feat: Phase 7A — semantic memory dedup ("sleep cycle" V1)

New table memory_merge_candidates + service functions to cluster
near-duplicate active memories within (project, memory_type) buckets,
draft a unified content via LLM, and merge on human approval. Source
memories become superseded (never deleted); merged memory carries
union of tags, max of confidence, sum of reference_count.

- schema migration for memory_merge_candidates
- atocore.memory.similarity: cosine + transitive clustering
- atocore.memory._dedup_prompt: stdlib-only LLM prompt preserving every specific
- service: merge_memories / create_merge_candidate / get_merge_candidates / reject_merge_candidate
- scripts/memory_dedup.py: host-side detector (HTTP-only, idempotent)
- 5 API endpoints under /admin/memory/merge-candidates* + /admin/memory/dedup-scan
- triage UI: purple "🔗 Merge Candidates" section + "🔗 Scan for duplicates" bar
- batch-extract.sh Step B3 (0.90 daily, 0.85 Sundays)
- deploy/dalidou/dedup-watcher.sh for UI-triggered scans
- 21 new tests (374 → 395)
- docs/PHASE-7-MEMORY-CONSOLIDATION.md covering 7A-7H roadmap

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

DEV-LEDGER.md

@@ -7,9 +7,9 @@
 ## Orientation
 
 - **live_sha** (Dalidou `/health` build_sha): `775960c` (verified 2026-04-16 via /health, build_time 2026-04-16T17:59:30Z)
-- **last_updated**: 2026-04-16 by Claude ("Make It Actually Useful" sprint — observability + Phase 10)
+- **last_updated**: 2026-04-18 by Claude (Phase 7A — Memory Consolidation "sleep cycle" V1 on branch, not yet deployed)
 - **main_tip**: `999788b`
-- **test_count**: 303 (4 new Phase 10 tests)
+- **test_count**: 395 (21 new Phase 7A dedup tests + accumulated Phase 5/6 tests since last ledger refresh)
 - **harness**: `17/18 PASS` on live Dalidou (p04-constraints expects "Zerodur" — retrieval content gap, not regression)
 - **vectors**: 33,253
 - **active_memories**: 84 (31 project, 23 knowledge, 10 episodic, 8 adaptation, 7 preference, 5 identity)
@@ -160,6 +160,8 @@ One branch `codex/extractor-eval-loop` for Day 1-5, a second `codex/retrieval-ha
 
 ## Session Log
 
+- **2026-04-18 Claude** **Phase 7A — Memory Consolidation V1 ("sleep cycle") landed on branch.** New `docs/PHASE-7-MEMORY-CONSOLIDATION.md` covers all 8 subphases (7A dedup, 7B contradictions, 7C tag canon, 7D confidence decay, 7E memory detail, 7F domain view, 7F re-extract, 7H vector hygiene). 7A implementation: schema migration `memory_merge_candidates`, `atocore.memory.similarity` (cosine + transitive cluster), stdlib-only `atocore.memory._dedup_prompt` (llm drafts unified content preserving all specifics), `merge_memories()` + `create_merge_candidate()` + `get_merge_candidates()` + `reject_merge_candidate()` in service.py, host-side `scripts/memory_dedup.py` (HTTP + claude -p, idempotent via sorted-id set), 5 new endpoints under `/admin/memory/merge-candidates*` + `/admin/memory/dedup-scan` + `/admin/memory/dedup-status`, purple-themed "🔗 Merge Candidates" section in /admin/triage with editable draft + approve/reject buttons, "🔗 Scan for duplicates" control bar with threshold slider, nightly Step B3 in batch-extract.sh (0.90 daily, 0.85 Sundays deep), `deploy/dalidou/dedup-watcher.sh` host watcher for UI-triggered scans (mirrors graduation-watcher pattern). 21 new tests (similarity, prompt parse, idempotency, merge happy path, override content/tags, audit rows, abort-if-source-tampered, reject leaves sources alone, schema). Tests 374 → 395. Not yet deployed; harness not re-run. Next: push + deploy, install `dedup-watcher.sh` in host cron, trigger first scan, review proposals in UI.
+
 - **2026-04-16 Claude** `b687e7f..999788b` **"Make It Actually Useful" sprint.** Two-part session: ops fixes then consolidation sprint.
 
   **Part 1 — Ops fixes:** Deployed `b687e7f` (project inference from cwd). Fixed cron logging (was `/dev/null` — redirected to `~/atocore-logs/`). Fixed OpenClaw gateway crash-loop (`discord.replyToMode: "any"` invalid → `"all"`). Deployed `atocore-capture` plugin on T420 OpenClaw using `before_agent_start` + `llm_output` hooks — verified end-to-end: 38 `client=openclaw` interactions captured. Backfilled project tags on 179/181 unscoped interactions (165 atocore, 8 p06, 6 p04).

deploy/dalidou/batch-extract.sh

@@ -166,6 +166,26 @@ curl -sSf -X POST "$ATOCORE_URL/admin/memory/extend-reinforced" \
     log "WARN: extend-reinforced failed (non-blocking)"
 }
 
+# Step B3: Memory dedup scan (Phase 7A)
+# Nightly at 0.90 (tight — only near-duplicates). Sundays run a deeper
+# pass at 0.85 to catch semantically-similar-but-differently-worded memories.
+if [[ "$(date -u +%u)" == "7" ]]; then
+    DEDUP_THRESHOLD="0.85"
+    DEDUP_BATCH="80"
+    log "Step B3: memory dedup (Sunday deep pass, threshold $DEDUP_THRESHOLD)"
+else
+    DEDUP_THRESHOLD="0.90"
+    DEDUP_BATCH="50"
+    log "Step B3: memory dedup (daily, threshold $DEDUP_THRESHOLD)"
+fi
+python3 "$APP_DIR/scripts/memory_dedup.py" \
+    --base-url "$ATOCORE_URL" \
+    --similarity-threshold "$DEDUP_THRESHOLD" \
+    --max-batch "$DEDUP_BATCH" \
+    2>&1 || {
+    log "WARN: memory dedup failed (non-blocking)"
+}
+
 # Step G: Integrity check (Phase 4 V1)
 log "Step G: integrity check"
 python3 "$APP_DIR/scripts/integrity_check.py" \

deploy/dalidou/dedup-watcher.sh

@@ -0,0 +1,110 @@
+#!/usr/bin/env bash
+#
+# deploy/dalidou/dedup-watcher.sh
+# -------------------------------
+# Host-side watcher for on-demand memory dedup scans (Phase 7A).
+#
+# The /admin/triage page has a "🔗 Scan for duplicates" button that POSTs
+# to /admin/memory/dedup-scan with {project, similarity_threshold, max_batch}.
+# The container writes this to project_state (atocore/config/dedup_requested_at).
+#
+# This script runs on the Dalidou HOST (where claude CLI lives), polls
+# for the flag, and runs memory_dedup.py when seen.
+#
+# Installed via cron every 2 minutes:
+#   */2 * * * * /srv/storage/atocore/app/deploy/dalidou/dedup-watcher.sh \
+#     >> /home/papa/atocore-logs/dedup-watcher.log 2>&1
+#
+# Mirrors deploy/dalidou/graduation-watcher.sh exactly.
+
+set -euo pipefail
+
+ATOCORE_URL="${ATOCORE_URL:-http://127.0.0.1:8100}"
+APP_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")/../.." && pwd)"
+LOCK_FILE="/tmp/atocore-dedup.lock"
+LOG_DIR="/home/papa/atocore-logs"
+mkdir -p "$LOG_DIR"
+
+TS="$(date -u +%Y-%m-%dT%H:%M:%SZ)"
+log() { printf '[%s] %s\n' "$TS" "$*"; }
+
+# Fetch the flag via API
+STATE_JSON=$(curl -sSf --max-time 5 "$ATOCORE_URL/project/state/atocore" 2>/dev/null || echo "{}")
+REQUESTED=$(echo "$STATE_JSON" | python3 -c "
+import sys, json
+try:
+    d = json.load(sys.stdin)
+    for e in d.get('entries', d.get('state', [])):
+        if e.get('category') == 'config' and e.get('key') == 'dedup_requested_at':
+            print(e.get('value', ''))
+            break
+except Exception:
+    pass
+" 2>/dev/null || echo "")
+
+if [[ -z "$REQUESTED" ]]; then
+    exit 0
+fi
+
+PROJECT=$(echo "$REQUESTED" | python3 -c "import sys,json; print(json.loads(sys.stdin.read() or '{}').get('project',''))" 2>/dev/null || echo "")
+THRESHOLD=$(echo "$REQUESTED" | python3 -c "import sys,json; print(json.loads(sys.stdin.read() or '{}').get('similarity_threshold',0.88))" 2>/dev/null || echo "0.88")
+MAX_BATCH=$(echo "$REQUESTED" | python3 -c "import sys,json; print(json.loads(sys.stdin.read() or '{}').get('max_batch',50))" 2>/dev/null || echo "50")
+
+# Acquire lock
+exec 9>"$LOCK_FILE" || exit 0
+if ! flock -n 9; then
+    log "dedup already running, skipping"
+    exit 0
+fi
+
+# Mark running
+curl -sSf -X POST "$ATOCORE_URL/project/state" \
+    -H 'Content-Type: application/json' \
+    -d "{\"project\":\"atocore\",\"category\":\"status\",\"key\":\"dedup_running\",\"value\":\"1\",\"source\":\"dedup watcher\"}" \
+    >/dev/null 2>&1 || true
+curl -sSf -X POST "$ATOCORE_URL/project/state" \
+    -H 'Content-Type: application/json' \
+    -d "{\"project\":\"atocore\",\"category\":\"status\",\"key\":\"dedup_last_started_at\",\"value\":\"$TS\",\"source\":\"dedup watcher\"}" \
+    >/dev/null 2>&1 || true
+
+LOG_FILE="$LOG_DIR/dedup-ondemand-$(date -u +%Y%m%d-%H%M%S).log"
+log "Starting dedup (project='$PROJECT' threshold=$THRESHOLD max_batch=$MAX_BATCH, log: $LOG_FILE)"
+
+# Clear the flag BEFORE running so duplicate clicks queue at most one
+curl -sSf -X DELETE "$ATOCORE_URL/project/state" \
+    -H 'Content-Type: application/json' \
+    -d "{\"project\":\"atocore\",\"category\":\"config\",\"key\":\"dedup_requested_at\"}" \
+    >/dev/null 2>&1 || true
+
+cd "$APP_DIR"
+export PYTHONPATH="$APP_DIR/src:${PYTHONPATH:-}"
+ARGS=(--base-url "$ATOCORE_URL" --similarity-threshold "$THRESHOLD" --max-batch "$MAX_BATCH")
+if [[ -n "$PROJECT" ]]; then
+    ARGS+=(--project "$PROJECT")
+fi
+
+if python3 scripts/memory_dedup.py "${ARGS[@]}" >> "$LOG_FILE" 2>&1; then
+    RESULT=$(grep "^summary:" "$LOG_FILE" | tail -1 || tail -1 "$LOG_FILE")
+    RESULT="${RESULT:-completed}"
+    log "dedup finished: $RESULT"
+else
+    RESULT="ERROR — see $LOG_FILE"
+    log "dedup FAILED"
+fi
+
+FINISH_TS="$(date -u +%Y-%m-%dT%H:%M:%SZ)"
+
+curl -sSf -X POST "$ATOCORE_URL/project/state" \
+    -H 'Content-Type: application/json' \
+    -d "{\"project\":\"atocore\",\"category\":\"status\",\"key\":\"dedup_running\",\"value\":\"0\",\"source\":\"dedup watcher\"}" \
+    >/dev/null 2>&1 || true
+curl -sSf -X POST "$ATOCORE_URL/project/state" \
+    -H 'Content-Type: application/json' \
+    -d "{\"project\":\"atocore\",\"category\":\"status\",\"key\":\"dedup_last_finished_at\",\"value\":\"$FINISH_TS\",\"source\":\"dedup watcher\"}" \
+    >/dev/null 2>&1 || true
+
+SAFE_RESULT=$(printf '%s' "$RESULT" | python3 -c "import sys,json; print(json.dumps(sys.stdin.read())[1:-1])")
+curl -sSf -X POST "$ATOCORE_URL/project/state" \
+    -H 'Content-Type: application/json' \
+    -d "{\"project\":\"atocore\",\"category\":\"status\",\"key\":\"dedup_last_result\",\"value\":\"$SAFE_RESULT\",\"source\":\"dedup watcher\"}" \
+    >/dev/null 2>&1 || true

docs/PHASE-7-MEMORY-CONSOLIDATION.md

@@ -0,0 +1,96 @@
+# Phase 7 — Memory Consolidation (the "Sleep Cycle")
+
+**Status**: 7A in progress · 7B-H scoped, deferred
+**Design principle**: *"Like human memory while sleeping, but more robotic — never discard relevant details. Consolidate, update, supersede — don't delete."*
+
+## Why
+
+Phases 1–6 built capture + triage + graduation + emerging-project detection. What they don't solve:
+
+| # | Problem | Fix |
+|---|---|---|
+| 1 | Redundancy — "APM uses NX" said 5 different ways across 5 memories | **7A** Semantic dedup |
+| 2 | Latent contradictions — "chose Zygo" + "switched from Zygo" both active | **7B** Pair contradiction detection |
+| 3 | Tag drift — `firmware`, `fw`, `firmware-control` fragment retrieval | **7C** Tag canonicalization |
+| 4 | Confidence staleness — 6-month unreferenced memory ranks as fresh | **7D** Confidence decay |
+| 5 | No memory drill-down page | **7E** `/wiki/memories/{id}` |
+| 6 | Domain knowledge siloed per project | **7F** `/wiki/domains/{tag}` |
+| 7 | Prompt upgrades (llm-0.5 → 0.6) don't re-process old interactions | **7G** Re-extraction on version bump |
+| 8 | Superseded memory vectors still in Chroma polluting retrieval | **7H** Vector hygiene |
+
+Collectively: the brain needs a nightly pass that looks at what it already knows and tidies up — dedup, resolve contradictions, canonicalize tags, decay stale facts — **without losing information**.
+
+## Subphases
+
+### 7A — Semantic dedup + consolidation *(this sprint)*
+
+Compute embeddings on active memories, find pairs within `(project, memory_type)` bucket above similarity threshold (default 0.88), cluster, draft a unified memory via LLM, human approves in triage UI. On approve: sources become `superseded`, new merged memory created with union of `source_refs`, sum of `reference_count`, max of `confidence`. **Ships first** because redundancy compounds — every new memory potentially duplicates an old one.
+
+Detailed spec lives in the working plan (`dapper-cooking-tower.md`) and across the files listed under "Files touched" below. Key decisions:
+
+- LLM drafts, human approves — no silent auto-merge.
+- Same `(project, memory_type)` bucket only. Cross-project merges are rare + risky → separate flow in 7B.
+- Recompute embeddings each scan (~2s / 335 memories). Persist only if scan time becomes a problem.
+- Cluster-based proposals (A~B~C → one merge), not pair-based.
+- `status=superseded` never deleted — still queryable with filter.
+
+**Schema**: new table `memory_merge_candidates` (pending | approved | rejected).
+**Cron**: nightly at threshold 0.90 (tight); weekly (Sundays) at 0.85 (deeper cleanup).
+**UI**: new "🔗 Merge Candidates" section in `/admin/triage`.
+
+**Files touched in 7A**:
+- `src/atocore/models/database.py` — migration
+- `src/atocore/memory/similarity.py` — new, `compute_memory_similarity()`
+- `src/atocore/memory/_dedup_prompt.py` — new, shared LLM prompt
+- `src/atocore/memory/service.py` — `merge_memories()`
+- `scripts/memory_dedup.py` — new, host-side detector (HTTP-only)
+- `src/atocore/api/routes.py` — 5 new endpoints under `/admin/memory/`
+- `src/atocore/engineering/triage_ui.py` — merge cards section
+- `deploy/dalidou/batch-extract.sh` — Step B3
+- `deploy/dalidou/dedup-watcher.sh` — new, UI-triggered scans
+- `tests/test_memory_dedup.py` — ~10-15 new tests
+
+### 7B — Memory-to-memory contradiction detection
+
+Same embedding-pair machinery as 7A but within a *different* band (similarity 0.70–0.88 — semantically related but different wording). LLM classifies each pair: `duplicate | complementary | contradicts | supersedes-older`. Contradictions write a `memory_conflicts` row + surface a triage badge. Clear supersessions (both tier 1 sonnet and tier 2 opus agree) auto-mark the older as `superseded`.
+
+### 7C — Tag canonicalization
+
+Weekly LLM pass over `domain_tags` distribution, proposes `alias → canonical` map (e.g. `fw → firmware`). Human approves via UI (one-click pattern, same as emerging-project registration). Bulk-rewrites `domain_tags` atomically across all memories.
+
+### 7D — Confidence decay
+
+Daily lightweight job. For memories with `reference_count=0` AND `last_referenced_at` older than 30 days: multiply confidence by 0.97/day (~2-month half-life). Reinforcement already bumps confidence. Below 0.3 → auto-supersede with reason `decayed, no references`. Reversible (tune half-life), non-destructive (still searchable with status filter).
+
+### 7E — Memory detail page `/wiki/memories/{id}`
+
+Provenance chain: source_chunk → interaction → graduated_to_entity. Audit trail (Phase 4 has the data). Related memories (same project + tag + semantic neighbors). Decay trajectory plot (if 7D ships). Link target from every memory surfaced anywhere in the wiki.
+
+### 7F — Cross-project domain view `/wiki/domains/{tag}`
+
+One page per `domain_tag` showing all memories + graduated entities with that tag, grouped by project. "Optics across p04+p05+p06" becomes a real navigable page. Answers the long-standing question the tag system was meant to enable.
+
+### 7G — Re-extraction on prompt upgrade
+
+`batch_llm_extract_live.py --force-reextract --since DATE`. Dedupe key: `(interaction_id, extractor_version)` — same run on same interaction doesn't double-create. Triggered manually when `LLM_EXTRACTOR_VERSION` bumps. Not automatic (destructive).
+
+### 7H — Vector store hygiene
+
+Nightly: scan `source_chunks` and `memory_embeddings` (added in 7A V2) for `status=superseded|invalid`. Delete matching vectors from Chroma. Fail-open — the retrieval harness catches any real regression.
+
+## Verification & ship order
+
+1. **7A** — ship + observe 1 week → validate merge proposals are high-signal, rejection rate acceptable
+2. **7D** — decay is low-risk + high-compounding value; ship second
+3. **7C** — clean up tag fragmentation before 7F depends on canonical tags
+4. **7E** + **7F** — UX surfaces; ship together once data is clean
+5. **7B** — contradictions flow (pairs harder than duplicates to classify; wait for 7A data to tune threshold)
+6. **7G** — on-demand; no ship until we actually bump the extractor prompt
+7. **7H** — housekeeping; after 7A + 7B + 7D have generated enough `superseded` rows to matter
+
+## Scope NOT in Phase 7
+
+- Graduated memories (entity-descended) are **frozen** — exempt from dedup/decay. Entity consolidation is a separate Phase (8+).
+- Auto-merging without human approval (always human-in-the-loop in V1).
+- Summarization / compression — a different problem (reducing the number of chunks per memory, not the number of memories).
+- Forgetting policies — there's no user-facing "delete this" flow in Phase 7. Supersede + filter covers the need.

scripts/memory_dedup.py

@@ -0,0 +1,278 @@
+#!/usr/bin/env python3
+"""Phase 7A — semantic memory dedup detector.
+
+Finds clusters of near-duplicate active memories and writes merge-
+candidate proposals for human review in the triage UI.
+
+Algorithm:
+  1. Fetch active memories via HTTP
+  2. Group by (project, memory_type) — cross-bucket merges are deferred
+     to Phase 7B contradiction flow
+  3. Within each group, embed contents via atocore.retrieval.embeddings
+  4. Greedy transitive cluster at similarity >= threshold
+  5. For each cluster of size >= 2, ask claude-p to draft unified content
+  6. POST the proposal to /admin/memory/merge-candidates/create (server-
+     side dedupes by the sorted memory-id set, so re-runs don't double-
+     create)
+
+Host-side because claude CLI lives on Dalidou, not the container. Reuses
+the same PYTHONPATH=src pattern as scripts/graduate_memories.py for
+atocore imports (embeddings, similarity, prompt module).
+
+Usage:
+  python3 scripts/memory_dedup.py --base-url http://127.0.0.1:8100 \\
+      --similarity-threshold 0.88 --max-batch 50
+
+Threshold conventions (see Phase 7 doc):
+  0.88  interactive / default — balanced precision/recall
+  0.90  nightly cron — tight, only near-duplicates
+  0.85  weekly cron — deeper cleanup
+"""
+
+from __future__ import annotations
+
+import argparse
+import json
+import os
+import shutil
+import subprocess
+import sys
+import tempfile
+import time
+import urllib.error
+import urllib.request
+from collections import defaultdict
+from typing import Any
+
+# Make src/ importable — same pattern as graduate_memories.py
+_SCRIPT_DIR = os.path.dirname(os.path.abspath(__file__))
+_SRC_DIR = os.path.abspath(os.path.join(_SCRIPT_DIR, "..", "src"))
+if _SRC_DIR not in sys.path:
+    sys.path.insert(0, _SRC_DIR)
+
+from atocore.memory._dedup_prompt import (  # noqa: E402
+    DEDUP_PROMPT_VERSION,
+    SYSTEM_PROMPT,
+    build_user_message,
+    normalize_merge_verdict,
+    parse_merge_verdict,
+)
+from atocore.memory.similarity import cluster_by_threshold  # noqa: E402
+
+DEFAULT_BASE_URL = os.environ.get("ATOCORE_BASE_URL", "http://127.0.0.1:8100")
+DEFAULT_MODEL = os.environ.get("ATOCORE_DEDUP_MODEL", "sonnet")
+DEFAULT_TIMEOUT_S = float(os.environ.get("ATOCORE_DEDUP_TIMEOUT_S", "60"))
+
+_sandbox_cwd = None
+
+
+def get_sandbox_cwd() -> str:
+    global _sandbox_cwd
+    if _sandbox_cwd is None:
+        _sandbox_cwd = tempfile.mkdtemp(prefix="ato-dedup-")
+    return _sandbox_cwd
+
+
+def api_get(base_url: str, path: str) -> dict:
+    req = urllib.request.Request(f"{base_url}{path}")
+    with urllib.request.urlopen(req, timeout=30) as resp:
+        return json.loads(resp.read().decode("utf-8"))
+
+
+def api_post(base_url: str, path: str, body: dict | None = None) -> dict:
+    data = json.dumps(body or {}).encode("utf-8")
+    req = urllib.request.Request(
+        f"{base_url}{path}", method="POST",
+        headers={"Content-Type": "application/json"}, data=data,
+    )
+    with urllib.request.urlopen(req, timeout=30) as resp:
+        return json.loads(resp.read().decode("utf-8"))
+
+
+def call_claude(system_prompt: str, user_message: str, model: str, timeout_s: float) -> tuple[str | None, str | None]:
+    """Shared CLI caller with retry + stderr capture (mirrors auto_triage)."""
+    if not shutil.which("claude"):
+        return None, "claude CLI not available"
+    args = [
+        "claude", "-p",
+        "--model", model,
+        "--append-system-prompt", system_prompt,
+        "--disable-slash-commands",
+        user_message,
+    ]
+    last_error = ""
+    for attempt in range(3):
+        if attempt > 0:
+            time.sleep(2 ** attempt)
+        try:
+            completed = subprocess.run(
+                args, capture_output=True, text=True,
+                timeout=timeout_s, cwd=get_sandbox_cwd(),
+                encoding="utf-8", errors="replace",
+            )
+        except subprocess.TimeoutExpired:
+            last_error = f"{model} timed out"
+            continue
+        except Exception as exc:
+            last_error = f"subprocess error: {exc}"
+            continue
+        if completed.returncode == 0:
+            return (completed.stdout or "").strip(), None
+        stderr = (completed.stderr or "").strip()[:200]
+        last_error = f"{model} exit {completed.returncode}: {stderr}" if stderr else f"{model} exit {completed.returncode}"
+    return None, last_error
+
+
+def fetch_active_memories(base_url: str, project: str | None) -> list[dict]:
+    # The /memory endpoint with active_only=true returns active memories.
+    # Graduated memories are exempt from dedup — they're frozen pointers
+    # to entities. Filter them out on the client side.
+    params = "active_only=true&limit=2000"
+    if project:
+        params += f"&project={urllib.request.quote(project)}"
+    try:
+        result = api_get(base_url, f"/memory?{params}")
+    except Exception as e:
+        print(f"ERROR: could not fetch memories: {e}", file=sys.stderr)
+        return []
+    mems = result.get("memories", [])
+    return [m for m in mems if (m.get("status") or "active") == "active"]
+
+
+def group_memories(mems: list[dict]) -> dict[tuple[str, str], list[dict]]:
+    """Bucket by (project, memory_type). Empty project is its own bucket."""
+    buckets: dict[tuple[str, str], list[dict]] = defaultdict(list)
+    for m in mems:
+        key = ((m.get("project") or "").strip().lower(), (m.get("memory_type") or "").strip().lower())
+        buckets[key].append(m)
+    return buckets
+
+
+def draft_merge(sources: list[dict], model: str, timeout_s: float) -> dict[str, Any] | None:
+    user_msg = build_user_message(sources)
+    raw, err = call_claude(SYSTEM_PROMPT, user_msg, model, timeout_s)
+    if err:
+        print(f"  WARN: claude call failed: {err}", file=sys.stderr)
+        return None
+    parsed = parse_merge_verdict(raw or "")
+    if parsed is None:
+        print(f"  WARN: could not parse verdict: {(raw or '')[:200]}", file=sys.stderr)
+        return None
+    return normalize_merge_verdict(parsed)
+
+
+def submit_candidate(
+    base_url: str,
+    memory_ids: list[str],
+    similarity: float,
+    verdict: dict[str, Any],
+    dry_run: bool,
+) -> str | None:
+    body = {
+        "memory_ids": memory_ids,
+        "similarity": similarity,
+        "proposed_content": verdict["content"],
+        "proposed_memory_type": verdict["memory_type"],
+        "proposed_project": verdict["project"],
+        "proposed_tags": verdict["domain_tags"],
+        "proposed_confidence": verdict["confidence"],
+        "reason": verdict["reason"],
+    }
+    if dry_run:
+        print(f"  [dry-run] would POST: {json.dumps(body)[:200]}...")
+        return "dry-run"
+    try:
+        result = api_post(base_url, "/admin/memory/merge-candidates/create", body)
+        return result.get("candidate_id")
+    except urllib.error.HTTPError as e:
+        print(f"  ERROR: submit failed: {e.code} {e.read().decode()[:200]}", file=sys.stderr)
+        return None
+    except Exception as e:
+        print(f"  ERROR: submit failed: {e}", file=sys.stderr)
+        return None
+
+
+def main() -> None:
+    parser = argparse.ArgumentParser(description="Phase 7A semantic dedup detector")
+    parser.add_argument("--base-url", default=DEFAULT_BASE_URL)
+    parser.add_argument("--project", default="", help="Only scan this project (empty = all)")
+    parser.add_argument("--similarity-threshold", type=float, default=0.88)
+    parser.add_argument("--max-batch", type=int, default=50,
+                        help="Max clusters to propose per run")
+    parser.add_argument("--model", default=DEFAULT_MODEL)
+    parser.add_argument("--timeout-s", type=float, default=DEFAULT_TIMEOUT_S)
+    parser.add_argument("--dry-run", action="store_true")
+    args = parser.parse_args()
+
+    base = args.base_url.rstrip("/")
+
+    print(f"memory_dedup {DEDUP_PROMPT_VERSION} | threshold={args.similarity_threshold} | model={args.model}")
+    mems = fetch_active_memories(base, args.project or None)
+    print(f"fetched {len(mems)} active memories")
+    if not mems:
+        return
+
+    buckets = group_memories(mems)
+    print(f"grouped into {len(buckets)} (project, memory_type) buckets")
+
+    clusters_found = 0
+    candidates_created = 0
+    skipped_existing = 0
+    llm_rejections = 0
+
+    for (proj, mtype), group in sorted(buckets.items()):
+        if len(group) < 2:
+            continue
+        if candidates_created >= args.max_batch:
+            print(f"reached max-batch={args.max_batch}, stopping")
+            break
+
+        texts = [(m.get("content") or "") for m in group]
+        clusters = cluster_by_threshold(texts, args.similarity_threshold)
+        # Keep only non-trivial clusters
+        clusters = [c for c in clusters if len(c) >= 2]
+        if not clusters:
+            continue
+
+        print(f"\n[{proj or '(global)'}/{mtype}] {len(group)} mems → {len(clusters)} cluster(s)")
+        for cluster in clusters:
+            if candidates_created >= args.max_batch:
+                break
+            clusters_found += 1
+            sources = [group[i] for i in cluster]
+            ids = [s["id"] for s in sources]
+            # Approximate cluster similarity = min pairwise within cluster.
+            # For reporting, just use threshold (we know all pairs >= threshold
+            # transitively; min may be lower). Keep it simple.
+            sim = args.similarity_threshold
+            print(f"  cluster of {len(cluster)}: {[s['id'][:8] for s in sources]}")
+
+            verdict = draft_merge(sources, args.model, args.timeout_s)
+            if verdict is None:
+                continue
+            if verdict["action"] == "reject":
+                llm_rejections += 1
+                print(f"    LLM rejected: {verdict['reason'][:100]}")
+                continue
+
+            cid = submit_candidate(base, ids, sim, verdict, args.dry_run)
+            if cid == "dry-run":
+                candidates_created += 1
+            elif cid:
+                candidates_created += 1
+                print(f"    → candidate {cid[:8]}")
+            else:
+                skipped_existing += 1
+
+            time.sleep(0.3)  # be kind to claude CLI
+
+    print(
+        f"\nsummary: clusters_found={clusters_found} "
+        f"candidates_created={candidates_created} "
+        f"llm_rejections={llm_rejections} "
+        f"skipped_existing={skipped_existing}"
+    )
+
+
+if __name__ == "__main__":
+    main()

src/atocore/api/routes.py

@@ -1530,6 +1530,169 @@ def api_extend_reinforced() -> dict:
     return {"extended_count": len(extended), "extensions": extended}
 
 
+# --- Phase 7A: memory dedup / merge-candidate lifecycle ---
+
+
+class MergeCandidateCreateBody(BaseModel):
+    memory_ids: list[str]
+    similarity: float = 0.0
+    proposed_content: str
+    proposed_memory_type: str = "knowledge"
+    proposed_project: str = ""
+    proposed_tags: list[str] = []
+    proposed_confidence: float = 0.6
+    reason: str = ""
+
+
+class MergeCandidateApproveBody(BaseModel):
+    actor: str = "human-triage"
+    content: str | None = None
+    domain_tags: list[str] | None = None
+
+
+class MergeCandidateRejectBody(BaseModel):
+    actor: str = "human-triage"
+    note: str = ""
+
+
+class DedupScanRequestBody(BaseModel):
+    project: str = ""
+    similarity_threshold: float = 0.88
+    max_batch: int = 50
+
+
+@router.get("/admin/memory/merge-candidates")
+def api_list_merge_candidates(status: str = "pending", limit: int = 100) -> dict:
+    """Phase 7A: list merge-candidate proposals for triage UI."""
+    from atocore.memory.service import get_merge_candidates
+    cands = get_merge_candidates(status=status, limit=limit)
+    return {"candidates": cands, "count": len(cands)}
+
+
+@router.post("/admin/memory/merge-candidates/create")
+def api_create_merge_candidate(body: MergeCandidateCreateBody) -> dict:
+    """Phase 7A: host-side dedup detector submits a proposal here.
+
+    Server-side idempotency: if a pending candidate already exists for
+    the same sorted memory_id set, returns the existing id.
+    """
+    from atocore.memory.service import create_merge_candidate
+    try:
+        cid = create_merge_candidate(
+            memory_ids=body.memory_ids,
+            similarity=body.similarity,
+            proposed_content=body.proposed_content,
+            proposed_memory_type=body.proposed_memory_type,
+            proposed_project=body.proposed_project,
+            proposed_tags=body.proposed_tags,
+            proposed_confidence=body.proposed_confidence,
+            reason=body.reason,
+        )
+    except ValueError as e:
+        raise HTTPException(status_code=400, detail=str(e))
+    if cid is None:
+        return {"candidate_id": None, "duplicate": True}
+    return {"candidate_id": cid, "duplicate": False}
+
+
+@router.post("/admin/memory/merge-candidates/{candidate_id}/approve")
+def api_approve_merge_candidate(candidate_id: str, body: MergeCandidateApproveBody) -> dict:
+    """Phase 7A: execute an approved merge. Sources → superseded; new
+    merged memory created. UI can pass content/tag edits via body."""
+    from atocore.memory.service import merge_memories
+    new_id = merge_memories(
+        candidate_id=candidate_id,
+        actor=body.actor,
+        override_content=body.content,
+        override_tags=body.domain_tags,
+    )
+    if new_id is None:
+        raise HTTPException(
+            status_code=409,
+            detail="Merge could not execute (candidate not pending, or source memory tampered)",
+        )
+    return {"status": "approved", "candidate_id": candidate_id, "result_memory_id": new_id}
+
+
+@router.post("/admin/memory/merge-candidates/{candidate_id}/reject")
+def api_reject_merge_candidate(candidate_id: str, body: MergeCandidateRejectBody) -> dict:
+    """Phase 7A: dismiss a merge candidate. Sources stay untouched."""
+    from atocore.memory.service import reject_merge_candidate
+    ok = reject_merge_candidate(candidate_id, actor=body.actor, note=body.note)
+    if not ok:
+        raise HTTPException(status_code=404, detail="Candidate not found or already resolved")
+    return {"status": "rejected", "candidate_id": candidate_id}
+
+
+@router.post("/admin/memory/dedup-scan")
+def api_request_dedup_scan(body: DedupScanRequestBody) -> dict:
+    """Phase 7A: request a host-side dedup scan.
+
+    Writes a flag in project_state with project + threshold + max_batch.
+    A host cron watcher picks it up within ~2 min and runs
+    scripts/memory_dedup.py. Mirrors /admin/graduation/request.
+    """
+    import json as _json
+    from datetime import datetime as _dt, timezone as _tz
+    from atocore.context.project_state import set_state
+
+    now = _dt.now(_tz.utc).strftime("%Y-%m-%dT%H:%M:%SZ")
+    payload = _json.dumps({
+        "project": (body.project or "").strip(),
+        "similarity_threshold": max(0.5, min(0.99, body.similarity_threshold)),
+        "max_batch": max(1, min(body.max_batch, 200)),
+        "requested_at": now,
+    })
+    set_state(
+        project_name="atocore",
+        category="config",
+        key="dedup_requested_at",
+        value=payload,
+        source="admin ui",
+    )
+    return {
+        "requested_at": now,
+        "project": body.project,
+        "similarity_threshold": body.similarity_threshold,
+        "max_batch": body.max_batch,
+        "note": "Host watcher picks up within ~2 min. Poll /admin/memory/dedup-status for progress.",
+    }
+
+
+@router.get("/admin/memory/dedup-status")
+def api_dedup_status() -> dict:
+    """Phase 7A: state of the dedup scan pipeline (UI polling)."""
+    import json as _json
+    from atocore.context.project_state import get_state
+    out = {
+        "requested": None,
+        "last_started_at": None,
+        "last_finished_at": None,
+        "last_result": None,
+        "is_running": False,
+    }
+    try:
+        for e in get_state("atocore"):
+            if e.category not in ("config", "status"):
+                continue
+            if e.key == "dedup_requested_at":
+                try:
+                    out["requested"] = _json.loads(e.value)
+                except Exception:
+                    out["requested"] = {"raw": e.value}
+            elif e.key == "dedup_last_started_at":
+                out["last_started_at"] = e.value
+            elif e.key == "dedup_last_finished_at":
+                out["last_finished_at"] = e.value
+            elif e.key == "dedup_last_result":
+                out["last_result"] = e.value
+            elif e.key == "dedup_running":
+                out["is_running"] = (e.value == "1")
+    except Exception:
+        pass
+    return out
+
+
 @router.get("/admin/graduation/stats")
 def api_graduation_stats() -> dict:
     """Phase 5F graduation stats for dashboard."""

src/atocore/engineering/triage_ui.py

@@ -377,6 +377,177 @@ _ENTITY_TRIAGE_CSS = """
 """
 
 
+# ---------------------------------------------------------------------
+# Phase 7A — Merge candidates (semantic dedup)
+# ---------------------------------------------------------------------
+
+_MERGE_TRIAGE_CSS = """
+<style>
+  .cand-merge { border-left: 3px solid #8b5cf6; }
+  .merge-type { background: #8b5cf6; color: white; padding: 0.1rem 0.5rem; border-radius: 3px; font-size: 0.75rem; }
+  .merge-sources { margin: 0.5rem 0 0.8rem 0; display: flex; flex-direction: column; gap: 0.35rem; }
+  .merge-source { background: var(--bg); border: 1px dashed var(--border); border-radius: 4px; padding: 0.4rem 0.6rem; font-size: 0.85rem; }
+  .merge-source-meta { font-family: monospace; font-size: 0.72rem; opacity: 0.7; margin-bottom: 0.2rem; }
+  .merge-arrow { text-align: center; font-size: 1.1rem; opacity: 0.5; margin: 0.3rem 0; }
+  .merge-proposed { background: var(--card); border: 1px solid #8b5cf6; border-radius: 4px; padding: 0.5rem; }
+  .btn-merge-approve { background: #8b5cf6; color: white; border-color: #8b5cf6; }
+  .btn-merge-approve:hover { background: #7c3aed; }
+</style>
+"""
+
+
+def _render_merge_card(cand: dict) -> str:
+    import json as _json
+    cid = _escape(cand.get("id", ""))
+    sim = cand.get("similarity") or 0.0
+    sources = cand.get("sources") or []
+    proposed_content = cand.get("proposed_content") or ""
+    proposed_tags = cand.get("proposed_tags") or []
+    proposed_project = cand.get("proposed_project") or ""
+    reason = cand.get("reason") or ""
+
+    src_html = "".join(
+        f"""
+        <div class="merge-source">
+          <div class="merge-source-meta">
+            {_escape(s.get('id','')[:8])} · [{_escape(s.get('memory_type',''))}]
+            · {_escape(s.get('project','') or '(global)')}
+            · conf {float(s.get('confidence',0)):.2f}
+            · refs {int(s.get('reference_count',0))}
+          </div>
+          <div>{_escape((s.get('content') or '')[:300])}</div>
+        </div>
+        """
+        for s in sources
+    )
+    tags_str = ", ".join(proposed_tags)
+    return f"""
+<div class="cand cand-merge" id="mcand-{cid}" data-merge-id="{cid}">
+  <div class="cand-head">
+    <span class="cand-type merge-type">[merge · {len(sources)} sources]</span>
+    <span class="cand-project">{_escape(proposed_project or '(global)')}</span>
+    <span class="cand-meta">sim ≥ {sim:.2f}</span>
+  </div>
+  <div class="merge-sources">{src_html}</div>
+  <div class="merge-arrow">↓ merged into ↓</div>
+  <div class="merge-proposed">
+    <textarea class="cand-content" id="mcontent-{cid}">{_escape(proposed_content)}</textarea>
+    <div class="cand-meta-row">
+      <label class="cand-field-label">Tags:
+        <input type="text" class="cand-tags-input" id="mtags-{cid}" value="{_escape(tags_str)}" placeholder="tag1, tag2">
+      </label>
+    </div>
+    {f'<div class="auto-triage-msg" style="margin-top:0.4rem;">💡 {_escape(reason)}</div>' if reason else ''}
+  </div>
+  <div class="cand-actions">
+    <button class="btn-merge-approve" data-merge-id="{cid}" title="Approve merge">✅ Approve Merge</button>
+    <button class="btn-reject" data-merge-id="{cid}" data-merge-reject="1" title="Keep separate">❌ Keep Separate</button>
+  </div>
+  <div class="cand-status" id="mstatus-{cid}"></div>
+</div>
+"""
+
+
+_MERGE_TRIAGE_SCRIPT = """
+<script>
+async function mergeApprove(id) {
+  const st = document.getElementById('mstatus-' + id);
+  st.textContent = 'Merging…';
+  st.className = 'cand-status ok';
+  const content = document.getElementById('mcontent-' + id).value;
+  const tagsRaw = document.getElementById('mtags-' + id).value;
+  const tags = tagsRaw.split(',').map(t => t.trim()).filter(Boolean);
+  const r = await fetch('/admin/memory/merge-candidates/' + encodeURIComponent(id) + '/approve', {
+    method: 'POST',
+    headers: {'Content-Type': 'application/json'},
+    body: JSON.stringify({actor: 'human-triage', content: content, domain_tags: tags}),
+  });
+  if (r.ok) {
+    const data = await r.json();
+    st.textContent = '✅ Merged → ' + (data.result_memory_id || '').slice(0, 8);
+    setTimeout(() => {
+      const card = document.getElementById('mcand-' + id);
+      if (card) { card.style.opacity = '0'; setTimeout(() => card.remove(), 300); }
+    }, 600);
+  } else {
+    const err = await r.text();
+    st.textContent = '❌ ' + r.status + ': ' + err.slice(0, 120);
+    st.className = 'cand-status err';
+  }
+}
+
+async function mergeReject(id) {
+  const st = document.getElementById('mstatus-' + id);
+  st.textContent = 'Rejecting…';
+  st.className = 'cand-status ok';
+  const r = await fetch('/admin/memory/merge-candidates/' + encodeURIComponent(id) + '/reject', {
+    method: 'POST',
+    headers: {'Content-Type': 'application/json'},
+    body: JSON.stringify({actor: 'human-triage'}),
+  });
+  if (r.ok) {
+    st.textContent = '❌ Kept separate';
+    setTimeout(() => {
+      const card = document.getElementById('mcand-' + id);
+      if (card) { card.style.opacity = '0'; setTimeout(() => card.remove(), 300); }
+    }, 400);
+  } else st.textContent = '❌ ' + r.status;
+}
+
+document.addEventListener('click', (e) => {
+  const mid = e.target.dataset?.mergeId;
+  if (!mid) return;
+  if (e.target.classList.contains('btn-merge-approve')) mergeApprove(mid);
+  else if (e.target.dataset?.mergeReject) mergeReject(mid);
+});
+
+async function requestDedupScan() {
+  const btn = document.getElementById('dedup-btn');
+  const status = document.getElementById('dedup-status');
+  btn.disabled = true;
+  btn.textContent = 'Queuing…';
+  status.textContent = '';
+  status.className = 'auto-triage-msg';
+  const threshold = parseFloat(document.getElementById('dedup-threshold').value || '0.88');
+  const r = await fetch('/admin/memory/dedup-scan', {
+    method: 'POST',
+    headers: {'Content-Type': 'application/json'},
+    body: JSON.stringify({project: '', similarity_threshold: threshold, max_batch: 50}),
+  });
+  if (r.ok) {
+    status.textContent = `✓ Queued dedup scan at threshold ${threshold}. Host watcher runs every 2 min; refresh in ~3 min to see merge candidates.`;
+    status.className = 'auto-triage-msg ok';
+  } else {
+    status.textContent = '✗ ' + r.status;
+    status.className = 'auto-triage-msg err';
+  }
+  setTimeout(() => {
+    btn.disabled = false;
+    btn.textContent = '🔗 Scan for duplicates';
+  }, 2000);
+}
+</script>
+"""
+
+
+def _render_dedup_bar() -> str:
+    return """
+<div class="auto-triage-bar">
+  <button id="dedup-btn" onclick="requestDedupScan()" title="Run semantic dedup scan on Dalidou host">
+    🔗 Scan for duplicates
+  </button>
+  <label class="cand-field-label" style="margin:0 0.5rem;">
+    Threshold:
+    <input id="dedup-threshold" type="number" min="0.70" max="0.99" step="0.01" value="0.88"
+           style="width:70px; padding:0.25rem; background:var(--bg); color:var(--text); border:1px solid var(--border); border-radius:3px;">
+  </label>
+  <span id="dedup-status" class="auto-triage-msg">
+    Finds semantically near-duplicate active memories and proposes LLM-drafted merges for review. Source memories become <code>superseded</code> on approve; nothing is deleted.
+  </span>
+</div>
+"""
+
+
 def _render_graduation_bar() -> str:
     """The 'Graduate memories → entity candidates' control bar."""
     from atocore.projects.registry import load_project_registry
@@ -478,26 +649,51 @@ def render_triage_page(limit: int = 100) -> str:
     except Exception as e:
         entity_candidates = []
 
-    total = len(mem_candidates) + len(entity_candidates)
+    try:
+        from atocore.memory.service import get_merge_candidates
+        merge_candidates = get_merge_candidates(status="pending", limit=limit)
+    except Exception:
+        merge_candidates = []
+
+    total = len(mem_candidates) + len(entity_candidates) + len(merge_candidates)
     graduation_bar = _render_graduation_bar()
+    dedup_bar = _render_dedup_bar()
 
     if total == 0:
-        body = _TRIAGE_CSS + _ENTITY_TRIAGE_CSS + f"""
+        body = _TRIAGE_CSS + _ENTITY_TRIAGE_CSS + _MERGE_TRIAGE_CSS + f"""
           <div class="triage-header">
             <h1>Triage Queue</h1>
           </div>
           {graduation_bar}
+          {dedup_bar}
           <div class="empty">
             <p>🎉 No candidates to review.</p>
             <p>The auto-triage pipeline keeps this queue empty unless something needs your judgment.</p>
-            <p>Use the 🎓 Graduate memories button above to propose new entity candidates from existing memories.</p>
+            <p>Use 🎓 Graduate memories to propose entity candidates, or 🔗 Scan for duplicates to find near-duplicate memories to merge.</p>
           </div>
-        """ + _GRADUATION_SCRIPT
+        """ + _GRADUATION_SCRIPT + _MERGE_TRIAGE_SCRIPT
         return render_html("Triage — AtoCore", body, breadcrumbs=[("Wiki", "/wiki"), ("Triage", "")])
 
     # Memory cards
     mem_cards = "".join(_render_candidate_card(c) for c in mem_candidates)
 
+    # Merge cards (Phase 7A)
+    merge_cards_html = ""
+    if merge_candidates:
+        merge_cards = "".join(_render_merge_card(c) for c in merge_candidates)
+        merge_cards_html = f"""
+        <div class="section-break">
+          <h2>🔗 Merge Candidates ({len(merge_candidates)})</h2>
+          <p class="auto-triage-msg">
+            Semantically near-duplicate active memories. Approving merges the sources
+            into the proposed unified memory; sources become <code>superseded</code>
+            (not deleted — still queryable). You can edit the draft content and tags
+            before approving.
+          </p>
+        </div>
+        {merge_cards}
+        """
+
     # Entity cards
     ent_cards_html = ""
     if entity_candidates:
@@ -513,11 +709,12 @@ def render_triage_page(limit: int = 100) -> str:
         {ent_cards}
         """
 
-    body = _TRIAGE_CSS + _ENTITY_TRIAGE_CSS + f"""
+    body = _TRIAGE_CSS + _ENTITY_TRIAGE_CSS + _MERGE_TRIAGE_CSS + f"""
       <div class="triage-header">
         <h1>Triage Queue</h1>
         <span class="count">
           <span id="cand-count">{len(mem_candidates)}</span> memory ·
+          {len(merge_candidates)} merge ·
           {len(entity_candidates)} entity
         </span>
       </div>
@@ -536,10 +733,12 @@ def render_triage_page(limit: int = 100) -> str:
         </span>
       </div>
       {graduation_bar}
+      {dedup_bar}
       <h2>📝 Memory Candidates ({len(mem_candidates)})</h2>
       {mem_cards}
+      {merge_cards_html}
       {ent_cards_html}
-    """ + _TRIAGE_SCRIPT + _ENTITY_TRIAGE_SCRIPT + _GRADUATION_SCRIPT
+    """ + _TRIAGE_SCRIPT + _ENTITY_TRIAGE_SCRIPT + _GRADUATION_SCRIPT + _MERGE_TRIAGE_SCRIPT
 
     return render_html(
         "Triage — AtoCore",

src/atocore/memory/_dedup_prompt.py

@@ -0,0 +1,156 @@
+"""Shared LLM prompt + parser for memory dedup (Phase 7A).
+
+Stdlib-only — must be importable from both the in-container service
+layer (when a user clicks "scan for duplicates" in the UI) and the
+host-side batch script (``scripts/memory_dedup.py``), which runs on
+Dalidou where the container's Python deps are not available.
+
+The prompt instructs the model to draft a UNIFIED memory that
+preserves every specific detail from the sources. We never want a
+merge to lose information — if two memories disagree on a number, the
+merged content should surface both with context.
+"""
+
+from __future__ import annotations
+
+import json
+from typing import Any
+
+DEDUP_PROMPT_VERSION = "dedup-0.1.0"
+MAX_CONTENT_CHARS = 1000
+MAX_SOURCES = 8  # cluster size cap — bigger clusters are suspicious
+
+SYSTEM_PROMPT = """You consolidate near-duplicate memories for AtoCore, a personal context engine.
+
+Given 2-8 memories that a semantic-similarity scan flagged as likely duplicates, draft a UNIFIED replacement that preserves every specific detail from every source.
+
+CORE PRINCIPLE: information never gets lost. If the sources disagree on a number, date, vendor, or spec, surface BOTH with attribution (e.g., "quoted at $3.2k on 2026-03-01, revised to $3.8k on 2026-04-10"). If one source is more specific than another, keep the specificity. If they say the same thing differently, pick the clearer wording.
+
+YOU MUST:
+- Produce content under 500 characters that reads as a single coherent statement
+- Keep all project/vendor/person/part names that appear in any source
+- Keep all numbers, dates, and identifiers
+- Keep the strongest claim wording ("ratified", "decided", "committed") if any source has it
+- Propose domain_tags as a UNION of the sources' tags (lowercase, deduped, cap 6)
+- Return valid_until = latest non-null valid_until across sources, or null if any source has null (permanent beats transient)
+
+REFUSE TO MERGE (return action="reject") if:
+- The memories are actually about DIFFERENT subjects that just share vocabulary (e.g., "p04 mirror" and "p05 mirror" — same project bucket means same project, but different components)
+- One memory CONTRADICTS another and you cannot reconcile them — flag for contradiction review instead
+- The sources span different time snapshots of a changing state that should stay as a timeline, not be collapsed
+
+OUTPUT — raw JSON, no prose, no markdown fences:
+{
+  "action": "merge" | "reject",
+  "content": "the unified memory content",
+  "memory_type": "knowledge|project|preference|adaptation|episodic|identity",
+  "project": "project-slug or empty",
+  "domain_tags": ["tag1", "tag2"],
+  "confidence": 0.5,
+  "reason": "one sentence explaining the merge (or the rejection)"
+}
+
+On action=reject, still fill content with a short explanation and set confidence=0."""
+
+
+def build_user_message(sources: list[dict[str, Any]]) -> str:
+    """Format N source memories for the model to consolidate.
+
+    Each source dict should carry id, content, project, memory_type,
+    domain_tags, confidence, valid_until, reference_count.
+    """
+    lines = [f"You have {len(sources)} source memories in the same (project, memory_type) bucket:\n"]
+    for i, src in enumerate(sources[:MAX_SOURCES], start=1):
+        tags = src.get("domain_tags") or []
+        if isinstance(tags, str):
+            try:
+                tags = json.loads(tags)
+            except Exception:
+                tags = []
+        lines.append(
+            f"--- Source {i} (id={src.get('id','?')[:8]}, "
+            f"refs={src.get('reference_count',0)}, "
+            f"conf={src.get('confidence',0):.2f}, "
+            f"valid_until={src.get('valid_until') or 'permanent'}) ---"
+        )
+        lines.append(f"project: {src.get('project','')}")
+        lines.append(f"type: {src.get('memory_type','')}")
+        lines.append(f"tags: {tags}")
+        lines.append(f"content: {(src.get('content') or '')[:MAX_CONTENT_CHARS]}")
+        lines.append("")
+    lines.append("Return the JSON object now.")
+    return "\n".join(lines)
+
+
+def parse_merge_verdict(raw_output: str) -> dict[str, Any] | None:
+    """Strip markdown fences / leading prose and return the parsed JSON
+    object. Returns None on parse failure."""
+    text = (raw_output or "").strip()
+    if text.startswith("```"):
+        text = text.strip("`")
+        nl = text.find("\n")
+        if nl >= 0:
+            text = text[nl + 1:]
+        if text.endswith("```"):
+            text = text[:-3]
+        text = text.strip()
+
+    if not text.lstrip().startswith("{"):
+        start = text.find("{")
+        end = text.rfind("}")
+        if start >= 0 and end > start:
+            text = text[start:end + 1]
+
+    try:
+        parsed = json.loads(text)
+    except json.JSONDecodeError:
+        return None
+    if not isinstance(parsed, dict):
+        return None
+    return parsed
+
+
+def normalize_merge_verdict(verdict: dict[str, Any]) -> dict[str, Any] | None:
+    """Validate + normalize a raw merge verdict. Returns None if the
+    verdict is unusable (no content, unknown action)."""
+    action = str(verdict.get("action") or "").strip().lower()
+    if action not in ("merge", "reject"):
+        return None
+
+    content = str(verdict.get("content") or "").strip()
+    if not content:
+        return None
+
+    memory_type = str(verdict.get("memory_type") or "knowledge").strip().lower()
+    project = str(verdict.get("project") or "").strip()
+
+    raw_tags = verdict.get("domain_tags") or []
+    if isinstance(raw_tags, str):
+        raw_tags = [t.strip() for t in raw_tags.split(",") if t.strip()]
+    if not isinstance(raw_tags, list):
+        raw_tags = []
+    tags: list[str] = []
+    for t in raw_tags[:6]:
+        if not isinstance(t, str):
+            continue
+        tt = t.strip().lower()
+        if tt and tt not in tags:
+            tags.append(tt)
+
+    try:
+        confidence = float(verdict.get("confidence", 0.5))
+    except (TypeError, ValueError):
+        confidence = 0.5
+    confidence = max(0.0, min(1.0, confidence))
+
+    reason = str(verdict.get("reason") or "").strip()[:500]
+
+    return {
+        "action": action,
+        "content": content[:1000],
+        "memory_type": memory_type,
+        "project": project,
+        "domain_tags": tags,
+        "confidence": confidence,
+        "reason": reason,
+    }

src/atocore/memory/service.py

@@ -925,3 +925,327 @@ def _row_to_memory(row) -> Memory:
 def _validate_confidence(confidence: float) -> None:
     if not 0.0 <= confidence <= 1.0:
         raise ValueError("Confidence must be between 0.0 and 1.0")
+
+
+# ---------------------------------------------------------------------
+# Phase 7A — Memory Consolidation: merge-candidate lifecycle
+# ---------------------------------------------------------------------
+#
+# The detector (scripts/memory_dedup.py) writes proposals into
+# memory_merge_candidates. The triage UI lists pending rows, a human
+# reviews, and on approve we execute the merge here — never at detect
+# time. This keeps the audit trail clean: every mutation is a human
+# decision.
+
+
+def create_merge_candidate(
+    memory_ids: list[str],
+    similarity: float,
+    proposed_content: str,
+    proposed_memory_type: str,
+    proposed_project: str,
+    proposed_tags: list[str] | None = None,
+    proposed_confidence: float = 0.6,
+    reason: str = "",
+) -> str | None:
+    """Insert a merge-candidate row. Returns the new row id, or None if
+    a pending candidate already covers this exact set of memory ids
+    (idempotent scan — re-running the detector doesn't double-create)."""
+    import json as _json
+
+    if not memory_ids or len(memory_ids) < 2:
+        raise ValueError("merge candidate requires at least 2 memory_ids")
+
+    memory_ids_sorted = sorted(set(memory_ids))
+    memory_ids_json = _json.dumps(memory_ids_sorted)
+    tags_json = _json.dumps(_normalize_tags(proposed_tags))
+    candidate_id = str(uuid.uuid4())
+
+    with get_connection() as conn:
+        # Idempotency: same sorted-id set already pending? skip.
+        existing = conn.execute(
+            "SELECT id FROM memory_merge_candidates "
+            "WHERE status = 'pending' AND memory_ids = ?",
+            (memory_ids_json,),
+        ).fetchone()
+        if existing:
+            return None
+
+        conn.execute(
+            "INSERT INTO memory_merge_candidates "
+            "(id, status, memory_ids, similarity, proposed_content, "
+            "proposed_memory_type, proposed_project, proposed_tags, "
+            "proposed_confidence, reason) "
+            "VALUES (?, 'pending', ?, ?, ?, ?, ?, ?, ?, ?)",
+            (
+                candidate_id, memory_ids_json, float(similarity or 0.0),
+                (proposed_content or "")[:2000],
+                (proposed_memory_type or "knowledge")[:50],
+                (proposed_project or "")[:100],
+                tags_json,
+                max(0.0, min(1.0, float(proposed_confidence))),
+                (reason or "")[:500],
+            ),
+        )
+    log.info(
+        "merge_candidate_created",
+        candidate_id=candidate_id,
+        memory_count=len(memory_ids_sorted),
+        similarity=round(similarity, 4),
+    )
+    return candidate_id
+
+
+def get_merge_candidates(status: str = "pending", limit: int = 100) -> list[dict]:
+    """List merge candidates with their source memories inlined."""
+    import json as _json
+
+    with get_connection() as conn:
+        rows = conn.execute(
+            "SELECT * FROM memory_merge_candidates "
+            "WHERE status = ? ORDER BY created_at DESC LIMIT ?",
+            (status, limit),
+        ).fetchall()
+
+        out = []
+        for r in rows:
+            try:
+                mem_ids = _json.loads(r["memory_ids"] or "[]")
+            except Exception:
+                mem_ids = []
+            try:
+                tags = _json.loads(r["proposed_tags"] or "[]")
+            except Exception:
+                tags = []
+
+            sources = []
+            for mid in mem_ids:
+                srow = conn.execute(
+                    "SELECT id, memory_type, content, project, confidence, "
+                    "status, reference_count, domain_tags, valid_until "
+                    "FROM memories WHERE id = ?",
+                    (mid,),
+                ).fetchone()
+                if srow:
+                    try:
+                        stags = _json.loads(srow["domain_tags"] or "[]")
+                    except Exception:
+                        stags = []
+                    sources.append({
+                        "id": srow["id"],
+                        "memory_type": srow["memory_type"],
+                        "content": srow["content"],
+                        "project": srow["project"] or "",
+                        "confidence": srow["confidence"],
+                        "status": srow["status"],
+                        "reference_count": int(srow["reference_count"] or 0),
+                        "domain_tags": stags,
+                        "valid_until": srow["valid_until"] or "",
+                    })
+
+            out.append({
+                "id": r["id"],
+                "status": r["status"],
+                "memory_ids": mem_ids,
+                "similarity": r["similarity"],
+                "proposed_content": r["proposed_content"] or "",
+                "proposed_memory_type": r["proposed_memory_type"] or "knowledge",
+                "proposed_project": r["proposed_project"] or "",
+                "proposed_tags": tags,
+                "proposed_confidence": r["proposed_confidence"],
+                "reason": r["reason"] or "",
+                "created_at": r["created_at"],
+                "resolved_at": r["resolved_at"],
+                "resolved_by": r["resolved_by"],
+                "result_memory_id": r["result_memory_id"],
+                "sources": sources,
+            })
+        return out
+
+
+def reject_merge_candidate(candidate_id: str, actor: str = "human-triage", note: str = "") -> bool:
+    """Mark a merge candidate as rejected. Source memories stay untouched."""
+    now_str = datetime.now(timezone.utc).strftime("%Y-%m-%d %H:%M:%S")
+    with get_connection() as conn:
+        result = conn.execute(
+            "UPDATE memory_merge_candidates "
+            "SET status = 'rejected', resolved_at = ?, resolved_by = ? "
+            "WHERE id = ? AND status = 'pending'",
+            (now_str, actor, candidate_id),
+        )
+        if result.rowcount == 0:
+            return False
+    log.info("merge_candidate_rejected", candidate_id=candidate_id, actor=actor, note=note[:100])
+    return True
+
+
+def merge_memories(
+    candidate_id: str,
+    actor: str = "human-triage",
+    override_content: str | None = None,
+    override_tags: list[str] | None = None,
+) -> str | None:
+    """Execute an approved merge candidate.
+
+    1. Validate all source memories still status=active
+    2. Create the new merged memory (status=active)
+    3. Mark each source status=superseded with an audit row pointing at
+       the new merged id
+    4. Mark the candidate status=approved, record result_memory_id
+    5. Write a consolidated audit row on the new memory
+
+    Returns the new merged memory's id, or None if the candidate cannot
+    be executed (already resolved, source tampered, etc.).
+
+    ``override_content`` and ``override_tags`` let the UI pass the human's
+    edits before clicking approve.
+    """
+    import json as _json
+
+    now_str = datetime.now(timezone.utc).strftime("%Y-%m-%d %H:%M:%S")
+
+    with get_connection() as conn:
+        row = conn.execute(
+            "SELECT * FROM memory_merge_candidates WHERE id = ?",
+            (candidate_id,),
+        ).fetchone()
+        if row is None or row["status"] != "pending":
+            log.warning("merge_candidate_not_pending", candidate_id=candidate_id)
+            return None
+
+        try:
+            mem_ids = _json.loads(row["memory_ids"] or "[]")
+        except Exception:
+            mem_ids = []
+        if not mem_ids or len(mem_ids) < 2:
+            log.warning("merge_candidate_invalid_memory_ids", candidate_id=candidate_id)
+            return None
+
+        # Snapshot sources + validate all active
+        source_rows = []
+        for mid in mem_ids:
+            srow = conn.execute(
+                "SELECT * FROM memories WHERE id = ?", (mid,)
+            ).fetchone()
+            if srow is None or srow["status"] != "active":
+                log.warning(
+                    "merge_source_not_active",
+                    candidate_id=candidate_id,
+                    memory_id=mid,
+                    actual_status=(srow["status"] if srow else "missing"),
+                )
+                return None
+            source_rows.append(srow)
+
+        # Build merged memory fields — prefer human overrides, then proposed
+        content = (override_content or row["proposed_content"] or "").strip()
+        if not content:
+            log.warning("merge_candidate_empty_content", candidate_id=candidate_id)
+            return None
+
+        merged_type = (row["proposed_memory_type"] or source_rows[0]["memory_type"]).lower()
+        if merged_type not in MEMORY_TYPES:
+            merged_type = source_rows[0]["memory_type"]
+
+        merged_project = row["proposed_project"] or source_rows[0]["project"] or ""
+        merged_project = resolve_project_name(merged_project)
+
+        # Tags: override wins, else proposed, else union of sources
+        if override_tags is not None:
+            merged_tags = _normalize_tags(override_tags)
+        else:
+            try:
+                proposed_tags = _json.loads(row["proposed_tags"] or "[]")
+            except Exception:
+                proposed_tags = []
+            if proposed_tags:
+                merged_tags = _normalize_tags(proposed_tags)
+            else:
+                union: list[str] = []
+                for srow in source_rows:
+                    try:
+                        stags = _json.loads(srow["domain_tags"] or "[]")
+                    except Exception:
+                        stags = []
+                    for t in stags:
+                        if isinstance(t, str) and t and t not in union:
+                            union.append(t)
+                merged_tags = union
+
+        # confidence = max; reference_count = sum
+        merged_confidence = max(float(s["confidence"]) for s in source_rows)
+        total_refs = sum(int(s["reference_count"] or 0) for s in source_rows)
+
+        # valid_until: if any source is permanent (None/empty), merged is permanent.
+        # Otherwise take the latest (lexical compare on ISO dates works).
+        merged_vu: str | None = ""  # placeholder
+        has_permanent = any(not (s["valid_until"] or "").strip() for s in source_rows)
+        if has_permanent:
+            merged_vu = None
+        else:
+            merged_vu = max((s["valid_until"] or "").strip() for s in source_rows) or None
+
+        new_id = str(uuid.uuid4())
+        tags_json = _json.dumps(merged_tags)
+
+        conn.execute(
+            "INSERT INTO memories (id, memory_type, content, project, "
+            "source_chunk_id, confidence, status, domain_tags, valid_until, "
+            "reference_count, last_referenced_at) "
+            "VALUES (?, ?, ?, ?, NULL, ?, 'active', ?, ?, ?, ?)",
+            (
+                new_id, merged_type, content[:2000], merged_project,
+                merged_confidence, tags_json, merged_vu, total_refs, now_str,
+            ),
+        )
+
+        # Mark sources superseded
+        for srow in source_rows:
+            conn.execute(
+                "UPDATE memories SET status = 'superseded', updated_at = ? "
+                "WHERE id = ?",
+                (now_str, srow["id"]),
+            )
+
+        # Mark candidate approved
+        conn.execute(
+            "UPDATE memory_merge_candidates SET status = 'approved', "
+            "resolved_at = ?, resolved_by = ?, result_memory_id = ? WHERE id = ?",
+            (now_str, actor, new_id, candidate_id),
+        )
+
+    # Audit rows (out of the transaction; fail-open via _audit_memory)
+    _audit_memory(
+        memory_id=new_id,
+        action="created_via_merge",
+        actor=actor,
+        after={
+            "memory_type": merged_type,
+            "content": content,
+            "project": merged_project,
+            "confidence": merged_confidence,
+            "domain_tags": merged_tags,
+            "reference_count": total_refs,
+            "merged_from": list(mem_ids),
+            "merge_candidate_id": candidate_id,
+        },
+        note=f"merged {len(mem_ids)} sources via candidate {candidate_id[:8]}",
+    )
+    for srow in source_rows:
+        _audit_memory(
+            memory_id=srow["id"],
+            action="superseded",
+            actor=actor,
+            before={"status": "active", "content": srow["content"]},
+            after={"status": "superseded", "superseded_by": new_id},
+            note=f"merged into {new_id}",
+        )
+
+    log.info(
+        "merge_executed",
+        candidate_id=candidate_id,
+        result_memory_id=new_id,
+        source_count=len(source_rows),
+        actor=actor,
+    )
+    return new_id

src/atocore/memory/similarity.py

@@ -0,0 +1,88 @@
+"""Phase 7A (Memory Consolidation): semantic similarity helpers.
+
+Thin wrapper over ``atocore.retrieval.embeddings`` that exposes
+pairwise + batch cosine similarity on normalized embeddings. Used by
+the dedup detector to cluster near-duplicate active memories.
+
+Embeddings from ``embed_texts()`` are already L2-normalized, so cosine
+similarity reduces to a dot product — no extra normalization needed.
+"""
+
+from __future__ import annotations
+
+from atocore.retrieval.embeddings import embed_texts
+
+
+def _dot(a: list[float], b: list[float]) -> float:
+    return sum(x * y for x, y in zip(a, b))
+
+
+def cosine(a: list[float], b: list[float]) -> float:
+    """Cosine similarity on already-normalized vectors. Clamped to [0,1]
+    (embeddings use paraphrase-multilingual-MiniLM which is unit-norm,
+    and we never want negative values leaking into thresholds)."""
+    return max(0.0, min(1.0, _dot(a, b)))
+
+
+def compute_memory_similarity(text_a: str, text_b: str) -> float:
+    """Return cosine similarity of two memory contents in [0,1].
+
+    Convenience helper for one-off checks + tests. For batch work (the
+    dedup detector), use ``embed_texts()`` directly and compute the
+    similarity matrix yourself to avoid re-embedding shared texts.
+    """
+    if not text_a or not text_b:
+        return 0.0
+    vecs = embed_texts([text_a, text_b])
+    return cosine(vecs[0], vecs[1])
+
+
+def similarity_matrix(texts: list[str]) -> list[list[float]]:
+    """N×N cosine similarity matrix. Diagonal is 1.0, symmetric."""
+    if not texts:
+        return []
+    vecs = embed_texts(texts)
+    n = len(vecs)
+    matrix = [[0.0] * n for _ in range(n)]
+    for i in range(n):
+        matrix[i][i] = 1.0
+        for j in range(i + 1, n):
+            s = cosine(vecs[i], vecs[j])
+            matrix[i][j] = s
+            matrix[j][i] = s
+    return matrix
+
+
+def cluster_by_threshold(texts: list[str], threshold: float) -> list[list[int]]:
+    """Greedy transitive clustering: if sim(i,j) >= threshold, merge.
+
+    Returns a list of clusters, each a list of indices into ``texts``.
+    Singletons are included. Used by the dedup detector to collapse
+    A~B~C into one merge proposal rather than three pair proposals.
+    """
+    if not texts:
+        return []
+    matrix = similarity_matrix(texts)
+    n = len(texts)
+    parent = list(range(n))
+
+    def find(x: int) -> int:
+        while parent[x] != x:
+            parent[x] = parent[parent[x]]
+            x = parent[x]
+        return x
+
+    def union(x: int, y: int) -> None:
+        rx, ry = find(x), find(y)
+        if rx != ry:
+            parent[rx] = ry
+
+    for i in range(n):
+        for j in range(i + 1, n):
+            if matrix[i][j] >= threshold:
+                union(i, j)
+
+    groups: dict[int, list[int]] = {}
+    for i in range(n):
+        groups.setdefault(find(i), []).append(i)
+    return list(groups.values())

src/atocore/models/database.py

@@ -251,6 +251,42 @@ def _apply_migrations(conn: sqlite3.Connection) -> None:
         "CREATE INDEX IF NOT EXISTS idx_interactions_created_at ON interactions(created_at)"
     )
 
+    # Phase 7A (Memory Consolidation — "sleep cycle"): merge candidates.
+    # When the dedup detector finds a cluster of semantically similar active
+    # memories within the same (project, memory_type) bucket, it drafts a
+    # unified content via LLM and writes a proposal here. The triage UI
+    # surfaces these for human approval. On approve, source memories become
+    # status=superseded and a new merged memory is created.
+    # memory_ids is a JSON array (length >= 2) of the source memory ids.
+    # proposed_* hold the LLM's draft; a human can edit before approve.
+    # result_memory_id is filled on approve with the new merged memory's id.
+    conn.execute(
+        """
+        CREATE TABLE IF NOT EXISTS memory_merge_candidates (
+            id TEXT PRIMARY KEY,
+            status TEXT DEFAULT 'pending',
+            memory_ids TEXT NOT NULL,
+            similarity REAL,
+            proposed_content TEXT,
+            proposed_memory_type TEXT,
+            proposed_project TEXT,
+            proposed_tags TEXT DEFAULT '[]',
+            proposed_confidence REAL,
+            reason TEXT DEFAULT '',
+            created_at DATETIME DEFAULT CURRENT_TIMESTAMP,
+            resolved_at DATETIME,
+            resolved_by TEXT,
+            result_memory_id TEXT
+        )
+        """
+    )
+    conn.execute(
+        "CREATE INDEX IF NOT EXISTS idx_mmc_status ON memory_merge_candidates(status)"
+    )
+    conn.execute(
+        "CREATE INDEX IF NOT EXISTS idx_mmc_created_at ON memory_merge_candidates(created_at)"
+    )
+
 
 def _column_exists(conn: sqlite3.Connection, table: str, column: str) -> bool:
     rows = conn.execute(f"PRAGMA table_info({table})").fetchall()

tests/test_memory_dedup.py

@@ -0,0 +1,380 @@
+"""Phase 7A — memory consolidation tests.
+
+Covers:
+  - similarity helpers (cosine bounds, matrix symmetry, clustering)
+  - _dedup_prompt parser / normalizer robustness
+  - create_merge_candidate idempotency
+  - get_merge_candidates inlines source memories
+  - merge_memories end-to-end happy path (sources → superseded,
+    new merged memory active, audit rows, result_memory_id)
+  - reject_merge_candidate leaves sources untouched
+"""
+
+from __future__ import annotations
+
+import pytest
+
+from atocore.memory._dedup_prompt import (
+    normalize_merge_verdict,
+    parse_merge_verdict,
+)
+from atocore.memory.service import (
+    create_memory,
+    create_merge_candidate,
+    get_memory_audit,
+    get_merge_candidates,
+    merge_memories,
+    reject_merge_candidate,
+)
+from atocore.memory.similarity import (
+    cluster_by_threshold,
+    cosine,
+    compute_memory_similarity,
+    similarity_matrix,
+)
+from atocore.models.database import get_connection, init_db
+
+
+# --- Similarity helpers ---
+
+
+def test_cosine_bounds():
+    assert cosine([1.0, 0.0], [1.0, 0.0]) == pytest.approx(1.0)
+    assert cosine([1.0, 0.0], [0.0, 1.0]) == pytest.approx(0.0)
+    # Negative dot product clamped to 0
+    assert cosine([1.0, 0.0], [-1.0, 0.0]) == 0.0
+
+
+def test_compute_memory_similarity_identical_high():
+    s = compute_memory_similarity("the sky is blue", "the sky is blue")
+    assert 0.99 <= s <= 1.0
+
+
+def test_compute_memory_similarity_unrelated_low():
+    s = compute_memory_similarity(
+        "APM integrates with NX via a Python bridge",
+        "the polisher firmware must use USB SSD not SD card",
+    )
+    assert 0.0 <= s < 0.7
+
+
+def test_similarity_matrix_symmetric():
+    texts = ["alpha beta gamma", "alpha beta gamma", "completely unrelated text"]
+    m = similarity_matrix(texts)
+    assert len(m) == 3 and all(len(r) == 3 for r in m)
+    for i in range(3):
+        assert m[i][i] == pytest.approx(1.0)
+    for i in range(3):
+        for j in range(3):
+            assert m[i][j] == pytest.approx(m[j][i])
+
+
+def test_cluster_by_threshold_transitive():
+    # Three near-paraphrases should land in one cluster
+    texts = [
+        "Antoine prefers OAuth over API keys",
+        "Antoine's preference is OAuth, not API keys",
+        "the polisher firmware uses USB SSD storage",
+    ]
+    clusters = cluster_by_threshold(texts, threshold=0.7)
+    # At least one cluster of size 2+ containing the paraphrases
+    big = [c for c in clusters if len(c) >= 2]
+    assert big, f"expected at least one multi-member cluster, got {clusters}"
+    assert 0 in big[0] and 1 in big[0]
+
+
+# --- Prompt parser robustness ---
+
+
+def test_parse_merge_verdict_strips_fences():
+    raw = "```json\n{\"action\":\"merge\",\"content\":\"x\"}\n```"
+    parsed = parse_merge_verdict(raw)
+    assert parsed == {"action": "merge", "content": "x"}
+
+
+def test_parse_merge_verdict_handles_prose_prefix():
+    raw = "Sure! Here's the result:\n{\"action\":\"reject\",\"content\":\"no\"}"
+    parsed = parse_merge_verdict(raw)
+    assert parsed is not None
+    assert parsed["action"] == "reject"
+
+
+def test_normalize_merge_verdict_fills_defaults():
+    v = normalize_merge_verdict({
+        "action": "merge",
+        "content": "unified text",
+    })
+    assert v is not None
+    assert v["memory_type"] == "knowledge"
+    assert v["project"] == ""
+    assert v["domain_tags"] == []
+    assert v["confidence"] == 0.5
+
+
+def test_normalize_merge_verdict_rejects_empty_content():
+    assert normalize_merge_verdict({"action": "merge", "content": ""}) is None
+
+
+def test_normalize_merge_verdict_rejects_unknown_action():
+    assert normalize_merge_verdict({"action": "?", "content": "x"}) is None
+
+
+# --- create_merge_candidate idempotency ---
+
+
+def test_create_merge_candidate_inserts_row(tmp_data_dir):
+    init_db()
+    m1 = create_memory("knowledge", "APM uses NX for DXF conversion")
+    m2 = create_memory("knowledge", "APM uses NX for DXF-to-STL")
+
+    cid = create_merge_candidate(
+        memory_ids=[m1.id, m2.id],
+        similarity=0.92,
+        proposed_content="APM uses NX for DXF→STL conversion",
+        proposed_memory_type="knowledge",
+        proposed_project="",
+        proposed_tags=["apm", "nx"],
+        proposed_confidence=0.6,
+        reason="near-paraphrase",
+    )
+    assert cid is not None
+
+    pending = get_merge_candidates(status="pending")
+    assert len(pending) == 1
+    assert pending[0]["id"] == cid
+    assert pending[0]["similarity"] == pytest.approx(0.92)
+    assert len(pending[0]["sources"]) == 2
+
+
+def test_create_merge_candidate_idempotent(tmp_data_dir):
+    init_db()
+    m1 = create_memory("knowledge", "Fact A")
+    m2 = create_memory("knowledge", "Fact A slightly reworded")
+
+    first = create_merge_candidate(
+        memory_ids=[m1.id, m2.id],
+        similarity=0.9,
+        proposed_content="merged",
+        proposed_memory_type="knowledge",
+        proposed_project="",
+    )
+    # Same id set, different order → dedupe skips
+    second = create_merge_candidate(
+        memory_ids=[m2.id, m1.id],
+        similarity=0.9,
+        proposed_content="merged (again)",
+        proposed_memory_type="knowledge",
+        proposed_project="",
+    )
+    assert first is not None
+    assert second is None
+
+
+def test_create_merge_candidate_requires_two_ids(tmp_data_dir):
+    init_db()
+    m1 = create_memory("knowledge", "lonely")
+    with pytest.raises(ValueError):
+        create_merge_candidate(
+            memory_ids=[m1.id],
+            similarity=1.0,
+            proposed_content="x",
+            proposed_memory_type="knowledge",
+            proposed_project="",
+        )
+
+
+# --- merge_memories end-to-end ---
+
+
+def test_merge_memories_happy_path(tmp_data_dir):
+    init_db()
+    m1 = create_memory(
+        "knowledge", "APM uses NX for DXF conversion",
+        project="apm", confidence=0.6, domain_tags=["apm", "nx"],
+    )
+    m2 = create_memory(
+        "knowledge", "APM does DXF to STL via NX bridge",
+        project="apm", confidence=0.8, domain_tags=["apm", "bridge"],
+    )
+    # Bump reference counts so sum is meaningful
+    with get_connection() as conn:
+        conn.execute("UPDATE memories SET reference_count = 3 WHERE id = ?", (m1.id,))
+        conn.execute("UPDATE memories SET reference_count = 5 WHERE id = ?", (m2.id,))
+
+    cid = create_merge_candidate(
+        memory_ids=[m1.id, m2.id],
+        similarity=0.92,
+        proposed_content="APM uses NX bridge for DXF→STL conversion",
+        proposed_memory_type="knowledge",
+        proposed_project="apm",
+        proposed_tags=["apm", "nx", "bridge"],
+        proposed_confidence=0.7,
+        reason="duplicates",
+    )
+    new_id = merge_memories(candidate_id=cid, actor="human-triage")
+    assert new_id is not None
+
+    # Sources superseded
+    with get_connection() as conn:
+        s1 = conn.execute("SELECT status FROM memories WHERE id = ?", (m1.id,)).fetchone()
+        s2 = conn.execute("SELECT status FROM memories WHERE id = ?", (m2.id,)).fetchone()
+        merged = conn.execute(
+            "SELECT content, status, confidence, reference_count, project "
+            "FROM memories WHERE id = ?", (new_id,)
+        ).fetchone()
+        cand = conn.execute(
+            "SELECT status, result_memory_id FROM memory_merge_candidates WHERE id = ?",
+            (cid,),
+        ).fetchone()
+    assert s1["status"] == "superseded"
+    assert s2["status"] == "superseded"
+    assert merged["status"] == "active"
+    assert merged["project"] == "apm"
+    # confidence = max of sources (0.8), not the proposed 0.7 (proposed is hint;
+    # merge_memories picks max of actual source confidences — verify).
+    assert merged["confidence"] == pytest.approx(0.8)
+    # reference_count = sum (3 + 5 = 8)
+    assert int(merged["reference_count"]) == 8
+    assert cand["status"] == "approved"
+    assert cand["result_memory_id"] == new_id
+
+
+def test_merge_memories_content_override(tmp_data_dir):
+    init_db()
+    m1 = create_memory("knowledge", "draft A", project="p05-interferometer")
+    m2 = create_memory("knowledge", "draft B", project="p05-interferometer")
+
+    cid = create_merge_candidate(
+        memory_ids=[m1.id, m2.id],
+        similarity=0.9,
+        proposed_content="AI draft",
+        proposed_memory_type="knowledge",
+        proposed_project="p05-interferometer",
+    )
+    new_id = merge_memories(
+        candidate_id=cid,
+        actor="human-triage",
+        override_content="human-edited final text",
+        override_tags=["optics", "custom"],
+    )
+    assert new_id is not None
+    with get_connection() as conn:
+        row = conn.execute(
+            "SELECT content, domain_tags FROM memories WHERE id = ?", (new_id,)
+        ).fetchone()
+    assert row["content"] == "human-edited final text"
+    # domain_tags JSON should contain the override
+    assert "optics" in row["domain_tags"]
+    assert "custom" in row["domain_tags"]
+
+
+def test_merge_memories_writes_audit(tmp_data_dir):
+    init_db()
+    m1 = create_memory("knowledge", "alpha")
+    m2 = create_memory("knowledge", "alpha variant")
+    cid = create_merge_candidate(
+        memory_ids=[m1.id, m2.id], similarity=0.9,
+        proposed_content="alpha merged",
+        proposed_memory_type="knowledge", proposed_project="",
+    )
+    new_id = merge_memories(candidate_id=cid)
+    assert new_id
+
+    audit_new = get_memory_audit(new_id)
+    actions_new = {a["action"] for a in audit_new}
+    assert "created_via_merge" in actions_new
+
+    audit_m1 = get_memory_audit(m1.id)
+    actions_m1 = {a["action"] for a in audit_m1}
+    assert "superseded" in actions_m1
+
+
+def test_merge_memories_aborts_if_source_not_active(tmp_data_dir):
+    init_db()
+    m1 = create_memory("knowledge", "one")
+    m2 = create_memory("knowledge", "two")
+    cid = create_merge_candidate(
+        memory_ids=[m1.id, m2.id], similarity=0.9,
+        proposed_content="merged",
+        proposed_memory_type="knowledge", proposed_project="",
+    )
+    # Tamper: supersede one source before the merge runs
+    with get_connection() as conn:
+        conn.execute("UPDATE memories SET status = 'superseded' WHERE id = ?", (m1.id,))
+    result = merge_memories(candidate_id=cid)
+    assert result is None
+
+    # Candidate still pending
+    pending = get_merge_candidates(status="pending")
+    assert any(c["id"] == cid for c in pending)
+
+
+def test_merge_memories_rejects_already_resolved(tmp_data_dir):
+    init_db()
+    m1 = create_memory("knowledge", "x")
+    m2 = create_memory("knowledge", "y")
+    cid = create_merge_candidate(
+        memory_ids=[m1.id, m2.id], similarity=0.9,
+        proposed_content="xy",
+        proposed_memory_type="knowledge", proposed_project="",
+    )
+    first = merge_memories(candidate_id=cid)
+    assert first is not None
+    # second call — already approved, should return None
+    second = merge_memories(candidate_id=cid)
+    assert second is None
+
+
+# --- reject_merge_candidate ---
+
+
+def test_reject_merge_candidate_leaves_sources_untouched(tmp_data_dir):
+    init_db()
+    m1 = create_memory("knowledge", "a")
+    m2 = create_memory("knowledge", "b")
+    cid = create_merge_candidate(
+        memory_ids=[m1.id, m2.id], similarity=0.9,
+        proposed_content="a+b",
+        proposed_memory_type="knowledge", proposed_project="",
+    )
+    ok = reject_merge_candidate(cid, actor="human-triage", note="false positive")
+    assert ok
+
+    # Sources still active
+    with get_connection() as conn:
+        s1 = conn.execute("SELECT status FROM memories WHERE id = ?", (m1.id,)).fetchone()
+        s2 = conn.execute("SELECT status FROM memories WHERE id = ?", (m2.id,)).fetchone()
+        cand = conn.execute(
+            "SELECT status FROM memory_merge_candidates WHERE id = ?", (cid,)
+        ).fetchone()
+    assert s1["status"] == "active"
+    assert s2["status"] == "active"
+    assert cand["status"] == "rejected"
+
+
+def test_reject_merge_candidate_idempotent(tmp_data_dir):
+    init_db()
+    m1 = create_memory("knowledge", "p")
+    m2 = create_memory("knowledge", "q")
+    cid = create_merge_candidate(
+        memory_ids=[m1.id, m2.id], similarity=0.9,
+        proposed_content="pq",
+        proposed_memory_type="knowledge", proposed_project="",
+    )
+    assert reject_merge_candidate(cid) is True
+    # second reject — already rejected, returns False
+    assert reject_merge_candidate(cid) is False
+
+
+# --- Schema sanity ---
+
+
+def test_merge_candidates_table_exists(tmp_data_dir):
+    init_db()
+    with get_connection() as conn:
+        cols = [r["name"] for r in conn.execute("PRAGMA table_info(memory_merge_candidates)").fetchall()]
+    expected = {"id", "status", "memory_ids", "similarity", "proposed_content",
+                "proposed_memory_type", "proposed_project", "proposed_tags",
+                "proposed_confidence", "reason", "created_at", "resolved_at",
+                "resolved_by", "result_memory_id"}
+    assert expected.issubset(set(cols))