@monoes/monomindcli 2.10.9 → 2.10.13

package/.claude/skills/mastermind-plan/SKILL.md

@@ -40,6 +40,7 @@ If the spec covers multiple independent subsystems, it should have been broken i
 
 Before defining tasks, map out which files will be created or modified and what each one is responsible for. This is where decomposition decisions get locked in.
 
+- Check `brain_context` for relevant prior lessons or patterns for this domain/project — past task-granularity mistakes, file-organization conventions, known gotchas — and apply them to the decomposition below. Note when a decomposition choice is informed by recalled history.
 - Design units with clear boundaries and well-defined interfaces. Each file should have one clear responsibility.
 - You reason best about code you can hold in context at once, and your edits are more reliable when files are focused. Prefer smaller, focused files over large ones that do too much.
 - Files that change together should live together. Split by responsibility, not by technical layer.
@@ -73,7 +74,7 @@ A task is the smallest unit that carries its own test cycle and is worth a fresh
 ```markdown
 # [Feature Name] Implementation Plan
 
-> **For agentic workers:** REQUIRED SUB-SKILL: Use `Skill("mastermind-taskdev")` (recommended) or `Skill("mastermind-execute")` to implement this plan task-by-task. Steps use checkbox (`- [ ]`) syntax for tracking.
+> **For agentic workers:** REQUIRED SUB-SKILL: Use `Skill("mastermind-execute")` to implement this plan task-by-task. Steps use checkbox (`- [ ]`) syntax for tracking.
 
 **Goal:** [One sentence describing what this builds]
 
@@ -196,234 +197,12 @@ Wait for the user's response. If they request changes, make them inline and re-r
 
 After the plan is approved (or in auto mode, after self-review):
 
-**In confirm mode (default):** Ask the user to choose execution mode:
+**In confirm mode (default):** Ask the user to confirm execution:
 
-**"Plan complete and saved to `docs/mastermind/plans/<filename>.md`. Two execution options:**
+**"Plan complete and saved to `docs/mastermind/plans/<filename>.md`. Execute it now with `mastermind-execute`?"**
 
-**1. Subagent-Driven (recommended)** — Invoke `Skill("mastermind-taskdev")`: dispatches a fresh subagent per task, reviews between tasks, fast iteration.
+**In auto mode:** Skip the question — invoke `Skill("mastermind-execute")` immediately.
 
-**2. Inline Execution** — Invoke `Skill("mastermind-execute")`: batch execution with checkpoints.
-
-**Which approach?"**
-
-**In auto mode:** Skip the question. Default to subagent-driven — invoke `Skill("mastermind-taskdev")` immediately.
-
-**If Subagent-Driven chosen (or auto mode):**
-- Invoke `Skill("mastermind-taskdev")`
-- Fresh subagent per task + two-stage review
-
-**If Inline Execution chosen:**
-- Invoke `Skill("mastermind-execute")`
-- Batch execution with checkpoints for review
-# monomind:start skills:claude:mastermind-plan
-# Mastermind Plan
-
-## Overview
-
-Write comprehensive implementation plans assuming the engineer has zero context for our codebase and questionable taste. Document everything they need to know: which files to touch for each task, code, testing, docs they might need to check, how to test it. Give them the whole plan as bite-sized tasks. DRY. YAGNI. TDD. Frequent commits.
-
-Assume they are a skilled developer, but know almost nothing about our toolset or problem domain. Assume they don't know good test design very well.
-
-**Announce at start:** "I'm using the mastermind:plan skill to create the implementation plan."
-
-**Context:** If working in an isolated worktree, it should have been created via the git worktrees workflow at execution time.
-
-**Save plans to:** `docs/mastermind/plans/YYYY-MM-DD-<feature-name>.md`
-- (User preferences for plan location override this default)
-
----
-
-## Inputs
-
-- `brain_context`: BRAIN CONTEXT block (loaded via mastermind-protocol/SKILL.md brain load)
-- `params`: spec text, feature description, or path to spec file
-- `mode`: auto | confirm
-
----
-
-## Scope Check
-
-If the spec covers multiple independent subsystems, it should have been broken into sub-project specs during brainstorming. If it wasn't, suggest breaking this into separate plans — one per subsystem. Each plan should produce working, testable software on its own.
-
----
-
-## File Structure
-
-Before defining tasks, map out which files will be created or modified and what each one is responsible for. This is where decomposition decisions get locked in.
-
-- Design units with clear boundaries and well-defined interfaces. Each file should have one clear responsibility.
-- You reason best about code you can hold in context at once, and your edits are more reliable when files are focused. Prefer smaller, focused files over large ones that do too much.
-- Files that change together should live together. Split by responsibility, not by technical layer.
-- In existing codebases, follow established patterns. If the codebase uses large files, don't unilaterally restructure — but if a file you're modifying has grown unwieldy, including a split in the plan is reasonable.
-
-This structure informs the task decomposition. Each task should produce self-contained changes that make sense independently.
-
----
-
-## Task Right-Sizing
-
-A task is the smallest unit that carries its own test cycle and is worth a fresh reviewer's gate. When drawing task boundaries: fold setup, configuration, scaffolding, and documentation steps into the task whose deliverable needs them; split only where a reviewer could meaningfully reject one task while approving its neighbor. Each task ends with an independently testable deliverable.
-
----
-
-## Bite-Sized Task Granularity
-
-**Each step is one action (2-5 minutes):**
-- "Write the failing test" — step
-- "Run it to make sure it fails" — step
-- "Implement the minimal code to make the test pass" — step
-- "Run the tests and make sure they pass" — step
-- "Commit" — step
-
----
-
-## Plan Document Header
-
-**Every plan MUST start with this header:**
-
-```markdown
-# [Feature Name] Implementation Plan
-
-> **For agentic workers:** REQUIRED SUB-SKILL: Use `Skill("mastermind-taskdev")` (recommended) or `Skill("mastermind-execute")` to implement this plan task-by-task. Steps use checkbox (`- [ ]`) syntax for tracking.
-
-**Goal:** [One sentence describing what this builds]
-
-**Architecture:** [2-3 sentences about approach]
-
-**Tech Stack:** [Key technologies/libraries]
-
-## Global Constraints
-
-[The spec's project-wide requirements — version floors, dependency limits,
-naming and copy rules, platform requirements — one line each, with exact
-values copied verbatim from the spec. Every task's requirements implicitly
-include this section.]
-
----
-```
-
----
-
-## Task Structure
-
-````markdown
-### Task N: [Component Name]
-
-**Files:**
-- Create: `exact/path/to/file.py`
-- Modify: `exact/path/to/existing.py:123-145`
-- Test: `tests/exact/path/to/test.py`
-
-**Interfaces:**
-- Consumes: [what this task uses from earlier tasks — exact signatures]
-- Produces: [what later tasks rely on — exact function names, parameter
-  and return types. A task's implementer sees only their own task; this
-  block is how they learn the names and types neighboring tasks use.]
-
-- [ ] **Step 1: Write the failing test**
-
-```python
-def test_specific_behavior():
-    result = function(input)
-    assert result == expected
-```
-
-- [ ] **Step 2: Run test to verify it fails**
-
-Run: `pytest tests/path/test.py::test_name -v`
-Expected: FAIL with "function not defined"
-
-- [ ] **Step 3: Write minimal implementation**
-
-```python
-def function(input):
-    return expected
-```
-
-- [ ] **Step 4: Run test to verify it passes**
-
-Run: `pytest tests/path/test.py::test_name -v`
-Expected: PASS
-
-- [ ] **Step 5: Commit**
-
-```bash
-git add tests/path/test.py src/path/file.py
-git commit -m "feat: add specific feature"
-```
-````
-
----
-
-## No Placeholders
-
-Every step must contain the actual content an engineer needs. These are **plan failures** — never write them:
-- "TBD", "TODO", "implement later", "fill in details"
-- "Add appropriate error handling" / "add validation" / "handle edge cases"
-- "Write tests for the above" (without actual test code)
-- "Similar to Task N" (repeat the code — the engineer may be reading tasks out of order)
-- Steps that describe what to do without showing how (code blocks required for code steps)
-- References to types, functions, or methods not defined in any task
-
----
-
-## Remember
-- Exact file paths always
-- Complete code in every step — if a step changes code, show the code
-- Exact commands with expected output
-- DRY, YAGNI, TDD, frequent commits
-
----
-
-## Self-Review
-
-After writing the complete plan, look at the spec with fresh eyes and check the plan against it. This is a checklist you run yourself — not a subagent dispatch.
-
-**1. Spec coverage:** Skim each section/requirement in the spec. Can you point to a task that implements it? List any gaps.
-
-**2. Placeholder scan:** Search your plan for red flags — any of the patterns from the "No Placeholders" section above. Fix them.
-
-**3. Type consistency:** Do the types, method signatures, and property names you used in later tasks match what you defined in earlier tasks? A function called `clearLayers()` in Task 3 but `clearFullLayers()` in Task 7 is a bug.
-
-If you find issues, fix them inline. No need to re-review — just fix and move on. If you find a spec requirement with no task, add the task.
-
----
-
-## User Review Gate
-
-After self-review:
-
-**In confirm mode (default):** Ask the user to review the written plan before proceeding:
-
-> "Plan written and saved to `docs/mastermind/plans/<filename>.md`. Please review it and let me know if you'd like any changes before we start execution."
-
-Wait for the user's response. If they request changes, make them inline and re-run the self-review. Only proceed once the user approves.
-
-**In auto mode:** Skip the wait. Proceed directly to execution handoff.
-
----
-
-## Execution Handoff
-
-After the plan is approved (or in auto mode, after self-review):
-
-**In confirm mode (default):** Ask the user to choose execution mode:
-
-**"Plan complete and saved to `docs/mastermind/plans/<filename>.md`. Two execution options:**
-
-**1. Subagent-Driven (recommended)** — Invoke `Skill("mastermind-taskdev")`: dispatches a fresh subagent per task, reviews between tasks, fast iteration.
-
-**2. Inline Execution** — Invoke `Skill("mastermind-execute")`: batch execution with checkpoints.
-
-**Which approach?"**
-
-**In auto mode:** Skip the question. Default to subagent-driven — invoke `Skill("mastermind-taskdev")` immediately.
-
-**If Subagent-Driven chosen (or auto mode):**
-- Invoke `Skill("mastermind-taskdev")`
-- Fresh subagent per task + two-stage review
-
-**If Inline Execution chosen:**
+**On execution:**
 - Invoke `Skill("mastermind-execute")`
-- Batch execution with checkpoints for review
-# monomind:end skills:claude:mastermind-plan
+- It loads the plan, reviews it critically, and works task-by-task with checkpoints for review

package/.claude/skills/mastermind-plan-to-tasks/SKILL.md

@@ -70,15 +70,24 @@ Read the plan carefully and apply these rules:
 
 **Decomposition output:**
 
-For each issue extracted from the plan, produce:
-```
-ISSUE: <title>
-  assignee:    <agent-id or UNASSIGNED>
-  priority:    low | medium | high | urgent
-  blockedBy:   <issue-title> (or none)
-  description: <1-2 sentence summary of the deliverable>
+For each issue extracted from the plan, emit one object into a JSON array and write the
+whole array to `.monomind/orgs/<org_name>-plan-decomposition.json` using the Write tool:
+
+```json
+[
+  {
+    "title": "Short imperative deliverable name",
+    "description": "1-2 sentence summary of the deliverable",
+    "assignee": "<agent-id from the org roster, or null if unassigned>",
+    "priority": "low | medium | high | urgent",
+    "blockedBy": ["<exact title of a blocking issue in this same array>"]
+  }
+]
 ```
 
+Titles must be unique within the array — `blockedBy` is resolved by exact title match.
+Use `[]` for `blockedBy` when an issue has no blockers.
+
 **Quality checklist before creating:**
 - [ ] Enough detail that assignees can act without re-asking
 - [ ] Every concrete deliverable is an issue
@@ -91,31 +100,130 @@ ISSUE: <title>
 
 ## Step 4 — Create Issues (unless dry_run=true)
 
+Reads the decomposition written in Step 3, assigns collision-free ids, validates assignees
+against the org roster, resolves `blockedBy` titles into `blockedByIssueIds`, and appends
+to the issues file atomically. Fails loudly on any unresolved or ambiguous reference —
+an unresolved blocker silently dropped is exactly what produces a `blocked` issue with an
+empty `blockedByIssueIds`, which `mastermind-liveness` reports as stalled.
+
 ```bash
 issuesFile=".monomind/orgs/${org_name}-issues.json"
+decompFile=".monomind/orgs/${org_name}-plan-decomposition.json"
 [ ! -f "$issuesFile" ] && echo '{"issues":[]}' > "$issuesFile"
+[ ! -f "$decompFile" ] && { echo "ERROR: $decompFile not found — Step 3 must write it first."; exit 1; }
 
 ts=$(date -u +%Y-%m-%dT%H:%M:%SZ)
 
-# Issues are created from the decomposition above
-# Each issue gets:
-#   id: issue-<timestamp>-<N>
-#   title, description, priority, status: open
-#   assigneeId (from matched agent)
-#   projectId, workspaceId (if provided)
-#   blockedByIssueIds (resolved after all issues are created)
-#   createdAt, updatedAt
-
-echo ""
-if [ "${dry_run:-false}" = "true" ]; then
-  echo "DRY RUN — no issues were created."
-  echo "Remove --dry-run to create these issues."
-else
-  echo "CREATED ISSUES:"
-  # (issue creation happens inline above as each issue is decomposed)
-fi
+python3 - "$issuesFile" "$decompFile" "$orgFile" "${project_id:-}" "${workspace_id:-}" "${dry_run:-false}" "$ts" <<'PYEOF'
+import json, os, sys, time
+
+issues_path, decomp_path, org_path, project_id, workspace_id, dry_run, ts = sys.argv[1:]
+dry = dry_run == "true"
+
+VALID_PRIORITY = {"low", "medium", "high", "urgent"}
+
+decomp = json.load(open(decomp_path))
+if not isinstance(decomp, list) or not decomp:
+    print("ERROR: decomposition must be a non-empty JSON array.")
+    sys.exit(1)
+
+roster = {r.get("id") for r in json.load(open(org_path)).get("roles", [])}
+data   = json.load(open(issues_path))
+issues = data.setdefault("issues", [])
+
+# --- Validate before mutating anything ---
+errors, titles = [], {}
+for n, item in enumerate(decomp, 1):
+    title = (item.get("title") or "").strip()
+    if not title:
+        errors.append(f"item {n}: missing title")
+        continue
+    if title in titles:
+        errors.append(f"item {n}: duplicate title {title!r} — titles must be unique to resolve blockedBy")
+    titles[title] = None
+    pri = item.get("priority") or "medium"
+    if pri not in VALID_PRIORITY:
+        errors.append(f"{title!r}: priority {pri!r} must be one of {sorted(VALID_PRIORITY)}")
+    asgn = item.get("assignee")
+    if asgn and asgn not in roster:
+        errors.append(f"{title!r}: assignee {asgn!r} is not a role in this org (roster: {sorted(roster)})")
+
+existing_by_title = {i.get("title"): i.get("id") for i in issues}
+for item in decomp:
+    for b in item.get("blockedBy") or []:
+        if b not in titles and b not in existing_by_title:
+            errors.append(f"{item.get('title')!r}: blockedBy {b!r} matches no issue in this batch or in the existing file")
+
+if errors:
+    print("ERROR: decomposition did not validate — no issues were created.")
+    for e in errors:
+        print(f"  · {e}")
+    sys.exit(1)
+
+# --- Assign ids (batch-stable, collision-free) ---
+batch = int(time.time() * 1000)
+for n, item in enumerate(decomp, 1):
+    titles[item["title"].strip()] = f"issue-{batch}-{n:03d}"
+
+def resolve(t):
+    return titles.get(t) or existing_by_title[t]
+
+created = []
+for item in decomp:
+    title = item["title"].strip()
+    created.append({
+        "id": titles[title],
+        "title": title,
+        "description": item.get("description") or "",
+        "status": "todo",
+        "priority": item.get("priority") or "medium",
+        "assigneeId": item.get("assignee") or None,
+        "assigneeAgentId": item.get("assignee") or None,
+        "assigneeUserId": None,
+        "parentId": None,
+        "projectId": project_id or None,
+        "workspaceId": workspace_id or None,
+        "blockedByIssueIds": [resolve(b) for b in (item.get("blockedBy") or [])],
+        "createdAt": ts,
+        "updatedAt": ts,
+    })
+
+if dry:
+    print("DRY RUN — no issues were created. Would create:")
+    for c in created:
+        blockers = ", ".join(c["blockedByIssueIds"]) or "none"
+        print(f"  {c['id']}  [{c['priority']:<6}] {c['title']}")
+        print(f"      assignee: {c['assigneeId'] or 'UNASSIGNED'}   blockedBy: {blockers}")
+    print(f"\n  {len(created)} issue(s) would be created.")
+    sys.exit(0)
+
+issues.extend(created)
+tmp = issues_path + ".tmp"
+with open(tmp, "w") as f:
+    json.dump(data, f, indent=2)
+os.replace(tmp, issues_path)
+
+print("CREATED ISSUES:")
+for c in created:
+    blockers = ", ".join(c["blockedByIssueIds"]) or "none"
+    print(f"  {c['id']}  [{c['priority']:<6}] {c['title']}")
+    print(f"      assignee: {c['assigneeId'] or 'UNASSIGNED'}   blockedBy: {blockers}")
+print(f"\n  {len(created)} issue(s) created in {issues_path}")
+
+roots = [c for c in created if not c["blockedByIssueIds"]]
+print(f"  {len(roots)} issue(s) are unblocked and can start in parallel.")
+PYEOF
+
+rm -f "$decompFile"
 ```
 
+> **These issues are not auto-consumed by the org runtime.** `OrgDaemon` builds its `TaskDag`
+> fresh in memory on every start and never reads `<org>-issues.json` (verified: zero references
+> in `packages/@monomind/cli/src/orgrt/`). Issues created here are tracked by the
+> `mastermind-issues` / `mastermind-my-issues` / `mastermind-liveness` skills and shown in the
+> dashboard, but running `monomind org run <org>` will not pick them up. Drive execution from
+> this file with `mastermind-execute`, or pass work explicitly via `org run --task`.
+
 ---
 
 ## Step 5 — Return Output

package/.claude/skills/mastermind-protocol/SKILL.md

@@ -30,17 +30,19 @@ Execute at the START of every mastermind run (master or standalone domain comman
 
 **Step A — Tier 3 core principles (all domains):**
 Try `mcp__monomind__memory_hierarchical-recall` with query `"mastermind principles"`, topK 20.
-If it returns `"LanceDB bridge not available"` or any error, fall back to:
+If it returns an error, fall back to:
 `mcp__monomind__memory_pattern-search` with query `"mastermind principles"`, namespace `"mastermind:principles"`, limit 20.
+Record the entry ids of whichever call actually returned results — `results[].id` from `memory_hierarchical-recall`, or `patterns[].id` from the `memory_pattern-search` fallback — into a running list: `recalled_entry_ids`.
 
 **Step B — Tier 2 weekly summary for this domain:**
 Try `mcp__monomind__memory_context-synthesize` with query `[current prompt keywords]`, maxEntries 10.
 If it fails, fall back to:
 `mcp__monomind__memory_pattern-search` with query `[current prompt keywords]`, namespace `"mastermind:<domain>:weekly"`, limit 10.
+`memory_context-synthesize` returns only a `context` string and a `sources` count — no entry ids — so it cannot add to `recalled_entry_ids`. Only the `memory_pattern-search` fallback contributes ids (`patterns[].id`); append them to `recalled_entry_ids` when that fallback runs.
 
 **Step C — Relevant graph nodes:**
 Call `mcp__monomind__monograph_query` with question `[3-5 keywords extracted from current prompt]`, depth 2.
-If the graph is not built yet (error: "No graph found"), skip this tier — continue without graph context.
+If the graph is not built yet (error: "No graph found"), skip this tier — continue without graph context. Graph nodes are not memory entries and never contribute to `recalled_entry_ids`.
 
 Combine all results into a **BRAIN CONTEXT** block. Insert this block before any planning, decomposition, or agent spawning step. Format:
 
@@ -54,19 +56,30 @@ Combine all results into a **BRAIN CONTEXT** block. Insert this block before any
 
 If any MCP call fails or returns empty, continue without that tier — do not abort the run.
 
+**Step D — Visible recall confirmation:**
+Immediately after loading, print one short line as part of the calling skill's normal output — not buried inside the BRAIN CONTEXT block or an internal note the user never sees:
+- If anything was recalled: `Remembered: <one-line gist of the single most relevant recalled item>`
+- If every tier came back empty: `Nothing relevant recalled.`
+
+Keep `recalled_entry_ids` in scope for the rest of the run — the Brain Write Procedure's feedback step below needs it. Memory tools are stateless between calls, so the calling skill (not the MCP server) is what carries this list forward.
+
 ---
 
 ## Brain Write Procedure
 
 Execute at the END of every mastermind run. Always runs even if execution was partial or blocked.
 
-**Step 1 — Score this run:**
-```
-score = confidence × (1 / (days_since_run + 1)) × log(uses + 1)
-```
-- `confidence`: average of all decision confidence values from the unified output schema
-- `days_since_run`: 0 (this is a fresh run)
-- `uses`: 1 (first write)
+**Step 1 — Close the feedback loop on recalled memory:**
+If `recalled_entry_ids` from the Brain Load Procedure is non-empty, call `mcp__monomind__memory_feedback`:
+- `taskId`: the `run_id` used in this run's unified output schema (ISO8601 timestamp) — also the idempotency key, so a retried or duplicated call for the same run never double-applies the rating
+- `entryIds`: `recalled_entry_ids`
+- `quality`: average of `decisions[].confidence` (0.0-1.0) from this run's unified output schema; omit if `decisions` is empty
+- `success`: `true` when `status: complete`, or `status: partial` with no `outcome: reverted` decisions and no user rejection/correction of the recalled guidance during the run; otherwise `false`
+- `agent`: `<domain>` (this skill's domain)
+
+This applies an EWMA update to `feedback_weight` on each recalled entry (idempotent per `taskId`) and appends a feedback-event record via the same tool call. If `recalled_entry_ids` is empty — e.g. Brain Load only got a hit from `memory_context-synthesize`, which returns no entry ids — skip this call; there is nothing to rate.
+
+`score` referenced in Steps 2-3 below is this same value: the average of `decisions[].confidence` from this run's unified output schema. The old day/use decay factors were already constants on a fresh write (day 0, first use), so this is equivalent and simpler.
 
 **Step 2 — Append to Tier 1 raw log:**
 Try `mcp__monomind__memory_hierarchical-store` with:
@@ -74,7 +87,7 @@ Try `mcp__monomind__memory_hierarchical-store` with:
 - content: [full unified output schema YAML from this run, as a string]
 - metadata: `{ score, project, run_id, date: ISO8601, domain }`
 
-If LanceDB is unavailable, fall back to `mcp__monomind__memory_pattern-store`:
+If hierarchical storage is unavailable, fall back to `mcp__monomind__memory_pattern-store`:
 - key: `mastermind:<domain>:run:<run_id>`
 - value: [JSON-encoded unified output schema]
 - namespace: `mastermind:<domain>:raw`
@@ -82,7 +95,7 @@ If LanceDB is unavailable, fall back to `mcp__monomind__memory_pattern-store`:
 
 **Step 3 — Check weekly compaction trigger:**
 Try `mcp__monomind__memory_health` on namespace `mastermind:<domain>:raw`.
-If unavailable, call `mcp__monomind__memory_stats` and check entry count manually.
+If unavailable, call `mcp__monomind__memory_health` and check entry count manually.
 If `entry_count >= 20` OR `days_since_last_compaction >= 7`:
 1. Retrieve all Tier 1 entries since last compaction
 2. Produce a per-domain weekly summary (use LLM synthesis: "Summarize the key decisions, patterns, and lessons from these run logs in under 300 words")
@@ -94,7 +107,7 @@ Call `mcp__monomind__monograph_community` for nodes matching `mastermind:<domain
 If 3+ similar memory nodes are detected in a cluster:
 1. Merge into a single principle via LLM: "Distill these memories into one clear principle in 1-2 sentences"
 2. Store principle: `mcp__monomind__memory_hierarchical-store` namespace `mastermind:principles`
-3. Add `EXCEPTION` edge in Monograph for any conflicting memory: `mcp__monomind__monograph_add_fact`
+3. Add `EXCEPTION` edge in Monograph for any conflicting memory: `mcp__monomind__memory_kg_ingest`
 
 ---
 
@@ -126,19 +139,17 @@ Empty fields use `[]`. The `status` field reflects the highest-level result: `co
 
 ---
 
-## Memory Scoring Formula
+## Memory Feedback Loop
 
-```
-score = confidence × (1 / (days_since_run + 1)) × log(uses + 1)
-```
+Recalled memories are rated for real via `mcp__monomind__memory_feedback` (Brain Write Procedure, Step 1) — not a locally-computed formula. `score` (used by Steps 2-3 for the newly-stored run entry) is the average of `decisions[].confidence` from the unified output schema.
 
-| Variable | Source |
+| Concept | Mechanism |
 |---|---|
-| `confidence` | Average of `decisions[].confidence` from output schema |
-| `days_since_run` | 0 on day of run, increases 1 per calendar day |
-| `uses` | Incremented each time this memory is returned by brain load |
-| Archive threshold | score < 0.1 |
-| Reinforcement | Increment `uses` on every brain load hit |
+| `feedback_weight` | Per-entry EWMA, updated by `memory_feedback`'s `quality`/`success` params, blended into future ranking (`bridgeApplyFeedback` in `memory-bridge.ts`) |
+| `frequency_weight` | Incremented automatically on every recall hit, inside the `memory_hierarchical-recall` / `memory_pattern-search` handlers themselves — protects frequently-used entries from GC |
+| Idempotency | `memory_feedback`'s `taskId` is the idempotency key — the same `run_id` never double-applies a rating |
+| Archive threshold | score < 0.1 (Step 3 above) |
+| GC protection | `bridgeConsolidate` never collects entries with `feedback_weight > 0.6` or `frequency_weight >= 3` |
 
 ---