSA
 participant "DB Pair A (oc-lib)" as DBA
@@ -8,41 +8,41 @@ participant "IndexerService (partagé)" as Indexer
 participant "DHT Kademlia" as DHT
 participant "NATS A (retour)" as NATSA2
 
-note over NodeA: Déclenché par : NATS PB_SEARCH PEER\nou callback SubscribeToSearch
+note over NodeA: Trigger : NATS PB_SEARCH PEER\nor callback SubscribeToSearch
 
-NodeA -> DBA: NewRequestAdmin(PEER).Search(DID_B ou PeerID_B)
-DBA --> NodeA: Peer B local (si connu) → résout DID_B + PeerID_B\nsinon utilise la valeur brute
+NodeA -> DBA: (PEER).Search(DID_B or PeerID_B)
+DBA --> NodeA: Local Peer B (if known) → solve DID_B + PeerID_B\nor use search value
 
-loop Pour chaque StaticIndexer
-    NodeA -> Indexer: TempStream /opencloud/record/get/1.0
-    NodeA -> Indexer: json.Encode(GetValue{Key: DID_B, PeerID: PeerID_B})
+loop For every Peer A Binded Indexer 
+    NodeA -> Indexer: TempStream /opencloud/record/get/1.0 -> streamAI
+    NodeA -> Indexer: streamAI.Encode(GetValue{Key: DID_B, PeerID: PeerID_B})
 
     Indexer -> Indexer: key = "/node/" + DID_B
     Indexer -> DHT: SearchValue(ctx 10s, "/node/"+DID_B)
     DHT --> Indexer: channel de bytes (PeerRecord B)
 
-    loop Pour chaque résultat DHT
-        Indexer -> Indexer: Unmarshal → PeerRecord B
+    loop Pour every results in DHT
+        Indexer -> Indexer: read → PeerRecord B
         alt PeerRecord.PeerID == PeerID_B
             Indexer -> Indexer: resp.Found=true, resp.Records[PeerID_B]=PeerRecord B
-            Indexer -> Indexer: StreamRecord B.LastSeen = now (si heartbeat actif)
+            Indexer -> Indexer: StreamRecord B.LastSeen = now (if active heartbeat)
         end
     end
 
-    Indexer -> NodeA: json.Encode(GetResponse{Found:true, Records:{PeerID_B: PeerRecord B}})
+    Indexer -> NodeA: streamAI.Encode(GetResponse{Found:true, Records:{PeerID_B: PeerRecord B}})
 end
 
-loop Pour chaque PeerRecord retourné
-    NodeA -> NodeA: rec.Verify() → valide signature de B
+loop For every PeerRecord founded
+    NodeA -> NodeA: rec.Verify() → valid B signature
     NodeA -> NodeA: rec.ExtractPeer(ourDID_A, DID_B, pubKey_B)
 
-    alt ourDID_A == DID_B (c'est notre propre entrée)
-        note over NodeA: Republier pour rafraîchir le TTL
+    alt ourDID_A == DID_B (it's our proper entry)
+        note over NodeA: Republish to refresh TTL
         NodeA -> Indexer: publishPeerRecord(rec) [refresh 2 min]
     end
 
     NodeA -> NATSA2: SetNATSPub(CREATE_RESOURCE, {PEER, Peer B JSON,\nSearchAttr:"peer_id"})
-    NATSA2 -> DBA: Upsert Peer B dans DB A
+    NATSA2 -> DBA: Upsert Peer B in DB A
     DBA --> NATSA2: ok
 end
 

docs/diagrams/06_native_registration.mmd

@@ -1,39 +0,0 @@
-sequenceDiagram
-    title Native Indexer — Enregistrement d'un Indexer auprès du Native
-
-    participant IndexerA as Indexer A
-    participant IndexerB as Indexer B
-    participant Native as Native Indexer (partagé)
-    participant DHT as DHT Kademlia
-    participant PubSub as GossipSub (oc-indexer-registry)
-
-    Note over IndexerA,IndexerB: Au démarrage + toutes les 60s (StartNativeRegistration)
-
-    par Indexer A s'enregistre
-        IndexerA->>IndexerA: Build IndexerRegistration{PeerID_A, Addr_A}
-        IndexerA->>Native: NewStream /opencloud/native/subscribe/1.0
-        IndexerA->>Native: json.Encode(IndexerRegistration A)
-
-        Native->>Native: Decode → liveIndexerEntry{PeerID_A, Addr_A, ExpiresAt=now+66s}
-        Native->>DHT: PutValue("/indexer/"+PeerID_A, entry A)
-        DHT-->>Native: ok
-        Native->>Native: liveIndexers[PeerID_A] = entry A
-        Native->>Native: knownPeerIDs[PeerID_A] = {}
-
-        Native->>PubSub: topic.Publish([]byte(PeerID_A))
-        Note over PubSub: Gossipé aux autres Natives<br/>→ ils ajoutent PeerID_A à knownPeerIDs<br/>→ refresh DHT au prochain tick 30s
-        IndexerA->>Native: stream.Close()
-    and Indexer B s'enregistre
-        IndexerB->>IndexerB: Build IndexerRegistration{PeerID_B, Addr_B}
-        IndexerB->>Native: NewStream /opencloud/native/subscribe/1.0
-        IndexerB->>Native: json.Encode(IndexerRegistration B)
-
-        Native->>Native: Decode → liveIndexerEntry{PeerID_B, Addr_B, ExpiresAt=now+66s}
-        Native->>DHT: PutValue("/indexer/"+PeerID_B, entry B)
-        DHT-->>Native: ok
-        Native->>Native: liveIndexers[PeerID_B] = entry B
-        Native->>PubSub: topic.Publish([]byte(PeerID_B))
-        IndexerB->>Native: stream.Close()
-    end
-
-    Note over Native: liveIndexers = {PeerID_A: entryA, PeerID_B: entryB}

docs/diagrams/06_native_registration.puml

@@ -1,41 +1,49 @@
-@startuml
-title Native Indexer — Enregistrement d'un Indexer auprès du Native
+@startuml native_registration
+title Native Indexer — Indexer Subscription (StartNativeRegistration)
 
 participant "Indexer A" as IndexerA
 participant "Indexer B" as IndexerB
-participant "Native Indexer (partagé)" as Native
+participant "Native Indexer" as Native
 participant "DHT Kademlia" as DHT
 participant "GossipSub (oc-indexer-registry)" as PubSub
 
-note over IndexerA,IndexerB: Au démarrage + toutes les 60s (StartNativeRegistration)
+note over IndexerA,IndexerB: At start + every 60s (RecommendedHeartbeatInterval)\\nStartNativeRegistration → RegisterWithNative
+
+par Indexer A subscribe
+    IndexerA -> IndexerA: fillRateFn()\\n= len(StreamRecords[HB]) / maxNodes
+
+    IndexerA -> IndexerA: Build IndexerRegistration{\\n  PeerID_A, Addr_A,\\n  Timestamp=now.UnixNano(),\\n  FillRate=fillRateFn(),\\n  PubKey, Signature\\n}\\nreg.Sign(h)
 
-par Indexer A s'enregistre
-    IndexerA -> IndexerA: Build IndexerRegistration{PeerID_A, Addr_A}
     IndexerA -> Native: NewStream /opencloud/native/subscribe/1.0
-    IndexerA -> Native: json.Encode(IndexerRegistration A)
+    IndexerA -> Native: stream.Encode(IndexerRegistration A)
+
+    Native -> Native: reg.Verify() — verify signature
+    Native -> Native: liveIndexerEntry{\\n  PeerID_A, Addr_A,\\n  ExpiresAt = now + IndexerTTL (90s),\\n  FillRate = reg.FillRate,\\n  PubKey, Signature\\n}
+    Native -> Native: liveIndexers[PeerID_A] = entry A
+    Native -> Native: knownPeerIDs[PeerID_A] = Addr_A
 
-    Native -> Native: Decode → liveIndexerEntry{PeerID_A, Addr_A, ExpiresAt=now+66s}
     Native -> DHT: PutValue("/indexer/"+PeerID_A, entry A)
     DHT --> Native: ok
-    Native -> Native: liveIndexers[PeerID_A] = entry A
-    Native -> Native: knownPeerIDs[PeerID_A] = {}
 
     Native -> PubSub: topic.Publish([]byte(PeerID_A))
-    note over PubSub: Gossipé aux autres Natives\n→ ils ajoutent PeerID_A à knownPeerIDs\n→ refresh DHT au prochain tick 30s
-    IndexerA -> Native: stream.Close()
-else Indexer B s'enregistre
-    IndexerB -> IndexerB: Build IndexerRegistration{PeerID_B, Addr_B}
-    IndexerB -> Native: NewStream /opencloud/native/subscribe/1.0
-    IndexerB -> Native: json.Encode(IndexerRegistration B)
+    note over PubSub: Gossip to other Natives\\n→ it adds PeerID_A to knownPeerIDs\\n→ refresh DHT next tick (30s)
 
-    Native -> Native: Decode → liveIndexerEntry{PeerID_B, Addr_B, ExpiresAt=now+66s}
-    Native -> DHT: PutValue("/indexer/"+PeerID_B, entry B)
-    DHT --> Native: ok
+    IndexerA -> Native: stream.Close()
+
+else Indexer B subscribe
+    IndexerB -> IndexerB: fillRateFn() + reg.Sign(h)
+    IndexerB -> Native: NewStream /opencloud/native/subscribe/1.0
+    IndexerB -> Native: stream.Encode(IndexerRegistration B)
+
+    Native -> Native: reg.Verify() + liveIndexerEntry{FillRate=reg.FillRate, ExpiresAt=now+90s}
     Native -> Native: liveIndexers[PeerID_B] = entry B
+    Native -> DHT: PutValue("/indexer/"+PeerID_B, entry B)
     Native -> PubSub: topic.Publish([]byte(PeerID_B))
     IndexerB -> Native: stream.Close()
 end par
 
-note over Native: liveIndexers = {PeerID_A: entryA, PeerID_B: entryB}
+note over Native: liveIndexers = {PeerID_A: {FillRate:0.3}, PeerID_B: {FillRate:0.6}}\\nTTL 90s — IndexerTTL
+
+note over Native: Explicit unsubcrive on stop :\\nUnregisterFromNative → /opencloud/native/unsubscribe/1.0\\nNative close all now.
 
 @enduml

docs/diagrams/07_native_get_consensus.mmd

@@ -1,60 +0,0 @@
-sequenceDiagram
-    title Native — ConnectToNatives + Consensus (Pair A bootstrap)
-
-    participant NodeA as Node A
-    participant Native1 as Native #1 (primary)
-    participant Native2 as Native #2
-    participant NativeN as Native #N
-    participant DHT as DHT Kademlia
-
-    Note over NodeA: NativeIndexerAddresses configuré<br/>Appelé pendant InitNode → ConnectToIndexers
-
-    NodeA->>NodeA: Parse NativeIndexerAddresses → StaticNatives
-    NodeA->>Native1: SendHeartbeat /opencloud/heartbeat/1.0 (20s tick)
-    NodeA->>Native2: SendHeartbeat /opencloud/heartbeat/1.0 (20s tick)
-
-    %% Étape 1 : récupérer un pool initial
-    NodeA->>Native1: Connect + NewStream /opencloud/native/indexers/1.0
-    NodeA->>Native1: json.Encode(GetIndexersRequest{Count: maxIndexer})
-
-    Native1->>Native1: reachableLiveIndexers()
-    Note over Native1: Filtre liveIndexers par TTL<br/>ping chaque candidat (PeerIsAlive)
-
-    alt Aucun indexer connu par Native1
-        Native1->>Native1: selfDelegate(NodeA.PeerID, resp)
-        Note over Native1: IsSelfFallback=true<br/>Indexers=[native1 addr]
-        Native1->>NodeA: GetIndexersResponse{IsSelfFallback:true, Indexers:[native1]}
-        NodeA->>NodeA: StaticIndexers[native1] = native1
-        Note over NodeA: Pas de consensus — native1 utilisé directement comme indexeur
-    else Indexers disponibles
-        Native1->>NodeA: GetIndexersResponse{Indexers:[Addr_IndexerA, Addr_IndexerB, ...]}
-
-        %% Étape 2 : consensus
-        Note over NodeA: clientSideConsensus(candidates)
-
-        par Requêtes consensus parallèles
-            NodeA->>Native1: NewStream /opencloud/native/consensus/1.0
-            NodeA->>Native1: ConsensusRequest{Candidates:[Addr_A, Addr_B]}
-            Native1->>Native1: Croiser avec liveIndexers propres
-            Native1->>NodeA: ConsensusResponse{Trusted:[Addr_A, Addr_B], Suggestions:[]}
-        and
-            NodeA->>Native2: NewStream /opencloud/native/consensus/1.0
-            NodeA->>Native2: ConsensusRequest{Candidates:[Addr_A, Addr_B]}
-            Native2->>Native2: Croiser avec liveIndexers propres
-            Native2->>NodeA: ConsensusResponse{Trusted:[Addr_A], Suggestions:[Addr_C]}
-        and
-            NodeA->>NativeN: NewStream /opencloud/native/consensus/1.0
-            NodeA->>NativeN: ConsensusRequest{Candidates:[Addr_A, Addr_B]}
-            NativeN->>NativeN: Croiser avec liveIndexers propres
-            NativeN->>NodeA: ConsensusResponse{Trusted:[Addr_A, Addr_B], Suggestions:[]}
-        end
-
-        Note over NodeA: Aggrège les votes (timeout 4s)<br/>Addr_A → 3/3 votes → confirmé ✓<br/>Addr_B → 2/3 votes → confirmé ✓
-
-        alt confirmed < maxIndexer && suggestions disponibles
-            Note over NodeA: Round 2 — rechallenge avec suggestions
-            NodeA->>NodeA: clientSideConsensus(confirmed + sample(suggestions))
-        end
-
-        NodeA->>NodeA: StaticIndexers = adresses confirmées à majorité
-    end

docs/diagrams/07_native_get_consensus.puml

@@ -1,62 +1,70 @@
-@startuml
-title Native — ConnectToNatives + Consensus (Pair A bootstrap)
+@startuml native_get_consensus
+title Native — ConnectToNatives : fetch pool + Phase 1 + Phase 2
 
-participant "Node A" as NodeA
-participant "Native #1 (primary)" as Native1
-participant "Native #2" as Native2
-participant "Native #N" as NativeN
-participant "DHT Kademlia" as DHT
+participant "Node / Indexer\\n(appelant)" as Caller
+participant "Native A" as NA
+participant "Native B" as NB
+participant "Indexer A\\n(stable voter)" as IA
 
-note over NodeA: NativeIndexerAddresses configuré\nAppelé pendant InitNode → ConnectToIndexers
+note over Caller: NativeIndexerAddresses configured\\nConnectToNatives() called from ConnectToIndexers
 
-NodeA -> NodeA: Parse NativeIndexerAddresses → StaticNatives
-NodeA -> Native1: SendHeartbeat /opencloud/heartbeat/1.0 (20s tick)
-NodeA -> Native2: SendHeartbeat /opencloud/heartbeat/1.0 (20s tick)
+== Step 1 : heartbeat to the native mesh (nativeHeartbeatOnce) ==
+Caller -> NA: SendHeartbeat /opencloud/heartbeat/1.0 
+Caller -> NB: SendHeartbeat /opencloud/heartbeat/1.0 
 
-' Étape 1 : récupérer un pool initial
-NodeA -> Native1: Connect + NewStream /opencloud/native/indexers/1.0
-NodeA -> Native1: json.Encode(GetIndexersRequest{Count: maxIndexer})
-
-Native1 -> Native1: reachableLiveIndexers()
-note over Native1: Filtre liveIndexers par TTL\nping chaque candidat (PeerIsAlive)
-
-alt Aucun indexer connu par Native1
-    Native1 -> Native1: selfDelegate(NodeA.PeerID, resp)
-    note over Native1: IsSelfFallback=true\nIndexers=[native1 addr]
-    Native1 -> NodeA: GetIndexersResponse{IsSelfFallback:true, Indexers:[native1]}
-    NodeA -> NodeA: StaticIndexers[native1] = native1
-    note over NodeA: Pas de consensus — native1 utilisé directement comme indexeur
-else Indexers disponibles
-    Native1 -> NodeA: GetIndexersResponse{Indexers:[Addr_IndexerA, Addr_IndexerB, ...]}
-
-    ' Étape 2 : consensus
-    note over NodeA: clientSideConsensus(candidates)
-
-    par Requêtes consensus parallèles
-        NodeA -> Native1: NewStream /opencloud/native/consensus/1.0
-        NodeA -> Native1: ConsensusRequest{Candidates:[Addr_A, Addr_B]}
-        Native1 -> Native1: Croiser avec liveIndexers propres
-        Native1 -> NodeA: ConsensusResponse{Trusted:[Addr_A, Addr_B], Suggestions:[]}
+== Step 2 : parrallel fetch pool (timeout 6s) ==
+par fetchIndexersFromNative — parallel
+    Caller -> NA: NewStream /opencloud/native/indexers/1.0\\nGetIndexersRequest{Count: maxIndexer, From: PeerID}
+    NA -> NA: reachableLiveIndexers()\\ntri par w(F) = fillRate×(1−fillRate) desc
+    NA --> Caller: GetIndexersResponse{Indexers:[IA,IB], FillRates:{IA:0.3,IB:0.6}}
 else
-        NodeA -> Native2: NewStream /opencloud/native/consensus/1.0
-        NodeA -> Native2: ConsensusRequest{Candidates:[Addr_A, Addr_B]}
-        Native2 -> Native2: Croiser avec liveIndexers propres
-        Native2 -> NodeA: ConsensusResponse{Trusted:[Addr_A], Suggestions:[Addr_C]}
-    else
-        NodeA -> NativeN: NewStream /opencloud/native/consensus/1.0
-        NodeA -> NativeN: ConsensusRequest{Candidates:[Addr_A, Addr_B]}
-        NativeN -> NativeN: Croiser avec liveIndexers propres
-        NativeN -> NodeA: ConsensusResponse{Trusted:[Addr_A, Addr_B], Suggestions:[]}
+    Caller -> NB: NewStream /opencloud/native/indexers/1.0
+    NB -> NB: reachableLiveIndexers()
+    NB --> Caller: GetIndexersResponse{Indexers:[IA,IB], FillRates:{IA:0.3,IB:0.6}}
 end par
 
-    note over NodeA: Aggrège les votes (timeout 4s)\nAddr_A → 3/3 votes → confirmé ✓\nAddr_B → 2/3 votes → confirmé ✓
+note over Caller: Fusion → candidates=[IA,IB]\\nisFallback=false
 
-    alt confirmed < maxIndexer && suggestions disponibles
-        note over NodeA: Round 2 — rechallenge avec suggestions
-        NodeA -> NodeA: clientSideConsensus(confirmed + sample(suggestions))
+alt isFallback=true (native give themself as Fallback indexer)
+    note over Caller: resolvePool : avoid consensus\\nadmittedAt = Now (zero)\\nStaticIndexers = {native_addr}
+else isFallback=false → Phase 1 + Phase 2
+    == Phase 1 — clientSideConsensus (timeout 3s/natif, 4s total) ==
+    par Parralel Consensus
+        Caller -> NA: NewStream /opencloud/native/consensus/1.0\\nConsensusRequest{Candidates:[IA,IB]}
+        NA -> NA: compare with clean liveIndexers
+        NA --> Caller: ConsensusResponse{Trusted:[IA,IB], Suggestions:[]}
+    else
+        Caller -> NB: NewStream /opencloud/native/consensus/1.0
+        NB --> Caller: ConsensusResponse{Trusted:[IA], Suggestions:[IC]}
+    end par
+
+    note over Caller: IA → 2/2 votes → confirmed ✓\\nIB → 1/2 vote → refusé ✗\\nIC → suggestion → round 2 if confirmed < maxIndexer
+
+    alt confirmed < maxIndexer && available suggestions 
+        note over Caller: Round 2 — rechallenge with confirmed + sample(suggestions)\\nclientSideConsensus([IA, IC])
     end
 
-    NodeA -> NodeA: StaticIndexers = adresses confirmées à majorité
+    note over Caller: admittedAt = time.Now()
+
+    == Phase 2 — indexerLivenessVote (timeout 3s/votant, 4s total) ==
+    note over Caller: Search for stable voters in Subscribed Indexers\\nAdmittedAt != zero && age >= MinStableAge (2min)
+
+    alt Stable Voters are available
+        par Phase 2 parrallel
+            Caller -> IA: NewStream /opencloud/indexer/consensus/1.0\\nIndexerConsensusRequest{Candidates:[IA]}
+            IA -> IA: StreamRecords[ProtocolHB][candidate]\\ntime.Since(LastSeen) <= 120s && LastScore >= 30.0
+            IA --> Caller: IndexerConsensusResponse{Alive:[IA]}
+        end par
+        note over Caller: alive IA confirmed per quorum > 0.5\\npool = {IA}
+    else No voters are stable (startup)
+        note over Caller: Phase 1 keep directly\\n(no indexer reaches MinStableAge)
     end
 
+    == Replacement pool ==
+    Caller -> Caller: replaceStaticIndexers(pool, admittedAt)\\nStaticIndexerMeta[IA].AdmittedAt = admittedAt
+end
+
+== Étape 3 : heartbeat to indexers pool (ConnectToIndexers) ==
+Caller -> Caller: SendHeartbeat /opencloud/heartbeat/1.0\\nvers StaticIndexers
+
 @enduml

docs/diagrams/08_nats_create_resource.mmd

@@ -1,49 +0,0 @@
-sequenceDiagram
-    title NATS — CREATE_RESOURCE : Pair A découvre Pair B et établit le stream
-
-    participant AppA as App Pair A (oc-api)
-    participant NATSA as NATS A
-    participant NodeA as Node A
-    participant StreamA as StreamService A
-    participant NodeB as Node B
-    participant StreamB as StreamService B
-    participant DBA as DB Pair A (oc-lib)
-
-    Note over AppA: Pair B vient d'être découvert<br/>(via indexeur ou manuel)
-
-    AppA->>NATSA: Publish(CREATE_RESOURCE, {<br/>  FromApp:"oc-api",<br/>  Datatype:PEER,<br/>  Payload: Peer B {StreamAddress_B, Relation:PARTNER}<br/>})
-
-    NATSA->>NodeA: ListenNATS callback → CREATE_RESOURCE
-
-    NodeA->>NodeA: resp.FromApp == "oc-discovery" ? → Non, continuer
-    NodeA->>NodeA: json.Unmarshal(payload) → peer.Peer B
-    NodeA->>NodeA: pp.AddrInfoFromString(B.StreamAddress)
-    Note over NodeA: ad_B = {ID: PeerID_B, Addrs: [...]}
-
-    NodeA->>StreamA: Mu.Lock()
-
-    alt peer B.Relation == PARTNER
-        NodeA->>StreamA: ConnectToPartner(B.StreamAddress)
-        StreamA->>StreamA: AddrInfoFromString(B.StreamAddress) → ad_B
-        StreamA->>NodeB: Connect (libp2p)
-        StreamA->>NodeB: NewStream /opencloud/resource/heartbeat/partner/1.0
-        StreamA->>NodeB: json.Encode(Heartbeat{Name_A, DID_A, PeerID_A})
-
-        NodeB->>StreamB: HandlePartnerHeartbeat(stream)
-        StreamB->>StreamB: CheckHeartbeat → bandwidth challenge
-        StreamB->>StreamA: Echo(payload)
-        StreamB->>StreamB: streams[ProtocolHeartbeatPartner][PeerID_A] = {DID_A, Expiry=now+10s}
-
-        StreamA->>StreamA: streams[ProtocolHeartbeatPartner][PeerID_B] = {DID_B, Expiry=now+10s}
-        Note over StreamA,StreamB: Stream partner long-lived établi<br/>dans les deux sens
-
-    else peer B.Relation != PARTNER (révocation / blacklist)
-        Note over NodeA: Supprimer tous les streams vers Pair B
-        loop Pour chaque protocole dans Streams
-            NodeA->>StreamA: streams[proto][PeerID_B].Stream.Close()
-            NodeA->>StreamA: delete(streams[proto], PeerID_B)
-        end
-    end
-
-    NodeA->>StreamA: Mu.Unlock()
-    NodeA->>DBA: (pas de write direct ici — géré par l'app source)

docs/diagrams/08_nats_create_update_peer.puml

@@ -1,50 +1,42 @@
 @startuml
-title NATS — CREATE_RESOURCE : Pair A découvre Pair B et établit le stream
+title NATS — CREATE_RESOURCE : Peer A Create/Update Peer B & establishing stream
 
-participant "App Pair A (oc-api)" as AppA
+participant "App Peer A (oc-api)" as AppA
 participant "NATS A" as NATSA
 participant "Node A" as NodeA
 participant "StreamService A" as StreamA
 participant "Node B" as NodeB
 participant "StreamService B" as StreamB
-participant "DB Pair A (oc-lib)" as DBA
+participant "DB Peer A (oc-lib)" as DBA
 
-note over AppA: Pair B vient d'être découvert\n(via indexeur ou manuel)
+note over AppA: Peer B is discovered\n(per indexer or manually)
 
 AppA -> NATSA: Publish(CREATE_RESOURCE, {\n  FromApp:"oc-api",\n  Datatype:PEER,\n  Payload: Peer B {StreamAddress_B, Relation:PARTNER}\n})
 
 NATSA -> NodeA: ListenNATS callback → CREATE_RESOURCE
 
-NodeA -> NodeA: resp.FromApp == "oc-discovery" ? → Non, continuer
+NodeA -> NodeA: if from himself ? → No, continue
 NodeA -> NodeA: json.Unmarshal(payload) → peer.Peer B
-NodeA -> NodeA: pp.AddrInfoFromString(B.StreamAddress)
-note over NodeA: ad_B = {ID: PeerID_B, Addrs: [...]}
-
-NodeA -> StreamA: Mu.Lock()
 
 alt peer B.Relation == PARTNER
     NodeA -> StreamA: ConnectToPartner(B.StreamAddress)
-    StreamA -> StreamA: AddrInfoFromString(B.StreamAddress) → ad_B
     StreamA -> NodeB: Connect (libp2p)
     StreamA -> NodeB: NewStream /opencloud/resource/heartbeat/partner/1.0
     StreamA -> NodeB: json.Encode(Heartbeat{Name_A, DID_A, PeerID_A})
 
     NodeB -> StreamB: HandlePartnerHeartbeat(stream)
     StreamB -> StreamB: CheckHeartbeat → bandwidth challenge
-    StreamB -> StreamA: Echo(payload)
     StreamB -> StreamB: streams[ProtocolHeartbeatPartner][PeerID_A] = {DID_A, Expiry=now+10s}
 
     StreamA -> StreamA: streams[ProtocolHeartbeatPartner][PeerID_B] = {DID_B, Expiry=now+10s}
-    note over StreamA,StreamB: Stream partner long-lived établi\ndans les deux sens
-else peer B.Relation != PARTNER (révocation / blacklist)
-    note over NodeA: Supprimer tous les streams vers Pair B
-    loop Pour chaque protocole dans Streams
+    note over StreamA,StreamB: Stream partner long-lived établi\nbi-directionnal
+else peer B.Relation != PARTNER (revoke / blacklist)
+    note over NodeA: Suppress all streams onto Peer B
+    loop For every Streams
         NodeA -> StreamA: streams[proto][PeerID_B].Stream.Close()
         NodeA -> StreamA: delete(streams[proto], PeerID_B)
     end
 end
-
-NodeA -> StreamA: Mu.Unlock()
-NodeA -> DBA: (pas de write direct ici — géré par l'app source)
+NodeA -> DBA: (no write — only app source manually add peer)
 
 @enduml

docs/diagrams/09_nats_propagation.mmd

@@ -1,66 +0,0 @@
-sequenceDiagram
-    title NATS — PROPALGATION_EVENT : Pair A propage vers Pair B
-
-    participant AppA as App Pair A
-    participant NATSA as NATS A
-    participant NodeA as Node A
-    participant StreamA as StreamService A
-    participant NodeB as Node B
-    participant NATSB as NATS B
-    participant DBB as DB Pair B (oc-lib)
-
-    AppA->>NATSA: Publish(PROPALGATION_EVENT, {Action, DataType, Payload})
-    NATSA->>NodeA: ListenNATS callback → PROPALGATION_EVENT
-    NodeA->>NodeA: resp.FromApp != "oc-discovery" ? → continuer
-    NodeA->>NodeA: json.Unmarshal → PropalgationMessage{Action, DataType, Payload}
-
-    alt Action == PB_DELETE
-        NodeA->>StreamA: ToPartnerPublishEvent(PB_DELETE, dt, user, payload)
-        StreamA->>StreamA: searchPeer(PARTNER) → [Pair B, ...]
-        StreamA->>NodeB: write(PeerID_B, addr_B, dt, user, payload, ProtocolDeleteResource)
-        Note over NodeB: /opencloud/resource/delete/1.0
-
-        NodeB->>NodeB: handleEventFromPartner(evt, ProtocolDeleteResource)
-        NodeB->>NATSB: SetNATSPub(REMOVE_RESOURCE, {DataType, resource JSON})
-        NATSB->>DBB: Supprimer ressource dans DB B
-
-    else Action == PB_UPDATE (via ProtocolUpdateResource)
-        NodeA->>StreamA: ToPartnerPublishEvent(PB_UPDATE, dt, user, payload)
-        StreamA->>NodeB: write → /opencloud/resource/update/1.0
-        NodeB->>NATSB: SetNATSPub(CREATE_RESOURCE, {DataType, resource JSON})
-        NATSB->>DBB: Upsert ressource dans DB B
-
-    else Action == PB_CONSIDERS + WORKFLOW_EXECUTION
-        NodeA->>NodeA: Unmarshal → executionConsidersPayload{PeerIDs:[PeerID_B, ...]}
-        loop Pour chaque peer_id cible
-            NodeA->>StreamA: PublishCommon(dt, user, PeerID_B, ProtocolConsidersResource, payload)
-            StreamA->>NodeB: write → /opencloud/resource/considers/1.0
-            NodeB->>NodeB: passConsidering(evt)
-            NodeB->>NATSB: SetNATSPub(PROPALGATION_EVENT, {PB_CONSIDERS, dt, payload})
-            NATSB->>DBB: (traité par oc-workflow sur NATS B)
-        end
-
-    else Action == PB_PLANNER (broadcast)
-        NodeA->>NodeA: Unmarshal → {peer_id: nil, ...payload}
-        loop Pour chaque stream ProtocolSendPlanner ouvert
-            NodeA->>StreamA: PublishCommon(nil, user, pid, ProtocolSendPlanner, payload)
-            StreamA->>NodeB: write → /opencloud/resource/planner/1.0
-        end
-
-    else Action == PB_CLOSE_PLANNER
-        NodeA->>NodeA: Unmarshal → {peer_id: PeerID_B}
-        NodeA->>StreamA: Streams[ProtocolSendPlanner][PeerID_B].Stream.Close()
-        NodeA->>StreamA: delete(Streams[ProtocolSendPlanner], PeerID_B)
-
-    else Action == PB_SEARCH + DataType == PEER
-        NodeA->>NodeA: Unmarshal → {search: "..."}
-        NodeA->>NodeA: GetPeerRecord(ctx, search)
-        Note over NodeA: Résolution via DB A + Indexer + DHT
-        NodeA->>NATSA: SetNATSPub(SEARCH_EVENT, {PEER, PeerRecord JSON})
-        NATSA->>NATSA: (AppA reçoit le résultat)
-
-    else Action == PB_SEARCH + autre DataType
-        NodeA->>NodeA: Unmarshal → {type:"all"|"known"|"partner", search:"..."}
-        NodeA->>NodeA: PubSubService.SearchPublishEvent(ctx, dt, type, user, search)
-        Note over NodeA: Voir diagrammes 10 et 11
-    end

docs/diagrams/09_nats_propagation.puml

@@ -1,50 +1,55 @@
 @startuml
-title NATS — PROPALGATION_EVENT : Pair A propage vers Pair B
+title NATS — PROPALGATION_EVENT : Peer A propalgate to Peer B lookup
 
 participant "App Pair A" as AppA
 participant "NATS A" as NATSA
 participant "Node A" as NodeA
 participant "StreamService A" as StreamA
-participant "Node B" as NodeB
+participant "Node Partner B" as PeerB
+participant "Node C" as PeerC
+
 participant "NATS B" as NATSB
 participant "DB Pair B (oc-lib)" as DBB
 
+note over App: only our proper resource (db data) can be propalgate : creator_id==self
+
 AppA -> NATSA: Publish(PROPALGATION_EVENT, {Action, DataType, Payload})
 NATSA -> NodeA: ListenNATS callback → PROPALGATION_EVENT
-NodeA -> NodeA: resp.FromApp != "oc-discovery" ? → continuer
+NodeA -> NodeA: propalgate from himself ? → no, continue
 NodeA -> NodeA: json.Unmarshal → PropalgationMessage{Action, DataType, Payload}
 
 alt Action == PB_DELETE
     NodeA -> StreamA: ToPartnerPublishEvent(PB_DELETE, dt, user, payload)
-    StreamA -> StreamA: searchPeer(PARTNER) → [Pair B, ...]
+    StreamA -> StreamA: searchPeer(PARTNER) → [Peer Partner B, ...]
     StreamA -> NodeB: write(PeerID_B, addr_B, dt, user, payload, ProtocolDeleteResource)
     note over NodeB: /opencloud/resource/delete/1.0
 
     NodeB -> NodeB: handleEventFromPartner(evt, ProtocolDeleteResource)
     NodeB -> NATSB: SetNATSPub(REMOVE_RESOURCE, {DataType, resource JSON})
-    NATSB -> DBB: Supprimer ressource dans DB B
+    NATSB -> DBB: Suppress ressource into DB B
 
-else Action == PB_UPDATE (via ProtocolUpdateResource)
+else Action == PB_UPDATE (per ProtocolUpdateResource)
     NodeA -> StreamA: ToPartnerPublishEvent(PB_UPDATE, dt, user, payload)
+    StreamA -> StreamA: searchPeer(PARTNER) → [Peer Partner B, ...]
     StreamA -> NodeB: write → /opencloud/resource/update/1.0
     NodeB -> NATSB: SetNATSPub(CREATE_RESOURCE, {DataType, resource JSON})
     NATSB -> DBB: Upsert ressource dans DB B
 
-else Action == PB_CONSIDERS + WORKFLOW_EXECUTION
+else Action == PB_CREATE (per ProtocolCreateResource)
+    NodeA -> StreamA: ToPartnerPublishEvent(PB_UPDATE, dt, user, payload)
+    StreamA -> StreamA: searchPeer(PARTNER) → [Peer Partner B, ...]
+    StreamA -> NodeB: write → /opencloud/resource/create/1.0
+    NodeB -> NATSB: SetNATSPub(CREATE_RESOURCE, {DataType, resource JSON})
+    NATSB -> DBB: Create ressource dans DB B
+
+else Action == PB_CONSIDERS (is a considering a previous action, such as planning or creating resource)
     NodeA -> NodeA: Unmarshal → executionConsidersPayload{PeerIDs:[PeerID_B, ...]}
-    loop Pour chaque peer_id cible
+    loop For every peer_id targeted
         NodeA -> StreamA: PublishCommon(dt, user, PeerID_B, ProtocolConsidersResource, payload)
         StreamA -> NodeB: write → /opencloud/resource/considers/1.0
         NodeB -> NodeB: passConsidering(evt)
         NodeB -> NATSB: SetNATSPub(PROPALGATION_EVENT, {PB_CONSIDERS, dt, payload})
-        NATSB -> DBB: (traité par oc-workflow sur NATS B)
-    end
-
-else Action == PB_PLANNER (broadcast)
-    NodeA -> NodeA: Unmarshal → {peer_id: nil, ...payload}
-    loop Pour chaque stream ProtocolSendPlanner ouvert
-        NodeA -> StreamA: PublishCommon(nil, user, pid, ProtocolSendPlanner, payload)
-        StreamA -> NodeB: write → /opencloud/resource/planner/1.0
+        NATSB -> DBB: (treat per emmitters app of a previous action on NATS B)
     end
 
 else Action == PB_CLOSE_PLANNER
@@ -53,16 +58,16 @@ else Action == PB_CLOSE_PLANNER
     NodeA -> StreamA: delete(Streams[ProtocolSendPlanner], PeerID_B)
 
 else Action == PB_SEARCH + DataType == PEER
-    NodeA -> NodeA: Unmarshal → {search: "..."}
+    NodeA -> NodeA: read → {search: "..."}
     NodeA -> NodeA: GetPeerRecord(ctx, search)
-    note over NodeA: Résolution via DB A + Indexer + DHT
+    note over NodeA: Resolved per DB A or Indexer + DHT
     NodeA -> NATSA: SetNATSPub(SEARCH_EVENT, {PEER, PeerRecord JSON})
-    NATSA -> NATSA: (AppA reçoit le résultat)
+    NATSA -> NATSA: (AppA retrieve results)
 
-else Action == PB_SEARCH + autre DataType
-    NodeA -> NodeA: Unmarshal → {type:"all"|"known"|"partner", search:"..."}
+else Action == PB_SEARCH + other DataType
+    NodeA -> NodeA: read → {type:"all"|"known"|"partner", search:"..."}
     NodeA -> NodeA: PubSubService.SearchPublishEvent(ctx, dt, type, user, search)
-    note over NodeA: Voir diagrammes 10 et 11
+    note over NodeA: Watch after pubsub_search & stream_search diagrams
 end
 
 @enduml

docs/diagrams/10_pubsub_search.mmd

@@ -1,52 +0,0 @@
-sequenceDiagram
-    title PubSub — Recherche gossip globale (type "all") : Pair A cherche, Pair B répond
-
-    participant AppA as App Pair A
-    participant NATSA as NATS A
-    participant NodeA as Node A
-    participant PubSubA as PubSubService A
-    participant GossipSub as GossipSub libp2p (mesh)
-    participant NodeB as Node B
-    participant PubSubB as PubSubService B
-    participant DBB as DB Pair B (oc-lib)
-    participant StreamB as StreamService B
-    participant StreamA as StreamService A
-
-    AppA->>NATSA: Publish(PROPALGATION_EVENT, {PB_SEARCH, type:"all", search:"gpu"})
-    NATSA->>NodeA: ListenNATS → PB_SEARCH (type "all")
-
-    NodeA->>PubSubA: SearchPublishEvent(ctx, dt, "all", user, "gpu")
-    PubSubA->>PubSubA: publishEvent(PB_SEARCH, user, {search:"gpu"})
-    PubSubA->>PubSubA: GenerateNodeID() → from = DID_A
-    PubSubA->>PubSubA: priv_A.Sign(event body) → sig
-    PubSubA->>PubSubA: Build Event{Type:"search", From:DID_A, Payload:{search:"gpu"}, Sig}
-
-    PubSubA->>GossipSub: topic.Join("search")
-    PubSubA->>GossipSub: topic.Publish(ctx, json(Event))
-
-    GossipSub-->>NodeB: Message propagé (gossip mesh)
-
-    NodeB->>PubSubB: subscribeEvents écoute topic "search#"
-    PubSubB->>PubSubB: json.Unmarshal → Event{From: DID_A}
-
-    PubSubB->>NodeB: GetPeerRecord(ctx, DID_A)
-    Note over NodeB: Résolution Pair A via DB B ou Indexer
-    NodeB-->>PubSubB: Peer A {PublicKey_A, Relation, ...}
-
-    PubSubB->>PubSubB: event.Verify(Peer A) → valide sig_A
-    PubSubB->>PubSubB: handleEventSearch(ctx, evt, PB_SEARCH)
-
-    PubSubB->>StreamB: SendResponse(Peer A, evt)
-    StreamB->>DBB: Search(COMPUTE + STORAGE + ..., filters{creator=self, access=PUBLIC OR partnerships[PeerID_A]}, search="gpu")
-    DBB-->>StreamB: [Resource1, Resource2, ...]
-
-    loop Pour chaque ressource matchée
-        StreamB->>StreamB: write(PeerID_A, addr_A, dt, resource JSON, ProtocolSearchResource)
-        StreamB->>StreamA: NewStream /opencloud/resource/search/1.0
-        StreamB->>StreamA: json.Encode(Event{Type:search, From:DID_B, DataType, Payload:resource})
-    end
-
-    StreamA->>StreamA: readLoop → handleEvent(ProtocolSearchResource, evt)
-    StreamA->>StreamA: retrieveResponse(evt)
-    StreamA->>NATSA: SetNATSPub(SEARCH_EVENT, {DataType, resource JSON})
-    NATSA->>AppA: Résultats de recherche de Pair B

docs/diagrams/10_pubsub_search.puml

@@ -1,54 +1,58 @@
 @startuml
-title PubSub — Recherche gossip globale (type "all") : Pair A cherche, Pair B répond
+title PubSub — Gossip Global search (type "all") : Peer A searching, Peer B answering
 
-participant "App Pair A" as AppA
+participant "App UI A" as UIA
+participant "App Peer A" as AppA
 participant "NATS A" as NATSA
 participant "Node A" as NodeA
+participant "StreamService A" as StreamA
 participant "PubSubService A" as PubSubA
 participant "GossipSub libp2p (mesh)" as GossipSub
 participant "Node B" as NodeB
 participant "PubSubService B" as PubSubB
-participant "DB Pair B (oc-lib)" as DBB
+participant "DB Peer B (oc-lib)" as DBB
 participant "StreamService B" as StreamB
-participant "StreamService A" as StreamA
 
-AppA -> NATSA: Publish(PROPALGATION_EVENT, {PB_SEARCH, type:"all", search:"gpu"})
+UIA -> AppA: websocket subscription, sending {type:"all", search:"search"} in query
+
+AppA -> NATSA: Publish(PROPALGATION_EVENT, {PB_SEARCH, type:"all", search:"search"})
 NATSA -> NodeA: ListenNATS → PB_SEARCH (type "all")
 
-NodeA -> PubSubA: SearchPublishEvent(ctx, dt, "all", user, "gpu")
-PubSubA -> PubSubA: publishEvent(PB_SEARCH, user, {search:"gpu"})
-PubSubA -> PubSubA: GenerateNodeID() → from = DID_A
+NodeA -> PubSubA: SearchPublishEvent(ctx, dt, "all", user, "search")
+PubSubA -> PubSubA: publishEvent(PB_SEARCH, user, {search:"search"})
 PubSubA -> PubSubA: priv_A.Sign(event body) → sig
-PubSubA -> PubSubA: Build Event{Type:"search", From:DID_A, Payload:{search:"gpu"}, Sig}
+PubSubA -> PubSubA: Build Event{Type:"search", From:DID_A, Payload:{search:"search"}, Sig}
 
 PubSubA -> GossipSub: topic.Join("search")
 PubSubA -> GossipSub: topic.Publish(ctx, json(Event))
 
-GossipSub --> NodeB: Message propagé (gossip mesh)
+GossipSub --> NodeB: Propalgate message (gossip mesh)
 
-NodeB -> PubSubB: subscribeEvents écoute topic "search#"
-PubSubB -> PubSubB: json.Unmarshal → Event{From: DID_A}
+NodeB -> PubSubB: subscribeEvents listen  to topic "search#"
+PubSubB -> PubSubB: read → Event{From: DID_A}
 
 PubSubB -> NodeB: GetPeerRecord(ctx, DID_A)
-note over NodeB: Résolution Pair A via DB B ou Indexer
+note over NodeB: Resolve Peer A per DB B or ask to Indexer
 NodeB --> PubSubB: Peer A {PublicKey_A, Relation, ...}
 
-PubSubB -> PubSubB: event.Verify(Peer A) → valide sig_A
+PubSubB -> PubSubB: event.Verify(Peer A) → valid sig_A
 PubSubB -> PubSubB: handleEventSearch(ctx, evt, PB_SEARCH)
 
 PubSubB -> StreamB: SendResponse(Peer A, evt)
-StreamB -> DBB: Search(COMPUTE + STORAGE + ..., filters{creator=self, access=PUBLIC OR partnerships[PeerID_A]}, search="gpu")
+StreamB -> DBB: Search(COMPUTE + STORAGE + ..., filters{creator=self, access=PUBLIC OR partnerships[PeerID_A]}, search="search")
 DBB --> StreamB: [Resource1, Resource2, ...]
 
-loop Pour chaque ressource matchée
+loop For every matching resource, only match our own resource creator_id=self_did
     StreamB -> StreamB: write(PeerID_A, addr_A, dt, resource JSON, ProtocolSearchResource)
     StreamB -> StreamA: NewStream /opencloud/resource/search/1.0
-    StreamB -> StreamA: json.Encode(Event{Type:search, From:DID_B, DataType, Payload:resource})
+    StreamB -> StreamA: stream.Encode(Event{Type:search, From:DID_B, DataType, Payload:resource})
 end
 
 StreamA -> StreamA: readLoop → handleEvent(ProtocolSearchResource, evt)
 StreamA -> StreamA: retrieveResponse(evt)
 StreamA -> NATSA: SetNATSPub(SEARCH_EVENT, {DataType, resource JSON})
-NATSA -> AppA: Résultats de recherche de Pair B
+NATSA -> AppA: Search results from Peer B
+
+AppA -> UIA: emit on websocket
 
 @enduml

docs/diagrams/11_stream_search.mmd

@@ -1,52 +0,0 @@
-sequenceDiagram
-    title Stream — Recherche directe (type "known"/"partner") : Pair A → Pair B
-
-    participant AppA as App Pair A
-    participant NATSA as NATS A
-    participant NodeA as Node A
-    participant PubSubA as PubSubService A
-    participant StreamA as StreamService A
-    participant DBA as DB Pair A (oc-lib)
-    participant NodeB as Node B
-    participant StreamB as StreamService B
-    participant DBB as DB Pair B (oc-lib)
-
-    AppA->>NATSA: Publish(PROPALGATION_EVENT, {PB_SEARCH, type:"partner", search:"gpu"})
-    NATSA->>NodeA: ListenNATS → PB_SEARCH (type "partner")
-    NodeA->>PubSubA: SearchPublishEvent(ctx, dt, "partner", user, "gpu")
-
-    PubSubA->>StreamA: SearchPartnersPublishEvent(dt, user, "gpu")
-    StreamA->>DBA: Search(PEER, PARTNER) + PeerIDS config
-    DBA-->>StreamA: [Peer B, ...]
-
-    loop Pour chaque pair partenaire (Pair B)
-        StreamA->>StreamA: json.Marshal({search:"gpu"}) → payload
-        StreamA->>StreamA: write(PeerID_B, addr_B, dt, user, payload, ProtocolSearchResource)
-        StreamA->>NodeB: TempStream /opencloud/resource/search/1.0
-        StreamA->>NodeB: json.Encode(Event{Type:search, From:DID_A, DataType, Payload:{search:"gpu"}})
-
-        NodeB->>StreamB: HandleResponse(stream) → readLoop
-        StreamB->>StreamB: handleEvent(ProtocolSearchResource, evt)
-        StreamB->>StreamB: handleEventFromPartner(evt, ProtocolSearchResource)
-
-        alt evt.DataType == -1 (toutes ressources)
-            StreamB->>DBA: Search(PEER, evt.From=DID_A)
-            Note over StreamB: Résolution locale ou via GetPeerRecord
-            StreamB->>StreamB: SendResponse(Peer A, evt)
-            StreamB->>DBB: Search(ALL_RESOURCES, filter{creator=B + public OR partner A + search:"gpu"})
-            DBB-->>StreamB: [Resource1, Resource2, ...]
-        else evt.DataType spécifié
-            StreamB->>DBB: Search(DataType, filter{creator=B + access + search:"gpu"})
-            DBB-->>StreamB: [Resource1, ...]
-        end
-
-        loop Pour chaque ressource
-            StreamB->>StreamA: write(PeerID_A, addr_A, dt, resource JSON, ProtocolSearchResource)
-            StreamA->>StreamA: readLoop → handleEvent(ProtocolSearchResource, evt)
-            StreamA->>StreamA: retrieveResponse(evt)
-            StreamA->>NATSA: SetNATSPub(SEARCH_EVENT, {DataType, resource JSON})
-            NATSA->>AppA: Résultat de Pair B
-        end
-    end
-
-    Note over NATSA,DBA: Optionnel: App A persiste<br/>les ressources découvertes dans DB A

docs/diagrams/11_stream_search.puml

@@ -1,6 +1,7 @@
 @startuml
-title Stream — Recherche directe (type "known"/"partner") : Pair A → Pair B
+title Stream — Direct search (type "known"/"partner") : Peer A → Peer B
 
+participant "App UI A" as UIA
 participant "App Pair A" as AppA
 participant "NATS A" as NATSA
 participant "Node A" as NodeA
@@ -11,6 +12,8 @@ participant "Node B" as NodeB
 participant "StreamService B" as StreamB
 participant "DB Pair B (oc-lib)" as DBB
 
+UIA -> AppA: websocket subscription, sending {type:"all", search:"search"} in query
+
 AppA -> NATSA: Publish(PROPALGATION_EVENT, {PB_SEARCH, type:"partner", search:"gpu"})
 NATSA -> NodeA: ListenNATS → PB_SEARCH (type "partner")
 NodeA -> PubSubA: SearchPublishEvent(ctx, dt, "partner", user, "gpu")
@@ -20,10 +23,9 @@ StreamA -> DBA: Search(PEER, PARTNER) + PeerIDS config
 DBA --> StreamA: [Peer B, ...]
 
 loop Pour chaque pair partenaire (Pair B)
-    StreamA -> StreamA: json.Marshal({search:"gpu"}) → payload
     StreamA -> StreamA: write(PeerID_B, addr_B, dt, user, payload, ProtocolSearchResource)
     StreamA -> NodeB: TempStream /opencloud/resource/search/1.0
-    StreamA -> NodeB: json.Encode(Event{Type:search, From:DID_A, DataType, Payload:{search:"gpu"}})
+    StreamA -> NodeB: stream.Encode(Event{Type:search, From:DID_A, DataType, Payload:{search:"gpu"}})
 
     NodeB -> StreamB: HandleResponse(stream) → readLoop
     StreamB -> StreamB: handleEvent(ProtocolSearchResource, evt)
@@ -31,11 +33,11 @@ loop Pour chaque pair partenaire (Pair B)
 
     alt evt.DataType == -1 (toutes ressources)
         StreamB -> DBA: Search(PEER, evt.From=DID_A)
-        note over StreamB: Résolution locale ou via GetPeerRecord
+        note over StreamB: Local Resolving (DB) or GetPeerRecord (Indexer Way)
         StreamB -> StreamB: SendResponse(Peer A, evt)
         StreamB -> DBB: Search(ALL_RESOURCES, filter{creator=B + public OR partner A + search:"gpu"})
         DBB --> StreamB: [Resource1, Resource2, ...]
-    else evt.DataType spécifié
+    else evt.DataType specified
         StreamB -> DBB: Search(DataType, filter{creator=B + access + search:"gpu"})
         DBB --> StreamB: [Resource1, ...]
     end
@@ -45,10 +47,8 @@ loop Pour chaque pair partenaire (Pair B)
         StreamA -> StreamA: readLoop → handleEvent(ProtocolSearchResource, evt)
         StreamA -> StreamA: retrieveResponse(evt)
         StreamA -> NATSA: SetNATSPub(SEARCH_EVENT, {DataType, resource JSON})
-        NATSA -> AppA: Résultat de Pair B
+        NATSA -> AppA: Peer B results
+        AppA -> UIA: emit on websocket
     end
 end
-
-note over NATSA,DBA: Optionnel: App A persiste\nles ressources découvertes dans DB A
-
 @enduml

docs/diagrams/12_partner_heartbeat.mmd

@@ -1,58 +0,0 @@
-sequenceDiagram
-    title Stream — Partner Heartbeat et propagation CRUD Pair A ↔ Pair B
-
-    participant DBA as DB Pair A (oc-lib)
-    participant StreamA as StreamService A
-    participant NodeA as Node A
-    participant NodeB as Node B
-    participant StreamB as StreamService B
-    participant NATSB as NATS B
-    participant DBB as DB Pair B (oc-lib)
-    participant NATSA as NATS A
-
-    Note over StreamA: Démarrage → connectToPartners()
-
-    StreamA->>DBA: Search(PEER, PARTNER) + PeerIDS config
-    DBA-->>StreamA: [Peer B, ...]
-
-    StreamA->>NodeB: Connect (libp2p)
-    StreamA->>NodeB: NewStream /opencloud/resource/heartbeat/partner/1.0
-    StreamA->>NodeB: json.Encode(Heartbeat{Name_A, DID_A, PeerID_A, IndexersBinded_A})
-
-    NodeB->>StreamB: HandlePartnerHeartbeat(stream)
-    StreamB->>StreamB: CheckHeartbeat → bandwidth challenge
-    StreamB->>StreamA: Echo(payload)
-    StreamB->>StreamB: streams[ProtocolHeartbeatPartner][PeerID_A] = {DID_A, Expiry=now+10s}
-
-    StreamA->>StreamA: streams[ProtocolHeartbeatPartner][PeerID_B] = {DID_B, Expiry=now+10s}
-
-    Note over StreamA,StreamB: Stream partner long-lived établi<br/>GC toutes les 8s (StreamService A)<br/>GC toutes les 30s (StreamService B)
-
-    Note over NATSA: Pair A reçoit PROPALGATION_EVENT{PB_DELETE, dt:"storage", payload:res}
-
-    NATSA->>NodeA: ListenNATS → ToPartnerPublishEvent(PB_DELETE, dt, user, payload)
-    NodeA->>StreamA: ToPartnerPublishEvent(ctx, PB_DELETE, dt_storage, user, payload)
-
-    alt dt == PEER (mise à jour relation partenaire)
-        StreamA->>StreamA: json.Unmarshal → peer.Peer B updated
-        alt B.Relation == PARTNER
-            StreamA->>NodeB: ConnectToPartner(B.StreamAddress)
-            Note over StreamA,NodeB: Reconnexion heartbeat si relation upgrade
-        else B.Relation != PARTNER
-            loop Tous les protocoles
-                StreamA->>StreamA: delete(streams[proto][PeerID_B])
-                StreamA->>NodeB: (streams fermés)
-            end
-        end
-    else dt != PEER (ressource ordinaire)
-        StreamA->>DBA: Search(PEER, PARTNER) → [Pair B, ...]
-        loop Pour chaque protocole partner (Create/Update/Delete)
-            StreamA->>NodeB: write(PeerID_B, addr_B, dt, user, payload, ProtocolDeleteResource)
-            Note over NodeB: /opencloud/resource/delete/1.0
-
-            NodeB->>StreamB: HandleResponse → readLoop
-            StreamB->>StreamB: handleEventFromPartner(evt, ProtocolDeleteResource)
-            StreamB->>NATSB: SetNATSPub(REMOVE_RESOURCE, {DataType, resource JSON})
-            NATSB->>DBB: Supprimer ressource dans DB B
-        end
-    end

docs/diagrams/13_planner_flow.mmd

@@ -1,49 +0,0 @@
-sequenceDiagram
-    title Stream — Session Planner : Pair A demande le plan de Pair B
-
-    participant AppA as App Pair A (oc-booking)
-    participant NATSA as NATS A
-    participant NodeA as Node A
-    participant StreamA as StreamService A
-    participant NodeB as Node B
-    participant StreamB as StreamService B
-    participant DBB as DB Pair B (oc-lib)
-    participant NATSB as NATS B
-
-    %% Ouverture session planner
-    AppA->>NATSA: Publish(PROPALGATION_EVENT, {PB_PLANNER, peer_id:PeerID_B, payload:{}})
-    NATSA->>NodeA: ListenNATS → PB_PLANNER
-
-    NodeA->>NodeA: Unmarshal → {peer_id: PeerID_B, payload: {}}
-    NodeA->>StreamA: PublishCommon(nil, user, PeerID_B, ProtocolSendPlanner, {})
-    Note over StreamA: WaitResponse=true, TTL=24h<br/>Stream long-lived vers Pair B
-    StreamA->>NodeB: TempStream /opencloud/resource/planner/1.0
-    StreamA->>NodeB: json.Encode(Event{Type:planner, From:DID_A, Payload:{}})
-
-    NodeB->>StreamB: HandleResponse → readLoop(ProtocolSendPlanner)
-    StreamB->>StreamB: handleEvent(ProtocolSendPlanner, evt)
-    StreamB->>StreamB: sendPlanner(evt)
-
-    alt evt.Payload vide (requête initiale)
-        StreamB->>DBB: planner.GenerateShallow(AdminRequest)
-        DBB-->>StreamB: plan (shallow booking plan de Pair B)
-        StreamB->>StreamA: PublishCommon(nil, user, DID_A, ProtocolSendPlanner, planJSON)
-        StreamA->>NodeA: json.Encode(Event{plan de B})
-        NodeA->>NATSA: (forwardé à AppA via SEARCH_EVENT ou PLANNER event)
-        NATSA->>AppA: Plan de Pair B
-    else evt.Payload non vide (mise à jour planner)
-        StreamB->>StreamB: m["peer_id"] = evt.From (DID_A)
-        StreamB->>NATSB: SetNATSPub(PROPALGATION_EVENT, {PB_PLANNER, peer_id:DID_A, payload:plan})
-        NATSB->>DBB: (oc-booking traite le plan sur NATS B)
-    end
-
-    %% Fermeture session planner
-    AppA->>NATSA: Publish(PROPALGATION_EVENT, {PB_CLOSE_PLANNER, peer_id:PeerID_B})
-    NATSA->>NodeA: ListenNATS → PB_CLOSE_PLANNER
-
-    NodeA->>NodeA: Unmarshal → {peer_id: PeerID_B}
-    NodeA->>StreamA: Mu.Lock()
-    NodeA->>StreamA: Streams[ProtocolSendPlanner][PeerID_B].Stream.Close()
-    NodeA->>StreamA: delete(Streams[ProtocolSendPlanner], PeerID_B)
-    NodeA->>StreamA: Mu.Unlock()
-    Note over StreamA,NodeB: Stream planner fermé — session terminée

docs/diagrams/14_native_offload_gc.mmd

@@ -1,59 +0,0 @@
-sequenceDiagram
-    title Native Indexer — Boucles background (offload, DHT refresh, GC streams)
-
-    participant IndexerA as Indexer A (enregistré)
-    participant IndexerB as Indexer B (enregistré)
-    participant Native as Native Indexer
-    participant DHT as DHT Kademlia
-    participant NodeA as Node A (responsible peer)
-
-    Note over Native: runOffloadLoop — toutes les 30s
-
-    loop Toutes les 30s
-        Native->>Native: len(responsiblePeers) > 0 ?
-        Note over Native: responsiblePeers = peers pour lesquels<br/>le native a fait selfDelegate (aucun indexer dispo)
-        alt Des responsible peers existent (ex: Node A)
-            Native->>Native: reachableLiveIndexers()
-            Note over Native: Filtre liveIndexers par TTL<br/>ping PeerIsAlive pour chaque candidat
-            alt Indexers A et B maintenant joignables
-                Native->>Native: responsiblePeers = {} (libère Node A et autres)
-                Note over Native: Node A se reconnectera<br/>au prochain ConnectToNatives
-            else Toujours aucun indexer
-                Note over Native: Node A reste sous la responsabilité du native
-            end
-        end
-    end
-
-    Note over Native: refreshIndexersFromDHT — toutes les 30s
-
-    loop Toutes les 30s
-        Native->>Native: Collecter tous les knownPeerIDs<br/>= {PeerID_A, PeerID_B, ...}
-        loop Pour chaque PeerID connu
-            Native->>Native: liveIndexers[PeerID] encore frais ?
-            alt Entrée manquante ou expirée
-                Native->>DHT: SearchValue(ctx 5s, "/indexer/"+PeerID)
-                DHT-->>Native: channel de bytes
-                loop Pour chaque résultat DHT
-                    Native->>Native: Unmarshal → liveIndexerEntry
-                    Native->>Native: Garder le meilleur (ExpiresAt le plus récent, valide)
-                end
-                Native->>Native: liveIndexers[PeerID] = best entry
-                Note over Native: "native: refreshed indexer from DHT"
-            end
-        end
-    end
-
-    Note over Native: LongLivedStreamRecordedService GC — toutes les 30s
-
-    loop Toutes les 30s
-        Native->>Native: gc() — lock StreamRecords[Heartbeat]
-        loop Pour chaque StreamRecord (Indexer A, B, ...)
-            Native->>Native: now > rec.Expiry ?<br/>OU timeSince(LastSeen) > 2×TTL restant ?
-            alt Pair périmé (ex: Indexer B disparu)
-                Native->>Native: Supprimer Indexer B de TOUS les maps de protocoles
-                Note over Native: Stream heartbeat fermé<br/>liveIndexers[PeerID_B] expirera naturellement
-            end
-        end
-    end
-
-    Note over IndexerA: Indexer A continue à heartbeater normalement<br/>et reste dans StreamRecords + liveIndexers

docs/diagrams/15_archi_config_nominale.puml

@@ -0,0 +1,49 @@
+@startuml 15_archi_config_nominale
+skinparam componentStyle rectangle
+skinparam backgroundColor white
+skinparam defaultTextAlignment center
+
+title C1 — Topologie nominale\n2 natifs · 2 indexeurs · 2 nœuds
+
+package "Couche 1 — Mesh natif" #E8F4FD {
+  component "Native A\n(hub autoritaire)" as NA #AED6F1
+  component "Native B\n(hub autoritaire)" as NB #AED6F1
+  NA <--> NB : heartbeat /opencloud/heartbeat/1.0 (20s)\n+ gossip PubSub oc-indexer-registry
+}
+
+package "Couche 2 — Indexeurs" #E9F7EF {
+  component "Indexer A\n(DHT server)" as IA #A9DFBF
+  component "Indexer B\n(DHT server)" as IB #A9DFBF
+}
+
+package "Couche 3 — Nœuds" #FEFBD8 {
+  component "Node 1" as N1 #FAF0BE
+  component "Node 2" as N2 #FAF0BE
+}
+
+' Enregistrements (one-shot, 60s)
+IA -[#117A65]--> NA : subscribe signé (60s)\n/opencloud/native/subscribe/1.0
+IA -[#117A65]--> NB : subscribe signé (60s)
+IB -[#117A65]--> NA : subscribe signé (60s)
+IB -[#117A65]--> NB : subscribe signé (60s)
+
+' Heartbeats indexeurs → natifs (long-lived, 20s)
+IA -[#27AE60]..> NA : heartbeat (20s)
+IA -[#27AE60]..> NB : heartbeat (20s)
+IB -[#27AE60]..> NA : heartbeat (20s)
+IB -[#27AE60]..> NB : heartbeat (20s)
+
+' Heartbeats nœuds → indexeurs (long-lived, 20s)
+N1 -[#E67E22]--> IA : heartbeat long-lived (20s)\n/opencloud/heartbeat/1.0
+N1 -[#E67E22]--> IB : heartbeat long-lived (20s)
+N2 -[#E67E22]--> IA : heartbeat long-lived (20s)
+N2 -[#E67E22]--> IB : heartbeat long-lived (20s)
+
+note as Legend
+  Légende :
+  ──► enregistrement one-shot (signé)
+  ···► heartbeat long-lived (20s)
+  ──► heartbeat nœud → indexeur (20s)
+end note
+
+@enduml

docs/diagrams/16_archi_config_seed.puml

@@ -0,0 +1,38 @@
+@startuml 16_archi_config_seed
+skinparam componentStyle rectangle
+skinparam backgroundColor white
+skinparam defaultTextAlignment center
+
+title C2 — Mode seed (sans natif)\nIndexerAddresses seuls · AdmittedAt = zero
+
+package "Couche 2 — Indexeurs seeds" #E9F7EF {
+  component "Indexer A\n(seed, AdmittedAt=0)" as IA #A9DFBF
+  component "Indexer B\n(seed, AdmittedAt=0)" as IB #A9DFBF
+}
+
+package "Couche 3 — Nœuds" #FEFBD8 {
+  component "Node 1" as N1 #FAF0BE
+  component "Node 2" as N2 #FAF0BE
+}
+
+note as NNative #FFDDDD
+  Aucun natif configuré.
+  AdmittedAt = zero → IsStableVoter() = false
+  Phase 2 sans votants : Phase 1 conservée directement.
+  Risque D20 : circularité du trust (seeds se valident mutuellement).
+end note
+
+' Heartbeats nœuds → indexeurs seeds
+N1 -[#E67E22]--> IA : heartbeat long-lived (20s)
+N1 -[#E67E22]--> IB : heartbeat long-lived (20s)
+N2 -[#E67E22]--> IA : heartbeat long-lived (20s)
+N2 -[#E67E22]--> IB : heartbeat long-lived (20s)
+
+note bottom of IA
+  Après 2s : goroutine async
+  fetchNativeFromIndexers → ?
+  Si natif trouvé → ConnectToNatives (upgrade vers C1)
+  Si non → mode indexeur pur (D20 actif)
+end note
+
+@enduml

docs/diagrams/17_startup_consensus_phase1_phase2.puml

@@ -0,0 +1,63 @@
+@startuml 17_startup_consensus_phase1_phase2
+title Démarrage avec natifs — Phase 1 (admission) + Phase 2 (vivacité)
+
+participant "Node / Indexer\n(appelant)" as Caller
+participant "Native A" as NA
+participant "Native B" as NB
+participant "Indexer A" as IA
+participant "Indexer B" as IB
+
+note over Caller: ConnectToNatives()\nNativeIndexerAddresses configuré
+
+== Étape 0 : heartbeat vers le mesh natif ==
+Caller -> NA: SendHeartbeat /opencloud/heartbeat/1.0 (goroutine longue durée)
+Caller -> NB: SendHeartbeat /opencloud/heartbeat/1.0 (goroutine longue durée)
+
+== Étape 1 : fetch pool en parallèle ==
+par Fetch parallèle (timeout 6s)
+    Caller -> NA: GET /opencloud/native/indexers/1.0\nGetIndexersRequest{Count: max, FillRates demandés}
+    NA -> NA: reachableLiveIndexers()\ntri par w(F) = fillRate×(1−fillRate)
+    NA --> Caller: GetIndexersResponse{Indexers:[IA,IB], FillRates:{IA:0.3, IB:0.6}}
+else
+    Caller -> NB: GET /opencloud/native/indexers/1.0
+    NB -> NB: reachableLiveIndexers()
+    NB --> Caller: GetIndexersResponse{Indexers:[IA,IB], FillRates:{IA:0.3, IB:0.6}}
+end par
+
+note over Caller: Fusion + dédup → candidates = [IA, IB]\nisFallback = false
+
+== Étape 2a : Phase 1 — Admission native (clientSideConsensus) ==
+par Consensus parallèle (timeout 3s par natif, 4s total)
+    Caller -> NA: /opencloud/native/consensus/1.0\nConsensusRequest{Candidates:[IA,IB]}
+    NA -> NA: croiser avec liveIndexers
+    NA --> Caller: ConsensusResponse{Trusted:[IA,IB], Suggestions:[]}
+else
+    Caller -> NB: /opencloud/native/consensus/1.0\nConsensusRequest{Candidates:[IA,IB]}
+    NB -> NB: croiser avec liveIndexers
+    NB --> Caller: ConsensusResponse{Trusted:[IA], Suggestions:[IC]}
+end par
+
+note over Caller: IA → 2/2 votes → confirmé ✓\nIB → 1/2 vote → refusé ✗\nadmittedAt = time.Now()
+
+== Étape 2b : Phase 2 — Liveness vote (indexerLivenessVote) ==
+note over Caller: Cherche votants stables dans StaticIndexerMeta\n(AdmittedAt != zero, age >= MinStableAge=2min)
+
+alt Votants stables disponibles
+    par Phase 2 parallèle (timeout 3s)
+        Caller -> IA: /opencloud/indexer/consensus/1.0\nIndexerConsensusRequest{Candidates:[IA]}
+        IA -> IA: vérifier StreamRecords[ProtocolHB][candidate]\nLastSeen ≤ 2×60s && LastScore ≥ 30
+        IA --> Caller: IndexerConsensusResponse{Alive:[IA]}
+    end par
+    note over Caller: IA confirmé vivant par quorum > 0.5
+else Aucun votant stable (premier démarrage)
+    note over Caller: Phase 1 conservée directement\n(aucun votant MinStableAge atteint)
+end
+
+== Étape 3 : remplacement StaticIndexers ==
+Caller -> Caller: replaceStaticIndexers(pool={IA}, admittedAt)\nStaticIndexerMeta[IA].AdmittedAt = time.Now()
+
+== Étape 4 : heartbeat long-lived vers pool ==
+Caller -> IA: SendHeartbeat /opencloud/heartbeat/1.0 (goroutine longue durée)
+note over Caller: Pool actif. NudgeIndexerHeartbeat()
+
+@enduml

docs/diagrams/18_startup_seed_discovers_native.puml

@@ -0,0 +1,51 @@
+@startuml 18_startup_seed_discovers_native
+title C2 → C1 — Seed découvre un natif (upgrade async)
+
+participant "Node / Indexer\\n(seed mode)" as Caller
+participant "Indexer A\\n(seed)" as IA
+participant "Indexer B\\n(seed)" as IB
+participant "Native A\\n(découvert)" as NA
+
+note over Caller: Démarrage sans NativeIndexerAddresses\\nStaticIndexers = [IA, IB] (AdmittedAt=0)
+
+== Phase initiale seed ==
+Caller -> IA: SendHeartbeat /opencloud/heartbeat/1.0 (goroutine longue durée)
+Caller -> IB: SendHeartbeat /opencloud/heartbeat/1.0 (goroutine longue durée)
+
+note over Caller: Pool actif en mode seed.\\nIsStableVoter() = false (AdmittedAt=0)\\nPhase 2 sans votants → Phase 1 conservée.
+
+== Goroutine async après 2s ==
+note over Caller: time.Sleep(2s)\\nfetchNativeFromIndexers()
+
+Caller -> IA: GET /opencloud/indexer/natives/1.0
+IA --> Caller: GetNativesResponse{Natives:[NA]}
+
+note over Caller: Natif découvert : NA\\nAppel ConnectToNatives([NA])
+
+== Upgrade vers mode nominal (ConnectToNatives) ==
+Caller -> NA: SendHeartbeat /opencloud/heartbeat/1.0 (goroutine longue durée)
+
+par Fetch pool depuis natif (timeout 6s)
+    Caller -> NA: GET /opencloud/native/indexers/1.0\\nGetIndexersRequest{Count: max}
+    NA -> NA: reachableLiveIndexers()\\ntri par w(F) = fillRate×(1−fillRate)
+    NA --> Caller: GetIndexersResponse{Indexers:[IA,IB], FillRates:{IA:0.4, IB:0.6}}
+end par
+
+note over Caller: candidates = [IA, IB], isFallback = false
+
+par Consensus Phase 1 (timeout 3s)
+    Caller -> NA: /opencloud/native/consensus/1.0\\nConsensusRequest{Candidates:[IA,IB]}
+    NA -> NA: croiser avec liveIndexers
+    NA --> Caller: ConsensusResponse{Trusted:[IA,IB], Suggestions:[]}
+end par
+
+note over Caller: IA ✓ IB ✓ (1/1 vote)\\nadmittedAt = time.Now()
+
+note over Caller: Aucun votant stable (AdmittedAt vient d'être posé)\\nPhase 2 sautée → Phase 1 conservée directement
+
+== Remplacement pool ==
+Caller -> Caller: replaceStaticIndexers(pool={IA,IB}, admittedAt)\\nStaticIndexerMeta[IA].AdmittedAt = time.Now()\\nStaticIndexerMeta[IB].AdmittedAt = time.Now()
+
+note over Caller: Pool upgradé dans la map partagée StaticIndexers.\\nLa goroutine heartbeat existante (démarrée en mode seed)\\ndétecte les nouveaux membres sur le prochain tick (20s).\\nAucune nouvelle goroutine créée.\\nIsStableVoter() deviendra true après MinStableAge (2min).\\nD20 (circularité seeds) éliminé.
+
+@enduml

docs/diagrams/19_failure_indexer_crash.puml

@@ -0,0 +1,55 @@
+@startuml failure_indexer_crash
+title Indexer Failure → replenish from a Native
+
+participant "Node" as N
+participant "Indexer A (alive)" as IA
+participant "Indexer B (crashed)" as IB
+participant "Native A" as NA
+participant "Native B" as NB
+
+note over N: Active Pool : Indexers = [IA, IB]\\nActive Heartbeat long-lived from IA & IB
+
+== IB Failure ==
+IB ->x N: heartbeat fails (sendHeartbeat err)
+note over N: doTick() dans SendHeartbeat triggers failure\\n→ delete(Indexers[IB])\\n→ delete(IndexerMeta[IB])\\nUnique heartbeat goroutine continue
+
+N -> N: go replenishIndexersFromNative(need=1)
+
+note over N: Reduced Pool to 1 indexers.\\nReplenish triggers with goroutine.
+
+== Replenish from natives ==
+par Fetch pool (timeout 6s)
+    N -> NA: GET /opencloud/native/indexers/1.0\\nGetIndexersRequest{Count: max}
+    NA -> NA: reachableLiveIndexers()\\n(IB absent because of a expired heartbeat)
+    NA --> N: GetIndexersResponse{Indexers:[IA,IC], FillRates:{IA:0.4,IC:0.2}}
+else
+    N -> NB: GET /opencloud/native/indexers/1.0
+    NB --> N: GetIndexersResponse{Indexers:[IA,IC]}
+end par
+
+note over N: Fusion + duplication → candidates = [IA, IC]\\n(IA already in pool → IC new candidate)
+
+par Consensus Phase 1 (timeout 4s)
+    N -> NA: /opencloud/native/consensus/1.0\\nConsensusRequest{Candidates:[IA,IC]}
+    NA --> N: ConsensusResponse{Trusted:[IA,IC]}
+else
+    N -> NB: /opencloud/native/consensus/1.0
+    NB --> N: ConsensusResponse{Trusted:[IA,IC]}
+end par
+
+note over N: IC → 2/2 votes → admit\\nadmittedAt = time.Now()
+
+par Phase 2 — liveness vote (if stable voters )
+    N -> IA: /opencloud/indexer/consensus/1.0\\nIndexerConsensusRequest{Candidates:[IC]}
+    IA -> IA: StreamRecords[ProtocolHB][IC]\\nLastSeen ≤ 120s && LastScore ≥ 30
+    IA --> N: IndexerConsensusResponse{Alive:[IC]}
+end par
+
+note over N: IC confirmed alive → add to pool
+
+N -> N: replaceStaticIndexers(pool={IA,IC})
+N -> IC: SendHeartbeat /opencloud/heartbeat/1.0 (goroutine long-live)
+
+note over N: Pool restaured to 2 indexers.
+
+@enduml

docs/diagrams/20_failure_indexers_native_falback.puml

@@ -0,0 +1,51 @@
+@startuml failure_indexers_native_falback
+title indexers failures → native IsSelfFallback 
+
+participant "Node" as N
+participant "Indexer A (crashed)" as IA
+participant "Indexer B (crashed)" as IB
+participant "Native A" as NA
+participant "Native B" as NB
+
+note over N: Active Pool : Indexers = [IA, IB]
+
+== Successive Failures on IA & IB ==
+IA ->x N: heartbeat failure (sendHeartbeat err)
+IB ->x N: heartbeat failure (sendHeartbeat err)
+
+note over N: doTick() in SendHeartbeat triggers failures\\n→ delete(StaticIndexers[IA]), delete(StaticIndexers[IB])\\n→ delete(StaticIndexerMeta[IA/IB])\\n unique heartbeat goroutine continue.
+
+N -> N: go replenishIndexersFromNative(need=2)
+
+== Replenish attempt — natives switches to self-fallback mode  ==
+par Fetch from natives (timeout 6s)
+    N -> NA: GET /opencloud/native/indexers/1.0
+    NA -> NA: reachableLiveIndexers() → 0 alive indexer\\nFallback : included as himself(IsSelfFallback=true)
+    NA --> N: GetIndexersResponse{Indexers:[NA_addr], IsSelfFallback:true}
+else
+    N -> NB: GET /opencloud/native/indexers/1.0
+    NB --> N: GetIndexersResponse{Indexers:[NB_addr], IsSelfFallback:true}
+end par
+
+note over N: isFallback=true → resolvePool avoids consensus\\nadmittedAt = time.Time{} (zero)\\nStaticIndexers = {NA_addr} (native as fallback)
+
+N -> NA: SendHeartbeat /opencloud/heartbeat/1.0\\n(native as temporary fallback indexers)
+
+note over NA: responsiblePeers[N] registered.\\nrunOffloadLoop look after real indexers.
+
+== Reprise IA → runOffloadLoop native side ==
+IA -> NA: /opencloud/native/subscribe/1.0\\nIndexerRegistration{FillRate: 0}
+note over NA: liveIndexers[IA] updated.\\nrunOffloadLoop triggers a real available indexer\\migrate from N to IA.
+
+== Replenish on next heartbeat tick ==
+N -> NA: GET /opencloud/native/indexers/1.0
+NA --> N: GetIndexersResponse{Indexers:[IA], IsSelfFallback:false}
+
+note over N: isFallback=false → Classic Phase 1 + Phase 2
+
+N -> N: replaceStaticIndexers(pool={IA}, admittedAt)
+N -> IA: SendHeartbeat /opencloud/heartbeat/1.0
+
+note over N: Pool restaured. Native self extracted as indexer.
+
+@enduml

docs/diagrams/21_failure_native_one_down.puml

@@ -0,0 +1,46 @@
+@startuml failure_native_one_down
+title Native failure, with one still alive
+
+participant "Indexer A" as IA
+participant "Indexer B" as IB
+participant "Native A (crashed)" as NA
+participant "Native B (alive)" as NB
+participant "Node" as N
+
+note over IA, NB: Native State : IA, IB heartbeats to NA & NB
+
+== Native A Failure ==
+NA ->x IA: stream reset
+NA ->x IB: stream reset
+NA ->x N: stream reset (heartbeat Node → NA)
+
+== Indexers side : replenishNativesFromPeers ==
+note over IA: SendHeartbeat(NA) détecte reset\\nAfterDelete(NA)\\nStaticNatives = [NB] (still 1)
+
+IA -> IA: replenishNativesFromPeers()\\nphase 1 : fetchNativeFromNatives
+
+IA -> NB: GET /opencloud/native/peers/1.0
+NB --> IA: GetPeersResponse{Peers:[NC]} /' new native if one known '/
+
+alt NC disponible
+    IA -> NC: SendHeartbeat /opencloud/heartbeat/1.0\\nSubscribe /opencloud/native/subscribe/1.0
+    note over IA: StaticNatives = [NB, NC]\\nNative Pool restored.
+else Aucun peer natif
+    IA -> IA: fetchNativeFromIndexers()\\nAsk to any indexers their natives
+    IB --> IA: GetNativesResponse{Natives:[]} /' IB also only got NB '/
+    note over IA: Impossible to find a 2e native.\\nStaticNatives = [NB] (degraded but alive).
+end
+
+== Node side : alive indexers pool ==
+note over N: Node heartbeats to IA & IB.\\nNA Failure does not affect indexers pool.\\nFuture Consensus did not use NB (1/1 vote = quorum OK).
+
+N -> NB: /opencloud/native/consensus/1.0\\nConsensusRequest{Candidates:[IA,IB]}
+NB --> N: ConsensusResponse{Trusted:[IA,IB]}
+note over N: Consensus 1/1 alive natif → admit.\\nAuto downgrade of the consensus floor (alive majority).
+
+== NB side : heartbeat to NA fails ==
+note over NB: EnsureNativePeers / SendHeartbeat to NA\\nfail (sendHeartbeat err)\\n→ delete(StaticNatives[NA])\\nreplenishNativesFromPeers(NA) triggers
+
+note over NB: Mesh natif downgraded to NB alone.\\Downgraded but functionnal.
+
+@enduml

docs/diagrams/22_failure_both_natives.puml

@@ -0,0 +1,60 @@
+@startuml 22_failure_both_natives
+title F4 — Panne des 2 natifs → fallback pool pré-validé
+
+participant "Node" as N
+participant "Indexer A\\n(vivant)" as IA
+participant "Indexer B\\n(vivant)" as IB
+participant "Native A\\n(crashé)" as NA
+participant "Native B\\n(crashé)" as NB
+
+note over N: Pool actif : StaticIndexers = [IA, IB]\\nStaticNatives = [NA, NB]\\nAdmittedAt[IA] et AdmittedAt[IB] posés (stables)
+
+== Panne simultanée NA et NB ==
+NA ->x N: stream reset
+NB ->x N: stream reset
+
+N -> N: AfterDelete(NA) + AfterDelete(NB)\\nStaticNatives = {} (vide)
+
+== replenishNativesFromPeers (sans résultat) ==
+N -> N: fetchNativeFromNatives() → aucun natif vivant
+N -> IA: GET /opencloud/indexer/natives/1.0
+IA --> N: GetNativesResponse{Natives:[NA,NB]}
+note over N: NA et NB connus mais non joignables.\\nAucun nouveau natif trouvé.
+
+== Fallback : pool d'indexeurs conservé ==
+note over N: isFallback = true\\nStaticIndexers conservé tel quel [IA, IB]\\n(dernier pool validé avec AdmittedAt != zero)\\nRisque D19 atténué : quorum natif = 0 → fallback accepté
+
+note over N: Heartbeats IA et IB continuent normalement.\\nPool d'indexeurs opérationnel sans natifs.
+
+N -> IA: SendHeartbeat /opencloud/heartbeat/1.0 (continue)
+N -> IB: SendHeartbeat /opencloud/heartbeat/1.0 (continue)
+
+== retryLostNative (30s ticker) ==
+loop toutes les 30s
+    N -> N: retryLostNative()\\ntente reconnexion NA et NB
+    N -> NA: dial (échec)
+    N -> NB: dial (échec)
+    note over N: Retry sans résultat.\\nPool indexeurs maintenu en fallback.
+end
+
+== Reprise natifs ==
+NA -> NA: redémarrage
+NB -> NB: redémarrage
+
+N -> NA: dial (succès)
+N -> NA: SendHeartbeat /opencloud/heartbeat/1.0
+N -> NB: SendHeartbeat /opencloud/heartbeat/1.0
+note over N: StaticNatives = [NA, NB] restauré\\nisFallback = false
+
+== Re-consensus pool indexeurs (optionnel) ==
+par Consensus Phase 1
+    N -> NA: /opencloud/native/consensus/1.0\\nConsensusRequest{Candidates:[IA,IB]}
+    NA --> N: ConsensusResponse{Trusted:[IA,IB]}
+else
+    N -> NB: /opencloud/native/consensus/1.0
+    NB --> N: ConsensusResponse{Trusted:[IA,IB]}
+end par
+
+note over N: Pool [IA,IB] reconfirmé.\\nisFallback = false. AdmittedAt[IA,IB] rafraîchi.
+
+@enduml

docs/diagrams/23_failure_native_plus_indexer.puml

@@ -0,0 +1,63 @@
+@startuml 23_failure_native_plus_indexer
+title F5 — Panne combinée : 1 natif + 1 indexeur
+
+participant "Node" as N
+participant "Indexer A\\n(vivant)" as IA
+participant "Indexer B\\n(crashé)" as IB
+participant "Native A\\n(vivant)" as NA
+participant "Native B\\n(crashé)" as NB
+
+note over N: Pool nominal : StaticIndexers=[IA,IB], StaticNatives=[NA,NB]
+
+== Pannes simultanées NB + IB ==
+NB ->x N: stream reset
+IB ->x N: stream reset
+
+N -> N: AfterDelete(NB) — StaticNatives = [NA]
+N -> N: AfterDelete(IB) — StaticIndexers = [IA]
+
+== Replenish natif (1 vivant) ==
+N -> N: replenishNativesFromPeers()
+N -> NA: GET /opencloud/native/peers/1.0
+NA --> N: GetPeersResponse{Peers:[]} /' NB seul pair, disparu '/
+note over N: Aucun natif alternatif.\\nStaticNatives = [NA] — dégradé.
+
+== Replenish indexeur depuis NA ==
+par Fetch pool (timeout 6s)
+    N -> NA: GET /opencloud/native/indexers/1.0
+    NA -> NA: reachableLiveIndexers()\\n(IB absent — heartbeat expiré)
+    NA --> N: GetIndexersResponse{Indexers:[IA,IC], FillRates:{IA:0.5,IC:0.3}}
+end par
+
+note over N: candidates = [IA, IC]
+
+par Consensus Phase 1 — 1 seul natif vivant (timeout 3s)
+    N -> NA: /opencloud/native/consensus/1.0\\nConsensusRequest{Candidates:[IA,IC]}
+    NA --> N: ConsensusResponse{Trusted:[IA,IC]}
+end par
+
+note over N: IC → 1/1 vote → admis (quorum sur vivants)\\nadmittedAt = time.Now()
+
+par Phase 2 liveness vote
+    N -> IA: /opencloud/indexer/consensus/1.0\\nIndexerConsensusRequest{Candidates:[IC]}
+    IA -> IA: StreamRecords[ProtocolHB][IC]\\nLastSeen ≤ 120s && LastScore ≥ 30
+    IA --> N: IndexerConsensusResponse{Alive:[IC]}
+end par
+
+N -> N: replaceStaticIndexers(pool={IA,IC})
+N -> IC: SendHeartbeat /opencloud/heartbeat/1.0
+
+note over N: Pool restauré à [IA,IC].\\nMode dégradé : 1 natif seulement.\\nretryLostNative(NB) actif (30s ticker).
+
+== retryLostNative pour NB ==
+loop toutes les 30s
+    N -> NB: dial (échec)
+end
+
+NB -> NB: redémarrage
+NB -> NA: heartbeat (mesh natif reconstruit)
+N -> NB: dial (succès)
+N -> NB: SendHeartbeat /opencloud/heartbeat/1.0
+note over N: StaticNatives = [NA,NB] restauré.\\nMode nominal retrouvé.
+
+@enduml

docs/diagrams/24_failure_retry_lost_native.puml

@@ -0,0 +1,45 @@
+@startuml 24_failure_retry_lost_native
+title F6 — retryLostNative : reconnexion natif après panne réseau
+
+participant "Node / Indexer" as Caller
+participant "Native A\\n(vivant)" as NA
+participant "Native B\\n(réseau instable)" as NB
+
+note over Caller: StaticNatives = [NA, NB]\\nHeartbeats actifs vers NA et NB
+
+== Panne réseau transitoire vers NB ==
+NB ->x Caller: stream reset (timeout réseau)
+
+Caller -> Caller: AfterDelete(NB)\\nStaticNatives = [NA]\\nlostNatives.Store(NB.addr)
+
+== replenishNativesFromPeers — phase 1 ==
+Caller -> NA: GET /opencloud/native/peers/1.0
+NA --> Caller: GetPeersResponse{Peers:[NB]}
+
+note over Caller: NB connu de NA, tentative de reconnexion directe
+
+Caller -> NB: dial (échec — réseau toujours coupé)
+note over Caller: Connexion impossible.\\nPassage en retryLostNative()
+
+== retryLostNative : ticker 30s ==
+loop toutes les 30s tant que NB absent
+    Caller -> Caller: retryLostNative()\\nParcourt lostNatives
+    Caller -> NB: StartNativeRegistration (dial + heartbeat + subscribe)
+    NB --> Caller: dial échoue
+    note over Caller: Retry loggé. Prochain essai dans 30s.
+end
+
+== Réseau rétabli ==
+note over NB: Réseau rétabli\\nNB de nouveau joignable
+
+Caller -> NB: StartNativeRegistration\\ndial (succès)
+Caller -> NB: SendHeartbeat /opencloud/heartbeat/1.0 (goroutine longue durée)
+Caller -> NB: /opencloud/native/subscribe/1.0\\nIndexerRegistration{FillRate: fillRateFn()}
+
+NB --> Caller: subscribe ack
+
+Caller -> Caller: lostNatives.Delete(NB.addr)\\nStaticNatives = [NA, NB] restauré
+
+note over Caller: Mode nominal retrouvé.\\nnativeHeartbeatOnce non utilisé (goroutine déjà active pour NA).\\nNouvelle goroutine SendHeartbeat pour NB uniquement.
+
+@enduml

docs/diagrams/25_failure_node_gc.puml

@@ -0,0 +1,42 @@
+@startuml 25_failure_node_gc
+title F7 — Crash nœud → GC indexeur + AfterDelete
+
+participant "Node\\n(crashé)" as N
+participant "Indexer A" as IA
+participant "Indexer B" as IB
+participant "Native A" as NA
+
+note over N, NA: État nominal : N heartbeatait vers IA et IB
+
+== Crash Node ==
+N ->x IA: stream reset (heartbeat coupé)
+N ->x IB: stream reset (heartbeat coupé)
+
+== GC côté Indexer A ==
+note over IA: HandleHeartbeat : stream reset détecté\\nStreamRecords[ProtocolHB][N].LastSeen figé
+
+loop ticker GC (30s) — StartGC(30*time.Second)
+    IA -> IA: gc()\\nnow.After(Expiry) où Expiry = lastHBTime + 2min\\n→ si 2min sans heartbeat → éviction
+    IA -> IA: delete(StreamRecords[ProtocolHB][N])\\nAfterDelete(N, name, did) appelé hors lock
+    note over IA: N retiré du registre vivant.\\nFillRate recalculé (n-1 / maxNodes).
+end
+
+== Impact sur le scoring / fill rate ==
+note over IA: FillRate diminue\\nProchain subscribe vers NA inclura FillRate mis à jour
+
+IA -> NA: /opencloud/native/subscribe/1.0\\nIndexerRegistration{FillRate: 0.3} /' était 0.5 '/
+
+NA -> NA: liveIndexerEntry[IA].FillRate = 0.3\\nPriorité de routage recalculée : w(0.3) = 0.21
+
+== Impact sur la Phase 2 (indexerLivenessVote) ==
+note over IA: Si un autre nœud demande consensus,\\nN n'est plus dans StreamRecords.\\nN absent de la réponse Alive[].
+
+note over IB: Même GC effectué côté IB.\\nN retiré de StreamRecords[ProtocolHB].
+
+== Reconnexion éventuelle du nœud ==
+N -> N: redémarrage
+N -> IA: SendHeartbeat /opencloud/heartbeat/1.0\\nHeartbeat{Score: X, IndexersBinded: 2}
+IA -> IA: HandleHeartbeat → nouveau UptimeTracker(FirstSeen=now)\\nStreamRecords[ProtocolHB][N] recréé
+note over IA: N de retour avec FirstSeen frais.\\ndynamicMinScore élevé tant que age < 24h.
+
+@enduml

docs/diagrams/README.md

@@ -1,39 +1,71 @@
-# OC-Discovery — Diagrammes de séquence
+# OC-Discovery — Diagrammes d'architecture et de séquence
 
-Tous les fichiers `.mmd` sont au format [Mermaid](https://mermaid.js.org/).
-Rendu possible via VS Code (extension Mermaid Preview), IntelliJ, ou [mermaid.live](https://mermaid.live).
+Tous les fichiers sont au format [PlantUML](https://plantuml.com/).
+Rendu possible via VS Code (extension PlantUML), IntelliJ, ou [plantuml.com/plantuml](https://www.plantuml.com/plantuml/uml/).
 
-## Vue d'ensemble des diagrammes
+## Diagrammes de séquence (flux internes)
 
 | Fichier | Description |
 |---------|-------------|
-| `01_node_init.mmd` | Initialisation complète d'un Node (libp2p host, GossipSub, indexers, StreamService, PubSubService, NATS) |
-| `02_node_claim.mmd` | Enregistrement du nœud auprès des indexeurs (`claimInfo` + `publishPeerRecord`) |
-| `03_indexer_heartbeat.mmd` | Protocole heartbeat avec calcul du score qualité (bande passante, uptime, diversité) |
-| `04_indexer_publish.mmd` | Publication d'un `PeerRecord` vers l'indexeur → DHT |
-| `05_indexer_get.mmd` | Résolution d'un pair via l'indexeur (`GetPeerRecord` + `handleNodeGet` + DHT) |
-| `06_native_registration.mmd` | Enregistrement d'un indexeur auprès d'un Native Indexer + gossip PubSub |
-| `07_native_get_consensus.mmd` | `ConnectToNatives` : pool d'indexeurs + protocole de consensus (vote majoritaire) |
-| `08_nats_create_resource.mmd` | Handler NATS `CREATE_RESOURCE` : connexion/déconnexion d'un partner |
-| `09_nats_propagation.mmd` | Handler NATS `PROPALGATION_EVENT` : delete, considers, planner, search |
-| `10_pubsub_search.mmd` | Recherche gossip globale (type `"all"`) via GossipSub |
-| `11_stream_search.mmd` | Recherche directe par stream (type `"known"` ou `"partner"`) |
-| `12_partner_heartbeat.mmd` | Heartbeat partner + propagation CRUD vers les partenaires |
-| `13_planner_flow.mmd` | Session planner (ouverture, échange, fermeture) |
-| `14_native_offload_gc.mmd` | Boucles background du Native Indexer (offload, DHT refresh, GC) |
+| `01_node_init.puml` | Initialisation complète d'un Node (libp2p host, GossipSub, indexers, StreamService, PubSubService, NATS) |
+| `02_node_claim.puml` | Enregistrement du nœud auprès des indexeurs (`claimInfo` + `publishPeerRecord`) |
+| `03_indexer_heartbeat.puml` | Protocole heartbeat avec score 5 composants (U/B/D/L/F), UptimeTracker, dynamicMinScore |
+| `04_indexer_publish.puml` | Publication d'un `PeerRecord` vers l'indexeur → DHT |
+| `05_indexer_get.puml` | Résolution d'un pair via l'indexeur (`GetPeerRecord` + `handleNodeGet` + DHT) |
+| `06_native_registration.puml` | Enregistrement d'un indexeur auprès du Native (FillRate, signature, TTL 90s, unsubscribe) |
+| `07_native_get_consensus.puml` | `ConnectToNatives` : fetch pool + Phase 1 (clientSideConsensus) + Phase 2 (indexerLivenessVote) |
+| `08_nats_create_resource.puml` | Handler NATS `CREATE_RESOURCE` : connexion/déconnexion d'un partner |
+| `09_nats_propagation.puml` | Handler NATS `PROPALGATION_EVENT` : delete, considers, planner, search |
+| `10_pubsub_search.puml` | Recherche gossip globale (type `"all"`) via GossipSub |
+| `11_stream_search.puml` | Recherche directe par stream (type `"known"` ou `"partner"`) |
+| `12_partner_heartbeat.puml` | Heartbeat partner + propagation CRUD vers les partenaires |
+| `13_planner_flow.puml` | Session planner (ouverture, échange, fermeture) |
+| `14_native_offload_gc.puml` | Boucles background du Native Indexer (offload, DHT refresh, GC) |
 
-## Protocoles libp2p utilisés
+## Diagrammes de topologie et flux de panne
+
+### Configurations réseau
+
+| Fichier | Description |
+|---------|-------------|
+| `15_archi_config_nominale.puml` | C1 — Topologie nominale : 2 natifs · 2 indexeurs · 2 nœuds, tous flux |
+| `16_archi_config_seed.puml` | C2 — Mode seed sans natif : indexeurs à AdmittedAt=0, risque D20 actif |
+
+### Flux de démarrage
+
+| Fichier | Description |
+|---------|-------------|
+| `17_startup_consensus_phase1_phase2.puml` | Démarrage nominal : Phase 1 (admission native) + Phase 2 (liveness vote) |
+| `18_startup_seed_discovers_native.puml` | Upgrade seed → nominal : goroutine async découvre un natif via l'indexeur |
+
+### Flux de panne
+
+| Fichier | Code | Description |
+|---------|------|-------------|
+| `19_failure_indexer_crash.puml` | F1 | Panne 1 indexeur → replenish depuis natif → IC admis |
+| `20_failure_both_indexers_selfdelegate.puml` | F2 | Panne 2 indexeurs → natif `IsSelfFallback=true`, runOffloadLoop |
+| `21_failure_native_one_down.puml` | F3 | Panne 1 natif → quorum 1/1 suffisant, mode dégradé |
+| `22_failure_both_natives.puml` | F4 | Panne 2 natifs → fallback pool pré-validé, retryLostNative |
+| `23_failure_native_plus_indexer.puml` | F5 | Panne combinée : 1 natif + 1 indexeur → double replenish |
+| `24_failure_retry_lost_native.puml` | F6 | Panne réseau transitoire → retryLostNative (30s ticker) |
+| `25_failure_node_gc.puml` | F7 | Crash nœud → GC indexeur (120s), AfterDelete, fill rate recalculé |
+
+## Protocoles libp2p utilisés (référence complète)
 
 | Protocole | Description |
 |-----------|-------------|
-| `/opencloud/heartbeat/1.0` | Heartbeat node → indexeur (long-lived) |
-| `/opencloud/heartbeat/indexer/1.0` | Heartbeat indexeur → native (long-lived) |
+| `/opencloud/heartbeat/1.0` | Heartbeat universel : node→indexeur, indexeur→native, native→native (long-lived) |
+| `/opencloud/probe/1.0` | Sonde de bande passante (echo, mesure latence + débit) |
 | `/opencloud/resource/heartbeat/partner/1.0` | Heartbeat node ↔ partner (long-lived) |
 | `/opencloud/record/publish/1.0` | Publication `PeerRecord` vers indexeur |
 | `/opencloud/record/get/1.0` | Requête `GetPeerRecord` vers indexeur |
-| `/opencloud/native/subscribe/1.0` | Enregistrement indexeur auprès du native |
-| `/opencloud/native/indexers/1.0` | Requête de pool d'indexeurs au native |
-| `/opencloud/native/consensus/1.0` | Validation de pool d'indexeurs (consensus) |
+| `/opencloud/native/subscribe/1.0` | Enregistrement indexeur auprès du native (+ FillRate) |
+| `/opencloud/native/unsubscribe/1.0` | Désenregistrement explicite indexeur → native |
+| `/opencloud/native/indexers/1.0` | Requête de pool d'indexeurs au native (tri par w(F)=F×(1-F)) |
+| `/opencloud/native/consensus/1.0` | Phase 1 : validation de pool d'indexeurs (vote majoritaire natifs) |
+| `/opencloud/native/peers/1.0` | Demande de pairs natifs connus (replenish mesh natif) |
+| `/opencloud/indexer/natives/1.0` | Demande d'adresses de natifs connus par un indexeur |
+| `/opencloud/indexer/consensus/1.0` | Phase 2 : liveness vote (LastSeen ≤ 120s && LastScore ≥ 30) |
 | `/opencloud/resource/search/1.0` | Recherche de ressources entre peers |
 | `/opencloud/resource/create/1.0` | Propagation création ressource vers partner |
 | `/opencloud/resource/update/1.0` | Propagation mise à jour ressource vers partner |

main.go

@@ -2,7 +2,6 @@ package main
 
 import (
 	"context"
-	"fmt"
 	"log"
 	"oc-discovery/conf"
 	"oc-discovery/daemons/node"
@@ -43,7 +42,6 @@ func main() {
 		syscall.SIGTERM,
 	)
 	defer stop()
-	fmt.Println(conf.GetConfig().NodeMode)
 	isNode := strings.Contains(conf.GetConfig().NodeMode, "node")
 	isIndexer := strings.Contains(conf.GetConfig().NodeMode, "indexer")
 	isNativeIndexer := strings.Contains(conf.GetConfig().NodeMode, "native-indexer")