@xdarkicex/openclaw-memory-libravdb 1.3.13 → 1.3.17

package/src/context-engine.ts

@@ -5,6 +5,7 @@ import {
   selectRecentTail,
 } from "./continuity.js";
 import {
+  detectRetrievalFailure,
   expandSection7HopCandidates,
   mergeSection7VariantCandidates,
   rankSection7VariantCandidates,
@@ -14,6 +15,7 @@ import { countTokens, estimateTokens, fitPromptBudget } from "./tokens.js";
 import type { RpcGetter } from "./plugin-runtime.js";
 import type {
   ContextAssembleArgs,
+  ContextAssembleResult,
   ContextBootstrapArgs,
   ContextCompactArgs,
   ContextIngestArgs,
@@ -26,6 +28,13 @@ import type {
 const AUTHORED_HARD_COLLECTION = "authored:hard";
 const AUTHORED_SOFT_COLLECTION = "authored:soft";
 const AUTHORED_VARIANT_COLLECTION = "authored:variant";
+const ELEVATED_USER_COLLECTION_PREFIX = "elevated:user:";
+const ELEVATED_SESSION_COLLECTION_PREFIX = "elevated:session:";
+const SESSION_RECALL_COLLECTION_PREFIX = "session_recall:";
+const SESSION_RAW_COLLECTION_PREFIX = "session_raw:";
+const SESSION_SUMMARY_COLLECTION_PREFIX = "session_summary:";
+const SESSION_EDGE_COLLECTION_PREFIX = "session_edge:";
+const SESSION_STATE_COLLECTION_PREFIX = "session_state:";
 
 export function buildContextEngineFactory(
   getRpc: RpcGetter,
@@ -35,6 +44,16 @@ export function buildContextEngineFactory(
   let authoredHardCache: SearchResult[] | null = null;
   let authoredSoftCache: SearchResult[] | null = null;
   let authoredVariantCache: SearchResult[] | null = null;
+  const authoredVariantRecallCache = new Map<string, SearchResult[]>();
+
+  // Session-scoped elevated-guidance cache keyed by sessionId + generation + userId + queryText
+  const elevatedRecallCache = new Map<string, SearchResult[]>();
+  const elevatedRecallGeneration = new Map<string, number>();
+
+  function clearElevatedCacheForSession(sessionId: string) {
+    const nextGeneration = (elevatedRecallGeneration.get(sessionId) ?? 0) + 1;
+    elevatedRecallGeneration.set(sessionId, nextGeneration);
+  }
 
   return {
     ownsCompaction: true,
@@ -43,6 +62,11 @@ export function buildContextEngineFactory(
       await rpc.call("ensure_collections", {
         collections: [
           `session:${sessionId}`,
+          sessionRawCollection(sessionId),
+          sessionSummaryCollection(sessionId),
+          sessionEdgeCollection(sessionId),
+          sessionStateCollection(sessionId),
+          ...(useSessionRecallProjection(cfg) ? [sessionRecallCollection(sessionId)] : []),
           `turns:${userId}`,
           `user:${userId}`,
           "global",
@@ -59,6 +83,10 @@ export function buildContextEngineFactory(
       authoredHardCache = authoredHard;
       authoredSoftCache = authoredSoft;
       authoredVariantCache = authoredVariantRecords;
+      authoredVariantRecallCache.clear();
+      if (useSessionRecallProjection(cfg)) {
+        await rebuildSessionRecallProjection(rpc, cfg, sessionId);
+      }
       validateSection7StartupHardReserve(cfg, authoredHard);
       return { ok: true };
     },
@@ -69,12 +97,34 @@ export function buildContextEngineFactory(
 
       const rpc = await getRpc();
       const ts = Date.now();
-      void rpc.call("insert_text", {
-        collection: `session:${sessionId}`,
-        id: `${sessionId}:${ts}`,
+      const sessionMeta = {
+        role: message.role,
+        ts,
+        userId,
+        sessionId,
+        type: "turn",
+        provenance_class: "session_turn",
+        stability_weight: stabilityWeightForMessage(message.role),
+      };
+      // Elevated cache is session-scoped, so invalidate immediately on every ingest
+      clearElevatedCacheForSession(sessionId);
+      const rawSessionId = `${sessionId}:${ts}`;
+      const rawSessionInsert = rpc.call("insert_session_turn", {
+        sessionId,
+        id: rawSessionId,
         text: message.content,
-        metadata: { role: message.role, ts, userId, sessionId, type: "turn" },
-      }).catch(console.error);
+        metadata: sessionMeta,
+      });
+      if (useSessionRecallProjection(cfg)) {
+        try {
+          await rawSessionInsert;
+          await rebuildSessionRecallProjection(rpc, cfg, sessionId);
+        } catch (error) {
+          console.error(error);
+        }
+      } else {
+        void rawSessionInsert.catch(console.error);
+      }
 
       if (message.role === "user") {
         try {
@@ -83,7 +133,10 @@ export function buildContextEngineFactory(
             collection: `turns:${userId}`,
             id: `${userId}:${ts}`,
             text: message.content,
-            metadata: { role: message.role, ts, userId, sessionId, type: "turn" },
+            metadata: {
+              ...sessionMeta,
+              provenance_class: "turn_index",
+            },
           });
 
           const gating = await rpc.call<GatingResult>("gating_scalar", {
@@ -102,6 +155,8 @@ export function buildContextEngineFactory(
                 sessionId,
                 type: "turn",
                 userId,
+                provenance_class: "durable_user_memory",
+                stability_weight: Math.max(stabilityWeightForMessage(message.role), gating.g),
                 gating_score: gating.g,
                 gating_t: gating.t,
                 gating_h: gating.h,
@@ -123,181 +178,427 @@ export function buildContextEngineFactory(
       return { ingested: true };
     },
     async assemble({ sessionId, userId, messages, tokenBudget }: ContextAssembleArgs) {
+      const PROFILE = process.env.OPENCLAW_PROFILE_ASSEMBLE === "1";
+
       const queryText = messages.at(-1)?.content ?? "";
       if (!queryText) {
         return {
           messages,
           estimatedTokens: countTokens(messages),
           systemPromptAddition: "",
-        };
+        } satisfies ContextAssembleResult;
       }
 
       const excluded = recentIds(messages, 4);
-      const cached = recallCache.get({ userId, queryText });
+      const cached = recallCache.take({ userId, queryText });
 
-      try {
-        const rpc = await getRpc();
-        const [authoredHard, authoredSoft, authoredVariantRecords] = await loadAuthoredCollections(rpc, {
+      const rpc = await getRpc();
+
+      // Use cached authored collections directly if available (bootstrap-loaded and sorted)
+      // Only load as fallback if caches are unexpectedly null
+      let authoredHard = authoredHardCache;
+      let authoredSoft = authoredSoftCache;
+      let authoredVariantRecords = authoredVariantCache;
+      if (!authoredHard || !authoredSoft || !authoredVariantRecords) {
+        const [loadedHard, loadedSoft, loadedVariant] = await loadAuthoredCollections(rpc, {
           hard: authoredHardCache,
           soft: authoredSoftCache,
           variant: authoredVariantCache,
         });
-        authoredHardCache = authoredHard;
-        authoredSoftCache = authoredSoft;
-        authoredVariantCache = authoredVariantRecords;
-
-        const memoryBudget = tokenBudget * (cfg.tokenBudgetFraction ?? 0.25);
-        const hardItems = authoredHard;
-        const hardUsed = tokenCostSum(hardItems);
-        const sessionRecords = await rpc.call<{ results: SearchResult[] }>("list_by_meta", {
-          collection: `session:${sessionId}`,
-          key: "sessionId",
-          value: sessionId,
-        });
-        const rawSessionTurns = sortChronological(
-          sessionRecords.results.filter((item) => item.metadata.type !== "summary"),
-        );
-        const minTurns = cfg.continuityMinTurns ?? DEFAULT_CONTINUITY_MIN_TURNS;
-        const tailTarget = cfg.continuityTailBudgetTokens ?? DEFAULT_CONTINUITY_TAIL_BUDGET_TOKENS;
-        const baseTail = selectRecentTail(rawSessionTurns, {
-          minTurns,
-          tailBudgetTokens: 0,
-          tokenCost,
-          sameBundle: isContinuityBundleCoupled,
-        });
-        const baseTailUsed = baseTail.baseTokens;
-        const configuredHardFraction = clampFraction(cfg.authoredHardBudgetFraction);
-        const hardBudget = configuredHardFraction > 0 ? memoryBudget * configuredHardFraction : hardUsed;
-        const degradedReasons: string[] = [];
-        if (hardUsed > hardBudget + 1e-9) {
-          degradedReasons.push("hard authored invariants exceed configured hard budget reserve");
-        }
-        if (hardUsed + baseTailUsed > memoryBudget + 1e-9) {
-          degradedReasons.push("hard authored invariants plus mandatory recent-tail base exceed available memory budget");
-        }
-        if (degradedReasons.length > 0) {
-          const degradedTail = markRecentTail(baseTail.base, baseTail.base.length);
-          const selected = [...hardItems, ...degradedTail];
-          const selectedMessages = selected.map((item) => ({
-            role: "system",
-            content: buildInjectedMemoryMessageContent(item),
-          }));
-          return {
-            messages: [...selectedMessages, ...messages],
-            estimatedTokens: countTokens(selectedMessages) + countTokens(messages),
-            systemPromptAddition: buildDegradedMemoryHeader(degradedReasons, selected),
-          };
-        }
-        const authoredSoftTarget = Math.max(0, memoryBudget * (cfg.authoredSoftBudgetFraction ?? 0.3));
-        const softBudget = Math.max(0, Math.min(authoredSoftTarget, memoryBudget - hardUsed - baseTailUsed));
-        const softItems = fitPromptBudget(authoredSoft, softBudget);
-        const remainingAfterHardSoft = Math.max(0, memoryBudget - hardUsed - tokenCostSum(softItems));
-        const effectiveTailBudget = Math.min(
-          Math.max(tailTarget, baseTailUsed),
-          remainingAfterHardSoft,
-        );
-        const recentTailSelection = selectRecentTail(rawSessionTurns, {
-          minTurns,
-          tailBudgetTokens: effectiveTailBudget,
-          tokenCost,
-          sameBundle: isContinuityBundleCoupled,
-        });
-        const recentTail = markRecentTail(
-          recentTailSelection.recent,
-          recentTailSelection.base.length,
-        );
-        const tailBaseItems = recentTail.slice(-recentTailSelection.base.length);
-        const tailExtensionItems = recentTail.slice(0, Math.max(0, recentTail.length - recentTailSelection.base.length));
-        const retrievalBudget = Math.max(0, memoryBudget - hardUsed - tokenCostSum(softItems) - tokenCostSum(recentTail));
-        const recentTailIDs = recentTail.map((item) => item.id);
-
-        const coarseTopK = Math.max(cfg.section7CoarseTopK ?? Math.max((cfg.topK ?? 8) * 2, 8), 1);
-        const secondPassTopK = Math.max(cfg.section7SecondPassTopK ?? (cfg.topK ?? 8), 1);
-        const [sessionHits, durableHits] = await Promise.all([
-          rpc.call<{ results: SearchResult[] }>("search_text", {
-            collection: `session:${sessionId}`,
-            text: queryText,
-            k: coarseTopK,
-            excludeIds: [...excluded, ...recentTailIDs],
-          }),
-          cached
-            ? Promise.resolve({ results: cached.durableVariantHits })
-            : rpc.call<{ results: SearchResult[] }>("search_text_collections", {
-                collections: [`user:${userId}`, "global", AUTHORED_VARIANT_COLLECTION],
-                text: queryText,
-                k: coarseTopK,
-                excludeByCollection: {},
-              }),
-        ]);
-
-        if (!cached) {
-          recallCache.put({
-            userId,
-            queryText,
-            durableVariantHits: durableHits.results,
-          });
-        }
+        authoredHard = loadedHard;
+        authoredSoft = loadedSoft;
+        authoredVariantRecords = loadedVariant;
+        authoredHardCache = loadedHard;
+        authoredSoftCache = loadedSoft;
+        authoredVariantCache = loadedVariant;
+      }
 
-        const ranked = rankSection7VariantCandidates(
-          [
-            ...annotateCollection(sessionHits.results, `session:${sessionId}`),
-            ...durableHits.results,
-          ],
-          {
-            queryText,
-            k1: coarseTopK,
-            k2: secondPassTopK,
-            theta1: cfg.section7Theta1,
-            kappa: cfg.section7Kappa,
-            authorityRecencyLambda: cfg.section7AuthorityRecencyLambda,
-            authorityRecencyWeight: cfg.section7AuthorityRecencyWeight,
-            authorityFrequencyWeight: cfg.section7AuthorityFrequencyWeight,
-            authorityAuthoredWeight: cfg.section7AuthorityAuthoredWeight,
-            sessionId,
-            userId,
-          },
-        );
-        const hopExpanded = expandSection7HopCandidates(
-          ranked,
-          annotateCollection(authoredVariantRecords, AUTHORED_VARIANT_COLLECTION),
-          {
-            etaHop: cfg.section7HopEta,
-            thetaHop: cfg.section7HopThreshold,
-          },
-        );
+      // Profiler: null when disabled (zero overhead), object when enabled
+      const profiler = PROFILE
+        ? (() => {
+            const marks: Array<[string, bigint]> = [];
+            return {
+              mark(label: string) {
+                marks.push([label, process.hrtime.bigint()]);
+              },
+              lines() {
+                const lines: string[] = [];
+                for (let i = 0; i < marks.length - 1; i++) {
+                  const [name, start] = marks[i];
+                  const [, end] = marks[i + 1];
+                  const ms = Number(end - start) / 1_000_000;
+                  lines.push(`assemble profile: ${name}=${ms.toFixed(2)}ms`);
+                }
+                return lines;
+              },
+              emit() {
+                for (const line of this.lines()) {
+                  console.log(line);
+                }
+              },
+            };
+          })()
+        : null;
 
-        const variantItems = fitPromptBudget(
-          mergeSection7VariantCandidates(ranked, hopExpanded),
-          retrievalBudget,
-        );
-        const selected = [
-          ...hardItems,
-          ...tailBaseItems,
-          ...softItems,
-          ...tailExtensionItems,
-          ...variantItems,
-        ];
-        void rpc.call("bump_access_counts", {
-          updates: groupAccessCountUpdates(variantItems),
-        }).catch(() => {});
+      try {
+        const result = await this.assembleCore({
+          rpc,
+          cfg,
+          recallCache,
+          authoredHard,
+          authoredSoft,
+          authoredVariantRecords,
+          cached,
+          excluded,
+          queryText,
+          sessionId,
+          userId,
+          messages,
+          tokenBudget,
+          profiler,
+        });
+
+        const profileLines = profiler?.lines() ?? [];
+        if (profiler) {
+          profiler.emit();
+        }
 
+        return profileLines.length > 0
+          ? { ...result, _profile: profileLines }
+          : result;
+      } catch {
+        return {
+          messages,
+          estimatedTokens: countTokens(messages),
+          systemPromptAddition: "",
+        } satisfies ContextAssembleResult;
+      }
+    },
+    async assembleCore({
+      rpc,
+      cfg,
+      recallCache,
+      authoredHard,
+      authoredSoft,
+      authoredVariantRecords,
+      cached,
+      excluded,
+      queryText,
+      sessionId,
+      userId,
+      messages,
+      tokenBudget,
+      profiler,
+    }: {
+      rpc: Awaited<ReturnType<RpcGetter>>;
+      cfg: PluginConfig;
+      recallCache: RecallCache<SearchResult>;
+      authoredHard: SearchResult[];
+      authoredSoft: SearchResult[];
+      authoredVariantRecords: SearchResult[];
+      cached: ReturnType<RecallCache<SearchResult>["take"]>;
+      excluded: string[];
+      queryText: string;
+      sessionId: string;
+      userId: string;
+      messages: Array<{ role: string; content: string }>;
+      tokenBudget: number;
+      profiler: { mark(label: string): void; emit(): void } | null;
+    }): Promise<ContextAssembleResult> {
+      const memoryBudget = tokenBudget * (cfg.tokenBudgetFraction ?? 0.25);
+      const hardItems = authoredHard;
+      const hardUsed = tokenCostSum(hardItems);
+
+      profiler?.mark("session");
+      const sessionRecords = await rpc.call<{ results: SearchResult[] }>("list_by_meta", {
+        collection: `session:${sessionId}`,
+        key: "sessionId",
+        value: sessionId,
+      });
+      const rawSessionTurns = sortChronological(
+        sessionRecords.results.filter((item) =>
+          // cascade_tier is ranking metadata (cascade search tier); exclude from session history
+          item.metadata.type !== "summary" &&
+          item.metadata.type !== "guidance_shard" &&
+          typeof item.metadata.cascade_tier !== "number"
+        ),
+      );
+      const minTurns = cfg.continuityMinTurns ?? DEFAULT_CONTINUITY_MIN_TURNS;
+      const tailTarget = cfg.continuityTailBudgetTokens ?? DEFAULT_CONTINUITY_TAIL_BUDGET_TOKENS;
+      const baseTail = selectRecentTail(rawSessionTurns, {
+        minTurns,
+        tailBudgetTokens: 0,
+        tokenCost,
+        sameBundle: isContinuityBundleCoupled,
+      });
+      const baseTailUsed = baseTail.baseTokens;
+      const configuredHardFraction = clampFraction(cfg.authoredHardBudgetFraction);
+      const hardBudget = configuredHardFraction > 0 ? memoryBudget * configuredHardFraction : hardUsed;
+      const degradedReasons: string[] = [];
+      if (hardUsed > hardBudget + 1e-9) {
+        degradedReasons.push("hard authored invariants exceed configured hard budget reserve");
+      }
+      if (hardUsed + baseTailUsed > memoryBudget + 1e-9) {
+        degradedReasons.push("hard authored invariants plus mandatory recent-tail base exceed available memory budget");
+      }
+      if (degradedReasons.length > 0) {
+        const degradedTail = markRecentTail(baseTail.base, baseTail.base.length);
+        const selected = [...hardItems, ...degradedTail];
         const selectedMessages = selected.map((item) => ({
           role: "system",
           content: buildInjectedMemoryMessageContent(item),
         }));
-
         return {
           messages: [...selectedMessages, ...messages],
           estimatedTokens: countTokens(selectedMessages) + countTokens(messages),
-          systemPromptAddition: buildMemoryHeader(selected),
-        };
-      } catch {
-        return {
-          messages,
-          estimatedTokens: countTokens(messages),
-          systemPromptAddition: "",
+          systemPromptAddition: buildDegradedMemoryHeader(degradedReasons, selected),
         };
       }
+      const authoredSoftTarget = Math.max(0, memoryBudget * (cfg.authoredSoftBudgetFraction ?? 0.3));
+      const softBudget = Math.max(0, Math.min(authoredSoftTarget, memoryBudget - hardUsed - baseTailUsed));
+      const softItems = fitPromptBudget(authoredSoft, softBudget);
+      const remainingAfterHardSoft = Math.max(0, memoryBudget - hardUsed - tokenCostSum(softItems));
+      const effectiveTailBudget = Math.min(
+        Math.max(tailTarget, baseTailUsed),
+        remainingAfterHardSoft,
+      );
+      const recentTailSelection = selectRecentTail(rawSessionTurns, {
+        minTurns,
+        tailBudgetTokens: effectiveTailBudget,
+        tokenCost,
+        sameBundle: isContinuityBundleCoupled,
+      });
+      const recentTail = markRecentTail(
+        recentTailSelection.recent,
+        recentTailSelection.base.length,
+      );
+      const tailBaseItems = recentTail.slice(-recentTailSelection.base.length);
+      const tailExtensionItems = recentTail.slice(0, Math.max(0, recentTail.length - recentTailSelection.base.length));
+      const retrievalBudget = Math.max(0, memoryBudget - hardUsed - tokenCostSum(softItems) - tokenCostSum(recentTail));
+      const recentTailIDs = recentTail.map((item) => item.id);
+
+      const coarseTopK = Math.max(cfg.section7CoarseTopK ?? Math.max((cfg.topK ?? 8) * 2, 8), 1);
+      const sessionSearchTopK = Math.max(cfg.topK ?? 8, 1);
+      const secondPassTopK = Math.max(cfg.section7SecondPassTopK ?? (cfg.topK ?? 8), 1);
+      const searchSessionRecall = useSessionRecallProjection(cfg);
+      const searchSessionSummary = useSessionSummarySearchExperiment(cfg);
+      let sessionSearchCollection = `session:${sessionId}`;
+      let sessionExcludeIds = [...excluded, ...recentTailIDs];
+      if (searchSessionSummary) {
+        const summaryCollection = sessionSummaryCollection(sessionId);
+        const summaryRecords = await rpc.call<{ results: SearchResult[] }>("list_collection", {
+          collection: summaryCollection,
+        });
+        if (summaryRecords.results.length > 0) {
+          sessionSearchCollection = summaryCollection;
+          sessionExcludeIds = [...excluded];
+        }
+      } else if (searchSessionRecall) {
+        sessionSearchCollection = sessionRecallCollection(sessionId);
+        sessionExcludeIds = [...excluded, ...recentTailIDs.map(sessionRecallId)];
+      }
+
+      profiler?.mark("session_search");
+      const [sessionHits] = await Promise.all([
+        rpc.call<{ results: SearchResult[] }>("search_text", {
+          collection: sessionSearchCollection,
+          text: queryText,
+          k: sessionSearchTopK,
+          excludeIds: sessionExcludeIds,
+        }),
+      ]);
+
+      profiler?.mark("recall_user_global");
+      const [userHits, globalHits] = await Promise.all([
+        cached?.userHits
+          ? Promise.resolve({ results: cached.userHits })
+          : rpc.call<{ results: SearchResult[] }>("search_text", {
+              collection: `user:${userId}`,
+              text: queryText,
+              k: Math.ceil((cfg.topK ?? 8) / 2),
+            }),
+        cached?.globalHits
+          ? Promise.resolve({ results: cached.globalHits })
+          : rpc.call<{ results: SearchResult[] }>("search_text", {
+              collection: "global",
+              text: queryText,
+              k: Math.ceil((cfg.topK ?? 8) / 4),
+            }),
+      ]);
+
+      if (!cached) {
+        recallCache.put({
+          userId,
+          queryText,
+          durableVariantHits: [],
+          userHits: userHits.results,
+          globalHits: globalHits.results,
+        });
+      }
+
+      profiler?.mark("recall_authored_variant");
+      const authoredVariantKey = `${queryText}\n${coarseTopK}`;
+      const cachedAuthoredVariantHits = authoredVariantRecallCache.get(authoredVariantKey);
+      const [authoredVariantHits] = await Promise.all([
+        cachedAuthoredVariantHits
+          ? Promise.resolve({ results: cachedAuthoredVariantHits })
+          : rpc.call<{ results: SearchResult[] }>("search_text", {
+              collection: AUTHORED_VARIANT_COLLECTION,
+              text: queryText,
+              k: coarseTopK,
+            }),
+      ]);
+      if (!cachedAuthoredVariantHits) {
+        authoredVariantRecallCache.set(authoredVariantKey, authoredVariantHits.results);
+      }
+
+      profiler?.mark("recall_elevated");
+      const elevatedGeneration = elevatedRecallGeneration.get(sessionId) ?? 0;
+      const elevatedKey = `${sessionId}\n${elevatedGeneration}\n${userId}\n${queryText}`;
+      const cachedElevated = elevatedRecallCache.get(elevatedKey);
+      const [elevatedHits] = await Promise.all([
+        cachedElevated
+          ? Promise.resolve({ results: cachedElevated })
+          : rpc.call<{ results: SearchResult[] }>("search_text_collections", {
+              collections: [
+                `${ELEVATED_USER_COLLECTION_PREFIX}${userId}`,
+                `${ELEVATED_SESSION_COLLECTION_PREFIX}${sessionId}`,
+              ],
+              text: queryText,
+              k: coarseTopK,
+              excludeByCollection: {},
+            }),
+      ]);
+      if (!cachedElevated) {
+        elevatedRecallCache.set(elevatedKey, elevatedHits.results);
+      }
+
+      profiler?.mark("rank");
+      const ranked = rankSection7VariantCandidates(
+        [
+          ...annotateCollection(sessionHits.results, `session:${sessionId}`),
+          ...elevatedHits.results,
+          ...userHits.results,
+          ...globalHits.results,
+          ...authoredVariantHits.results,
+        ],
+        {
+          queryText,
+          k1: coarseTopK,
+          k2: secondPassTopK,
+          theta1: cfg.section7Theta1,
+          kappa: cfg.section7Kappa,
+          authorityRecencyLambda: cfg.section7AuthorityRecencyLambda,
+          authorityRecencyWeight: cfg.section7AuthorityRecencyWeight,
+          authorityFrequencyWeight: cfg.section7AuthorityFrequencyWeight,
+          authorityAuthoredWeight: cfg.section7AuthorityAuthoredWeight,
+          sessionId,
+          userId,
+        },
+      );
+
+      profiler?.mark("hop");
+      const hopExpanded = expandSection7HopCandidates(
+        ranked,
+        annotateCollection(authoredVariantRecords, AUTHORED_VARIANT_COLLECTION),
+        {
+          etaHop: cfg.section7HopEta,
+          thetaHop: cfg.section7HopThreshold,
+        },
+      );
+
+      profiler?.mark("fit");
+      const mergedCandidates = mergeSection7VariantCandidates(ranked, hopExpanded);
+      // Recovery trigger is evaluated before variant fitting so healthy sessions
+      // do not lose recall budget to an unused recovery reserve.
+      profiler?.mark("recovery_trigger");
+      const recoveryTrigger = detectRetrievalFailure(mergedCandidates, {
+        floorScore: cfg.recoveryFloorScore ?? 0.15,
+        minTopK: cfg.recoveryMinTopK ?? 4,
+        meanConfidenceThresh: cfg.recoveryMinConfidenceMean ?? 0.5,
+      });
+      const recoveryReserveTokens = recoveryTrigger.fire
+        ? Math.min(memoryBudget, Math.max(Math.floor(memoryBudget * 0.10), 16), 128)
+        : 0;
+      const elevatedGuidanceBudget = Math.max(
+        0,
+        Math.min(
+          memoryBudget * (cfg.elevatedGuidanceBudgetFraction ?? 0.15),
+          retrievalBudget,
+        ),
+      );
+      const elevatedItems = fitPromptBudget(
+        mergedCandidates.filter((item) => item.metadata.elevated_guidance === true),
+        elevatedGuidanceBudget,
+      );
+      const remainingAfterElevated = Math.max(0, retrievalBudget - tokenCostSum(elevatedItems));
+      const remainingForVariant = Math.max(0, remainingAfterElevated - recoveryReserveTokens);
+      const variantItems = fitPromptBudget(
+        mergedCandidates.filter((item) => item.metadata.elevated_guidance !== true),
+        remainingForVariant,
+      );
+
+      // Build set of theorem-selected IDs for recovery deduplication.
+      // Recovery should only append NEW raw evidence, not re-inject content already
+      // selected by the normal assembly path (hard/soft/tail/elevated/variant).
+      const theoremSelectedIDs = new Set([
+        ...hardItems.map((i) => i.id),
+        ...softItems.map((i) => i.id),
+        ...tailBaseItems.map((i) => i.id),
+        ...tailExtensionItems.map((i) => i.id),
+        ...elevatedItems.map((i) => i.id),
+        ...variantItems.map((i) => i.id),
+      ]);
+
+      // Recovery is a policy overlay — it appends raw content only when triggered,
+      // it never modifies the C_total(q) output and does not spend from tau_V.
+      let recoveryItems: SearchResult[] = [];
+      if (recoveryTrigger.fire) {
+        profiler?.mark("recovery_expand");
+        // Recovery searches immutable raw history directly — never the active view, elevated shards,
+        // or authored collections. Raw turns are immutable (storage axiom, unchanged).
+        const recoveryExcludeIDs = [...excluded, ...recentTailIDs, ...theoremSelectedIDs];
+        const rawResults = await rpc.call<{ results: SearchResult[] }>("query_raw_session", {
+          sessionId,
+          text: queryText,
+          k: Math.max(cfg.topK ?? 8, 4),
+          excludeIds: recoveryExcludeIDs,
+        });
+        // Fit recovered raw items to the reserved recovery budget — never exceed it.
+        const fittedRecovery = fitPromptBudget(rawResults.results ?? [], recoveryReserveTokens);
+        recoveryItems = fittedRecovery.map((item: SearchResult) => ({
+          ...item,
+          metadata: {
+            ...item.metadata,
+            recovery_fallback: true,
+          },
+        }));
+      }
+
+      const selected = [
+        ...hardItems,
+        ...tailBaseItems,
+        ...softItems,
+        ...tailExtensionItems,
+        ...elevatedItems,
+        ...variantItems,
+        ...recoveryItems,
+      ];
+      void rpc.call("bump_access_counts", {
+        updates: groupAccessCountUpdates([...elevatedItems, ...variantItems]),
+      }).catch(() => {});
+
+      profiler?.mark("render");
+      const selectedMessages = selected.map((item) => ({
+        role: "system",
+        content: buildInjectedMemoryMessageContent(item),
+      }));
+
+      return {
+        messages: [...selectedMessages, ...messages],
+        estimatedTokens: countTokens(selectedMessages) + countTokens(messages),
+        systemPromptAddition: buildMemoryHeader(selected),
+      };
     },
     async compact({ sessionId, force, targetSize }: ContextCompactArgs) {
       const rpc = await getRpc();
@@ -312,6 +613,9 @@ export function buildContextEngineFactory(
       const compacted = "didCompact" in result
         ? (result.didCompact ?? result.compacted ?? false)
         : (result.compacted ?? false);
+      if (compacted && useSessionRecallProjection(cfg)) {
+        await rebuildSessionRecallProjection(rpc, cfg, sessionId);
+      }
 
       return {
         ok: true,
@@ -321,12 +625,103 @@ export function buildContextEngineFactory(
   };
 }
 
+function useSessionRecallProjection(cfg: PluginConfig): boolean {
+  return cfg.useSessionRecallProjection === true;
+}
+
+function useSessionSummarySearchExperiment(cfg: PluginConfig): boolean {
+  return cfg.useSessionSummarySearchExperiment === true;
+}
+
+function sessionRecallCollection(sessionId: string): string {
+  return `${SESSION_RECALL_COLLECTION_PREFIX}${sessionId}`;
+}
+
+function sessionRawCollection(sessionId: string): string {
+  return `${SESSION_RAW_COLLECTION_PREFIX}${sessionId}`;
+}
+
+function sessionSummaryCollection(sessionId: string): string {
+  return `${SESSION_SUMMARY_COLLECTION_PREFIX}${sessionId}`;
+}
+
+function sessionEdgeCollection(sessionId: string): string {
+  return `${SESSION_EDGE_COLLECTION_PREFIX}${sessionId}`;
+}
+
+function sessionStateCollection(sessionId: string): string {
+  return `${SESSION_STATE_COLLECTION_PREFIX}${sessionId}`;
+}
+
+function sessionRecallId(sourceId: string): string {
+  return `recall:${sourceId}`;
+}
+
+async function rebuildSessionRecallProjection(
+  rpc: Awaited<ReturnType<RpcGetter>>,
+  cfg: PluginConfig,
+  sessionId: string,
+): Promise<void> {
+  const rawCollection = `session:${sessionId}`;
+  const projectionCollection = sessionRecallCollection(sessionId);
+  const sessionRecords = await rpc.call<{ results: SearchResult[] }>("list_by_meta", {
+    collection: rawCollection,
+    key: "sessionId",
+    value: sessionId,
+  });
+  const rawSessionTurns = sortChronological(
+    sessionRecords.results.filter((item) =>
+      // cascade_tier is ranking metadata (cascade search tier); exclude from session history
+      item.metadata.type !== "summary" &&
+      item.metadata.type !== "guidance_shard" &&
+      typeof item.metadata.cascade_tier !== "number"
+    ),
+  );
+  const recentTail = selectRecentTail(rawSessionTurns, {
+    minTurns: cfg.continuityMinTurns ?? DEFAULT_CONTINUITY_MIN_TURNS,
+    tailBudgetTokens: cfg.continuityTailBudgetTokens ?? DEFAULT_CONTINUITY_TAIL_BUDGET_TOKENS,
+    tokenCost,
+    sameBundle: isContinuityBundleCoupled,
+  });
+  const projectionItems = recentTail.older;
+  const existingProjection = await rpc.call<{ results: SearchResult[] }>("list_collection", {
+    collection: projectionCollection,
+  });
+  const existingIds = existingProjection.results
+    .map((item) => item.id)
+    .filter((id): id is string => typeof id === "string" && id.length > 0);
+  if (existingIds.length > 0) {
+    await rpc.call("delete_batch", {
+      collection: projectionCollection,
+      ids: existingIds,
+    });
+  }
+  await Promise.all(projectionItems.map((item) =>
+    rpc.call("insert_text", {
+      collection: projectionCollection,
+      id: sessionRecallId(item.id),
+      score: item.score,
+      text: item.text,
+      metadata: {
+        ...item.metadata,
+        projection_class: "session_recall",
+        source_turn_id: item.id,
+        source_turn_ts: metadataTimestamp(item),
+      },
+    })
+  ));
+}
+
 async function loadAuthoredCollections(
   rpc: Awaited<ReturnType<RpcGetter>>,
   cached: { hard: SearchResult[] | null; soft: SearchResult[] | null; variant: SearchResult[] | null },
 ): Promise<[SearchResult[], SearchResult[], SearchResult[]]> {
   if (cached.hard && cached.soft && cached.variant) {
-    return [cached.hard, cached.soft, cached.variant];
+    return [
+      sortAuthoredItems(cached.hard),
+      sortAuthoredItems(cached.soft),
+      sortAuthoredItems(cached.variant),
+    ];
   }
 
   const [hard, soft, variant] = await Promise.all([
@@ -341,7 +736,11 @@ async function loadAuthoredCollections(
       : rpc.call<{ results: SearchResult[] }>("list_collection", { collection: AUTHORED_VARIANT_COLLECTION }),
   ]);
 
-  return [hard.results, soft.results, variant.results];
+  return [
+    sortAuthoredItems(hard.results),
+    sortAuthoredItems(soft.results),
+    sortAuthoredItems(variant.results),
+  ];
 }
 
 function tokenCostSum(items: SearchResult[]): number {
@@ -372,6 +771,11 @@ function metadataTimestamp(item: SearchResult): number {
   return typeof raw === "number" && Number.isFinite(raw) ? raw : 0;
 }
 
+function metadataNumber(item: SearchResult, key: string): number {
+  const raw = item.metadata[key];
+  return typeof raw === "number" && Number.isFinite(raw) ? raw : 0;
+}
+
 function markRecentTail(items: SearchResult[], baseCount: number): SearchResult[] {
   const baseStart = Math.max(0, items.length - baseCount);
   return items.map((item, idx) => ({
@@ -394,6 +798,30 @@ function annotateCollection(items: SearchResult[], collection: string): SearchRe
   }));
 }
 
+function sortAuthoredItems(items: SearchResult[]): SearchResult[] {
+  return [...items].sort((left, right) => {
+    const leftDoc = typeof left.metadata.source_doc === "string" ? left.metadata.source_doc : "";
+    const rightDoc = typeof right.metadata.source_doc === "string" ? right.metadata.source_doc : "";
+    if (leftDoc !== rightDoc) {
+      return leftDoc.localeCompare(rightDoc);
+    }
+
+    const leftPosition = metadataNumber(left, "position");
+    const rightPosition = metadataNumber(right, "position");
+    if (leftPosition !== rightPosition) {
+      return leftPosition - rightPosition;
+    }
+
+    const leftOrdinal = metadataNumber(left, "ordinal");
+    const rightOrdinal = metadataNumber(right, "ordinal");
+    if (leftOrdinal !== rightOrdinal) {
+      return leftOrdinal - rightOrdinal;
+    }
+
+    return left.id.localeCompare(right.id);
+  });
+}
+
 function groupAccessCountUpdates(items: SearchResult[]): Array<{ collection: string; ids: string[] }> {
   const grouped = new Map<string, string[]>();
   for (const item of items) {
@@ -464,3 +892,14 @@ function isContinuityBundleCoupled(left: SearchResult, right: SearchResult): boo
     (leftRole === "assistant" && rightRole === "user")
   );
 }
+
+function stabilityWeightForMessage(role: string): number {
+  switch (role) {
+    case "user":
+      return 0.5;
+    case "assistant":
+      return 0.25;
+    default:
+      return 0.2;
+  }
+}