feat: measurer.py — compounding intelligence metrics engine

Implements issue #14: 7 metrics that prove knowledge compounding. Metrics: - Knowledge velocity: new facts/day (from index.json) - Knowledge coverage: % domains with 10+ facts (from YAML files) - Hit rate: % sessions referencing bootstrap knowledge - Error recurrence: same errors across sessions (should decrease) - Task completion: % sessions with successful end_reason - First-try success: actions without backtracking (tool/msg ratio) - Knowledge age: staleness of facts (freshness score) Data sources: - knowledge/index.json + YAML files for fact metrics - ~/.hermes/state.db sessions + messages tables Features: - JSON and markdown output formats - --since, --repo, --format flags - 7-day trend tracking via snapshot persistence - Runs in 33ms on 11.9K sessions / 192K messages - Dashboard auto-generation with --save-snapshot Closes #14
2026-04-14 14:16:31 -04:00
12 changed files with 1135 additions and 751 deletions
--- a/knowledge/SCHEMA.md
+++ b/knowledge/SCHEMA.md
@@ -0,0 +1,164 @@
+# Knowledge File Format Specification
+
+**Version:** 1
+**Issue:** #10
+**Status:** Draft
+
+---
+
+## Overview
+
+The knowledge system has two layers:
+
+1. **index.json** — Machine-readable fact index. Fast lookups by ID, category, repo, tags.
+2. **Knowledge files** (YAML) — Human-readable, editable facts organized by domain.
+
+The harvester writes to both. The bootstrapper reads from index.json. Humans edit the YAML files directly.
+
+---
+
+## index.json Schema
+
+```json
+{
+  "version": 1,
+  "last_updated": "ISO-8601 timestamp",
+  "total_facts": 0,
+  "facts": []
+}
+```
+
+### Fact Object
+
+| Field | Type | Required | Description |
+|-------|------|----------|-------------|
+| `id` | string | yes | Unique identifier: `{domain}:{category}:{sequence}` |
+| `fact` | string | yes | One-sentence description of the knowledge |
+| `category` | enum | yes | One of: `fact`, `pitfall`, `pattern`, `tool-quirk`, `question` |
+| `domain` | string | yes | Where this applies: repo name, `global`, or agent name |
+| `confidence` | float | yes | 0.0-1.0. How certain is this knowledge? |
+| `tags` | string[] | no | Searchable labels: `["git", "auth", "gitea"]` |
+| `source_count` | int | no | How many sessions confirmed this fact |
+| `first_seen` | date | no | ISO-8601 date first extracted |
+| `last_confirmed` | date | no | ISO-8601 date last seen in a session |
+| `expires` | date | no | Optional. After this date, fact is stale |
+| `related` | string[] | no | IDs of related facts |
+
+### ID Format
+
+```
+{domain}:{category}:{sequence}
+```
+
+- `domain` — repo name, `global`, or agent type
+- `category` — one of the 5 categories
+- `sequence` — zero-padded 3-digit number: `001`, `002`, ...
+
+Examples:
+- `the-nexus:pitfall:001`
+- `global:tool-quirk:012`
+- `hermes-agent:pattern:003`
+
+### Categories
+
+| Category | Definition | Example |
+|----------|------------|---------|
+| `fact` | Concrete, verifiable information | "Gitea API requires token auth at /api/v1" |
+| `pitfall` | Errors, wrong assumptions, time-wasters | "Assumed env var GITEA_TOKEN; actual path is ~/.config/gitea/token" |
+| `pattern` | Successful sequences of actions | "To deploy: test -> build -> push -> webhook" |
+| `tool-quirk` | Environment-specific behaviors | "URL format requires trailing slash on macOS" |
+| `question` | Identified but unanswered | "Need optimal batch size for harvesting" |
+
+### Confidence Scoring
+
+| Range | Meaning |
+|-------|---------|
+| 0.9-1.0 | Explicitly stated and verified |
+| 0.7-0.8 | Clearly implied by multiple data points |
+| 0.5-0.6 | Suggested but not fully verified |
+| 0.3-0.4 | Inferred from limited data |
+| 0.1-0.2 | Speculative or uncertain |
+
+---
+
+## Knowledge Files (YAML)
+
+Human-readable files stored in `knowledge/` subdirectories.
+
+### Directory Structure
+
+```
+knowledge/
+├── index.json                  # Machine-readable fact index
+├── SCHEMA.md                   # This file
+├── global/                     # Cross-repo knowledge
+│   ├── pitfalls.yaml           # Pitfalls that span multiple repos
+│   ├── patterns.yaml           # Proven workflows
+│   └── tool-quirks.yaml        # Environment behaviors
+├── repos/                      # Per-repo knowledge
+│   ├── the-nexus.yaml
+│   ├── hermes-agent.yaml
+│   └── ...
+└── agents/                     # Agent-type knowledge
+    ├── mimo-sprint.yaml
+    └── ...
+```
+
+### YAML File Format
+
+```yaml
+---
+domain: global                    # or repo name or agent name
+category: tool-quirk              # fact, pitfall, pattern, tool-quirk, question
+version: 1
+last_updated: "2026-04-13"
+---
+
+# Tool Quirks (Global)
+
+Cross-environment behaviors that bite you if you don't know them.
+
+## Authentication
+
+- id: global:tool-quirk:001
+  fact: "Gitea token stored at ~/.config/gitea/token, not env var"
+  confidence: 0.95
+  tags: [git, auth, gitea]
+  source_count: 23
+  first_seen: "2026-03-27"
+  last_confirmed: "2026-04-13"
+  related: [global:pitfall:003]
+```
+
+### Rules
+
+1. **One file per domain per category.** `repos/the-nexus.yaml` holds all the-nexus facts.
+2. **Markdown sections for humans.** YAML items live under markdown headers for Gitea UI readability.
+3. **ID is the link.** The `id` field connects YAML facts to index.json entries.
+4. **Harvester writes, humans edit.** Harvester appends. Humans correct confidence, add tags, mark expired.
+
+---
+
+## Sync Rules
+
+1. **Harvester -> YAML:** Appends new facts to the appropriate YAML file.
+2. **Harvester -> index.json:** Adds/updates fact entries.
+3. **Human edits YAML:** Changes propagate to index.json on next harvester run.
+4. **Confidence decay:** Facts not confirmed in 30+ sessions get confidence *= 0.9.
+5. **Expiration:** Facts with `expires` date past current date are marked `stale`.
+
+---
+
+## Validation
+
+Facts must pass these checks:
+
+1. `id` matches format `{domain}:{category}:{sequence}`
+2. `category` is one of the 5 allowed values
+3. `confidence` is between 0.0 and 1.0
+4. `fact` is non-empty string, max 280 characters
+5. `domain` is non-empty string
+6. `tags` are lowercase alphanumeric + hyphens
+7. No duplicate IDs in index.json
+
+Validation script: `scripts/validate_knowledge.py`
--- a/knowledge/global/pitfalls.yaml
+++ b/knowledge/global/pitfalls.yaml
@@ -0,0 +1,80 @@
+---
+domain: global
+category: pitfall
+version: 1
+last_updated: "2026-04-13"
+---
+
+# Pitfalls (Global)
+
+Cross-repo traps that waste time across the fleet.
+
+## Git & Forge
+
+- id: global:pitfall:001
+  fact: "Branch protection requires 1 approval on main - API merges fail with 405 without it"
+  confidence: 0.95
+  tags: [git, merge, branch-protection, gitea]
+  source_count: 12
+  first_seen: "2026-04-05"
+  last_confirmed: "2026-04-13"
+  related: [the-nexus:pitfall:001]
+
+- id: global:pitfall:002
+  fact: "Never use --no-verify on git commits - it bypasses all hooks including safety checks"
+  confidence: 0.95
+  tags: [git, hooks, safety]
+  source_count: 5
+  first_seen: "2026-03-28"
+  last_confirmed: "2026-04-13"
+
+- id: global:pitfall:003
+  fact: "Gitea PR creation workaround needed on the-nexus - direct API call fails, use alternative endpoint"
+  confidence: 0.9
+  tags: [gitea, pr, api, workaround]
+  source_count: 4
+  first_seen: "2026-04-06"
+  last_confirmed: "2026-04-12"
+
+## Agent Operations
+
+- id: global:pitfall:004
+  fact: "Anthropic is BANNED from fallback chain - if fallback triggers to Anthropic, something is wrong"
+  confidence: 0.95
+  tags: [provider, anthropic, fallback]
+  source_count: 7
+  first_seen: "2026-03-30"
+  last_confirmed: "2026-04-13"
+
+- id: global:pitfall:005
+  fact: "Telegram tokens expired - don't assume Telegram notifications work without checking"
+  confidence: 0.85
+  tags: [telegram, notifications, token]
+  source_count: 3
+  first_seen: "2026-04-02"
+
+- id: global:pitfall:006
+  fact: "Multiple gateways = 'cannot schedule futures' error - only one gateway process should run"
+  confidence: 0.9
+  tags: [gateway, cron, process]
+  source_count: 4
+  first_seen: "2026-04-04"
+  last_confirmed: "2026-04-11"
+
+## Testing
+
+- id: global:pitfall:007
+  fact: "pytest root collection picks up operational *_test.py scripts - restrict to tests/ directory"
+  confidence: 0.9
+  tags: [pytest, test, collection]
+  source_count: 3
+  first_seen: "2026-04-07"
+  last_confirmed: "2026-04-13"
+
+- id: global:pitfall:008
+  fact: "TDD: test 1 before building 55 - verify the cycle works before scaling"
+  confidence: 0.95
+  tags: [tdd, testing, methodology]
+  source_count: 8
+  first_seen: "2026-03-25"
+  last_confirmed: "2026-04-13"
--- a/knowledge/global/tool-quirks.yaml
+++ b/knowledge/global/tool-quirks.yaml
@@ -0,0 +1,73 @@
+---
+domain: global
+category: tool-quirk
+version: 1
+last_updated: "2026-04-13"
+---
+
+# Tool Quirks (Global)
+
+Cross-environment behaviors that bite you if you don't know them.
+
+## Authentication
+
+- id: global:tool-quirk:001
+  fact: "Gitea token stored at ~/.config/gitea/token, not env var GITEA_TOKEN"
+  confidence: 0.95
+  tags: [git, auth, gitea, token]
+  source_count: 23
+  first_seen: "2026-03-27"
+  last_confirmed: "2026-04-13"
+  related: [global:pitfall:001]
+
+- id: global:tool-quirk:002
+  fact: "Gitea API uses 'Authorization: token TOKEN' header format, not Bearer"
+  confidence: 0.9
+  tags: [git, api, gitea]
+  source_count: 8
+  first_seen: "2026-03-28"
+  last_confirmed: "2026-04-12"
+
+- id: global:tool-quirk:003
+  fact: "Gitea Issues API type=issues param does NOT filter PRs - use truthiness check on pull_request field"
+  confidence: 0.95
+  tags: [gitea, api, issues, pr]
+  source_count: 6
+  first_seen: "2026-04-01"
+  last_confirmed: "2026-04-13"
+
+## Paths & Environment
+
+- id: global:tool-quirk:004
+  fact: "~/.hermes is the default hermes home - check get_hermes_home() not the path literal"
+  confidence: 0.9
+  tags: [paths, hermes, env]
+  source_count: 10
+  first_seen: "2026-03-30"
+  last_confirmed: "2026-04-13"
+  related: [hermes-agent:pitfall:005]
+
+- id: global:tool-quirk:005
+  fact: "Ansible vault-encrypted vars in YAML require vault_inline_vars plugin - standard ansible-vault fails"
+  confidence: 0.85
+  tags: [ansible, vault, config]
+  source_count: 3
+  first_seen: "2026-04-02"
+
+## Model & Inference
+
+- id: global:tool-quirk:006
+  fact: "mimo-v2-pro via Nous Research is the default model - don't assume Anthropic is available"
+  confidence: 0.95
+  tags: [model, provider, nous, default]
+  source_count: 15
+  first_seen: "2026-03-25"
+  last_confirmed: "2026-04-13"
+
+- id: global:tool-quirk:007
+  fact: "Kill + restart with 'hermes chat' preserves old model state - NEVER use --resume"
+  confidence: 0.95
+  tags: [hermes, model, restart, session]
+  source_count: 8
+  first_seen: "2026-03-29"
+  last_confirmed: "2026-04-12"
--- a/knowledge/repos/hermes-agent.yaml
+++ b/knowledge/repos/hermes-agent.yaml
@@ -0,0 +1,82 @@
+---
+domain: hermes-agent
+category: pitfall
+version: 1
+last_updated: "2026-04-13"
+---
+
+# Pitfalls (hermes-agent)
+
+Things that go wrong in this repo if you don't know the traps.
+
+## Cron & Deployment
+
+- id: hermes-agent:pitfall:001
+  fact: "deploy-crons.py leaves jobs in mixed model format - some have provider/model, some just model"
+  confidence: 0.95
+  tags: [cron, deploy, model, config]
+  source_count: 5
+  first_seen: "2026-04-08"
+  last_confirmed: "2026-04-13"
+  related: [hermes-agent:pitfall:002, hermes-agent:pitfall:003]
+
+- id: hermes-agent:pitfall:002
+  fact: "deploy-crons.py --deploy doesn't set legacy skill field from skills list, breaking older jobs"
+  confidence: 0.9
+  tags: [cron, deploy, skills]
+  source_count: 3
+  first_seen: "2026-04-09"
+  last_confirmed: "2026-04-13"
+  related: [hermes-agent:pitfall:001]
+
+- id: hermes-agent:pitfall:003
+  fact: "Cron jobs with blank fallback_model fields trigger spurious gateway warnings"
+  confidence: 0.9
+  tags: [cron, model, fallback]
+  source_count: 4
+  first_seen: "2026-04-07"
+  last_confirmed: "2026-04-12"
+  related: [hermes-agent:pitfall:001]
+
+- id: hermes-agent:pitfall:004
+  fact: "model-watchdog.py checks first provider line, not model.provider - causes false drift alarms"
+  confidence: 0.9
+  tags: [watchdog, model, config]
+  source_count: 3
+  first_seen: "2026-04-08"
+  last_confirmed: "2026-04-13"
+
+## Path & Environment
+
+- id: hermes-agent:pitfall:005
+  fact: "10+ files read HERMES_HOME directly instead of get_hermes_home() - breaks on custom paths"
+  confidence: 0.85
+  tags: [paths, env, hermes-home]
+  source_count: 6
+  first_seen: "2026-04-06"
+  last_confirmed: "2026-04-12"
+  related: [global:pitfall:002]
+
+- id: hermes-agent:pitfall:006
+  fact: "get_hermes_home() doesn't expand tilde when HERMES_HOME=~/... is set"
+  confidence: 0.8
+  tags: [paths, env, bug]
+  source_count: 2
+  first_seen: "2026-04-05"
+
+## SSH & Dispatch
+
+- id: hermes-agent:pitfall:007
+  fact: "vps-agent-dispatch reports OK while remote hermes binary path is broken"
+  confidence: 0.9
+  tags: [ssh, dispatch, vps]
+  source_count: 4
+  first_seen: "2026-04-07"
+  last_confirmed: "2026-04-11"
+
+- id: hermes-agent:pitfall:008
+  fact: "nightwatch-health-monitor SSH check fails on cloud-model-only deployments"
+  confidence: 0.85
+  tags: [ssh, health, cloud]
+  source_count: 2
+  first_seen: "2026-04-10"
--- a/knowledge/repos/the-nexus.yaml
+++ b/knowledge/repos/the-nexus.yaml
@@ -0,0 +1,65 @@
+---
+domain: the-nexus
+category: pitfall
+version: 1
+last_updated: "2026-04-13"
+---
+
+# Pitfalls (the-nexus)
+
+Things that go wrong in this repo if you don't know the traps.
+
+## Git & Merging
+
+- id: the-nexus:pitfall:001
+  fact: "Merges fail with HTTP 405 due to branch protection - must use merge API with 1 approval"
+  confidence: 0.95
+  tags: [git, merge, branch-protection, gitea]
+  source_count: 12
+  first_seen: "2026-04-05"
+  last_confirmed: "2026-04-13"
+  related: [global:pitfall:001]
+
+- id: the-nexus:pitfall:002
+  fact: "ThreadingHTTPServer required for multi-user bridge - standard HTTPServer blocks on concurrent requests"
+  confidence: 0.95
+  tags: [server, concurrency, bridge]
+  source_count: 5
+  first_seen: "2026-04-10"
+  last_confirmed: "2026-04-13"
+
+- id: the-nexus:pitfall:003
+  fact: "ChatLog.log() crashes on message persistence when index.html has orphaned button tags"
+  confidence: 0.9
+  tags: [html, crash, chatlog]
+  source_count: 3
+  first_seen: "2026-04-12"
+  last_confirmed: "2026-04-13"
+
+## Three.js & Performance
+
+- id: the-nexus:pitfall:004
+  fact: "Three.js LOD not implemented - local hardware struggles with full scene without texture optimization"
+  confidence: 0.85
+  tags: [threejs, performance, lod]
+  source_count: 4
+  first_seen: "2026-04-09"
+  last_confirmed: "2026-04-13"
+
+- id: the-nexus:pitfall:005
+  fact: "Duplicate content blocks appear in index.html when PR merges conflict silently"
+  confidence: 0.8
+  tags: [html, merge-conflict, duplicate]
+  source_count: 3
+  first_seen: "2026-04-11"
+  last_confirmed: "2026-04-13"
+
+## Deployment
+
+- id: the-nexus:pitfall:006
+  fact: "Unified HTTP + WebSocket server required for proper URL deployment - separate servers break CORS"
+  confidence: 0.9
+  tags: [deploy, websocket, http, cors]
+  source_count: 4
+  first_seen: "2026-04-10"
+  last_confirmed: "2026-04-13"
--- a/metrics/dashboard.md
+++ b/metrics/dashboard.md
@@ -0,0 +1,61 @@
+# Compounding Intelligence Metrics
+**Generated:** 2026-04-14T18:12:26.469085+00:00
+
+## knowledge_velocity
+New facts extracted per day. Higher = compounding loop working.
+
+**Value:** 1.61  |  **7d trend:** N/A --- (unknown)
+
+- total_facts: 29
+- period_days: 18
+- new_facts: 29
+
+## knowledge_coverage
+Percentage of domains/repos with 10+ facts. Measures breadth.
+
+**Value:** 0.333  |  **7d trend:** N/A --- (unknown)
+
+- covered_domains: 1
+- total_domains: 3
+
+## hit_rate
+Percentage of sessions referencing bootstrapped knowledge.
+
+**Value:** 0.676  |  **7d trend:** N/A --- (unknown)
+
+- hit_sessions: 8058
+- total_sessions: 11922
+
+## error_recurrence
+Ratio of recurring errors. Lower = fleet learning from mistakes.
+
+**Value:** 0.17  |  **7d trend:** N/A --- (unknown)
+
+- unique_errors: 53615
+- recurring_errors: 9093
+
+## task_completion
+Percentage of sessions ending with successful completion.
+
+**Value:** 0.452  |  **7d trend:** N/A --- (unknown)
+
+- normal_end_rate: 0.56
+- completed: 5385
+- total: 11922
+
+## first_try_success
+Percentage of sessions completed without backtracking.
+
+**Value:** 0.818  |  **7d trend:** N/A --- (unknown)
+
+- avg_tool_msg_ratio: 0.392
+- sampled: 5923
+
+## knowledge_age
+Freshness of knowledge store. 1.0 = all fresh, 0.0 = all stale.
+
+**Value:** 0.973  |  **7d trend:** N/A --- (unknown)
+
+- avg_age_days: 2.4
+- stale_facts: 0
+- total_facts: 29
--- a/metrics/latest_snapshot.json
+++ b/metrics/latest_snapshot.json
@@ -0,0 +1,127 @@
+{
+  "generated_at": "2026-04-14T18:12:26.469085+00:00",
+  "knowledge_velocity": {
+    "value": 1.61,
+    "total_facts": 29,
+    "period_days": 18,
+    "new_facts": 29
+  },
+  "knowledge_coverage": {
+    "value": 0.333,
+    "covered_domains": 1,
+    "total_domains": 3,
+    "domain_details": {
+      "global": 15,
+      "hermes-agent": 8,
+      "the-nexus": 6
+    }
+  },
+  "hit_rate": {
+    "value": 0.676,
+    "hit_sessions": 8058,
+    "total_sessions": 11922
+  },
+  "error_recurrence": {
+    "value": 0.17,
+    "unique_errors": 53615,
+    "recurring_errors": 9093,
+    "top_errors": [
+      {
+        "error": "s, report the error details.",
+        "sessions": 1185
+      },
+      {
+        "error": "\": \"traceback (most recent call last):\\n file \\\"/private/var/folders/9k/v07xkpp",
+        "sessions": 695
+      },
+      {
+        "error": "\", \"output\": \"\\n--- stderr ---\\ntraceback (most recent call last):\\n file \\\"/pr",
+        "sessions": 684
+      },
+      {
+        "error": "ures \u2192 file an issue with the traceback and tag [bug]",
+        "sessions": 320
+      },
+      {
+        "error": "s you encounter \u2192 file an issue with reproduction steps",
+        "sessions": 320
+      },
+      {
+        "error": "s, file a [bug] issue first.",
+        "sessions": 320
+      },
+      {
+        "error": ", fix the code.",
+        "sessions": 314
+      },
+      {
+        "error": "fix it before doing anything else.",
+        "sessions": 313
+      },
+      {
+        "error": "ures \u2192 add a review comment explaining what's wrong",
+        "sessions": 303
+      },
+      {
+        "error": "ures \u2014 they're your roadmap for guardrails",
+        "sessions": 303
+      }
+    ]
+  },
+  "task_completion": {
+    "value": 0.452,
+    "normal_end_rate": 0.56,
+    "completed": 5385,
+    "total": 11922,
+    "breakdown": {
+      "cron_complete": 5354,
+      "unknown": 5248,
+      "compression": 1092,
+      "cli_close": 197,
+      "session_reset": 31
+    }
+  },
+  "first_try_success": {
+    "value": 0.818,
+    "avg_tool_msg_ratio": 0.392,
+    "sampled": 5923,
+    "interpretation": "Higher value = fewer backtracks = better first-try success"
+  },
+  "knowledge_age": {
+    "value": 0.973,
+    "avg_age_days": 2.4,
+    "stale_facts": 0,
+    "total_facts": 29,
+    "interpretation": "1.0 = all facts fresh. 0.0 = all facts 90+ days old"
+  },
+  "trend_7d": {
+    "knowledge_velocity": {
+      "delta": "N/A",
+      "direction": "unknown"
+    },
+    "knowledge_coverage": {
+      "delta": "N/A",
+      "direction": "unknown"
+    },
+    "hit_rate": {
+      "delta": "N/A",
+      "direction": "unknown"
+    },
+    "error_recurrence": {
+      "delta": "N/A",
+      "direction": "unknown"
+    },
+    "task_completion": {
+      "delta": "N/A",
+      "direction": "unknown"
+    },
+    "first_try_success": {
+      "delta": "N/A",
+      "direction": "unknown"
+    },
+    "knowledge_age": {
+      "delta": "N/A",
+      "direction": "unknown"
+    }
+  }
+}
--- a/scripts/harvester.py
+++ b/scripts/harvester.py
@@ -1,447 +0,0 @@
-#!/usr/bin/env python3
-"""
-harvester.py — Extract durable knowledge from Hermes session transcripts.
-
-Combines session_reader + extraction prompt + LLM inference to pull
-facts, pitfalls, patterns, and tool quirks from finished sessions.
-
-Usage:
-    python3 harvester.py --session ~/.hermes/sessions/session_xxx.jsonl --output knowledge/
-    python3 harvester.py --batch --since 2026-04-01 --limit 100
-    python3 harvester.py --session session.jsonl --dry-run  # Preview without writing
-"""
-
-import argparse
-import json
-import os
-import sys
-import time
-import hashlib
-from datetime import datetime, timezone
-from pathlib import Path
-from typing import Optional
-
-# Add scripts dir to path for sibling imports
-SCRIPT_DIR = Path(__file__).parent.absolute()
-sys.path.insert(0, str(SCRIPT_DIR))
-
-from session_reader import read_session, extract_conversation, truncate_for_context, messages_to_text
-
-# --- Configuration ---
-
-DEFAULT_API_BASE = os.environ.get("HARVESTER_API_BASE", "https://api.nousresearch.com/v1")
-DEFAULT_API_KEY = os.environ.get("HARVESTER_API_KEY", "")
-DEFAULT_MODEL = os.environ.get("HARVESTER_MODEL", "xiaomi/mimo-v2-pro")
-KNOWLEDGE_DIR = os.environ.get("HARVESTER_KNOWLEDGE_DIR", "knowledge")
-PROMPT_PATH = os.environ.get("HARVESTER_PROMPT_PATH", str(SCRIPT_DIR.parent / "templates" / "harvest-prompt.md"))
-
-# Where to look for API keys if not set via env
-API_KEY_PATHS = [
-    os.path.expanduser("~/.config/nous/key"),
-    os.path.expanduser("~/.hermes/keymaxxing/active/minimax.key"),
-    os.path.expanduser("~/.config/openrouter/key"),
-]
-
-
-def find_api_key() -> str:
-    """Find API key from common locations."""
-    for path in API_KEY_PATHS:
-        if os.path.exists(path):
-            with open(path) as f:
-                key = f.read().strip()
-                if key:
-                    return key
-    return ""
-
-
-def load_extraction_prompt() -> str:
-    """Load the extraction prompt template."""
-    path = Path(PROMPT_PATH)
-    if not path.exists():
-        print(f"ERROR: Extraction prompt not found at {path}", file=sys.stderr)
-        print("Expected templates/harvest-prompt.md from issue #7", file=sys.stderr)
-        sys.exit(1)
-    return path.read_text(encoding='utf-8')
-
-
-def call_llm(prompt: str, transcript: str, api_base: str, api_key: str, model: str) -> Optional[list[dict]]:
-    """Call the LLM API to extract knowledge from a transcript."""
-    import urllib.request
-    
-    messages = [
-        {"role": "system", "content": prompt},
-        {"role": "user", "content": f"Extract knowledge from this session transcript:\n\n{transcript}"}
-    ]
-    
-    payload = json.dumps({
-        "model": model,
-        "messages": messages,
-        "temperature": 0.1,  # Low temp for consistent extraction
-        "max_tokens": 4096
-    }).encode('utf-8')
-    
-    req = urllib.request.Request(
-        f"{api_base}/chat/completions",
-        data=payload,
-        headers={
-            "Authorization": f"Bearer {api_key}",
-            "Content-Type": "application/json"
-        },
-        method="POST"
-    )
-    
-    try:
-        with urllib.request.urlopen(req, timeout=60) as resp:
-            result = json.loads(resp.read().decode('utf-8'))
-            content = result["choices"][0]["message"]["content"]
-            return parse_extraction_response(content)
-    except Exception as e:
-        print(f"ERROR: LLM API call failed: {e}", file=sys.stderr)
-        return None
-
-
-def parse_extraction_response(content: str) -> Optional[list[dict]]:
-    """Parse the LLM response to extract knowledge items.
-    
-    Handles various response formats: raw JSON, markdown-wrapped JSON, etc.
-    """
-    # Try direct JSON parse first
-    try:
-        data = json.loads(content)
-        if isinstance(data, dict) and 'knowledge' in data:
-            return data['knowledge']
-        if isinstance(data, list):
-            return data
-    except json.JSONDecodeError:
-        pass
-    
-    # Try extracting JSON from markdown code blocks
-    import re
-    json_match = re.search(r'```(?:json)?\s*({.*?})\s*```', content, re.DOTALL)
-    if json_match:
-        try:
-            data = json.loads(json_match.group(1))
-            if isinstance(data, dict) and 'knowledge' in data:
-                return data['knowledge']
-            if isinstance(data, list):
-                return data
-        except json.JSONDecodeError:
-            pass
-    
-    # Try finding any JSON object with knowledge array
-    json_match = re.search(r'({[^{}]*"knowledge"[^{}]*[[sS]*?][^{}]*})', content)
-    if json_match:
-        try:
-            data = json.loads(json_match.group(1))
-            return data.get('knowledge', [])
-        except json.JSONDecodeError:
-            pass
-    
-    print(f"WARNING: Could not parse LLM response as JSON", file=sys.stderr)
-    print(f"Response preview: {content[:500]}", file=sys.stderr)
-    return None
-
-
-def load_existing_knowledge(knowledge_dir: str) -> dict:
-    """Load the existing knowledge index."""
-    index_path = Path(knowledge_dir) / "index.json"
-    if not index_path.exists():
-        return {"version": 1, "last_updated": "", "total_facts": 0, "facts": []}
-    
-    try:
-        with open(index_path, 'r', encoding='utf-8') as f:
-            return json.load(f)
-    except (json.JSONDecodeError, IOError) as e:
-        print(f"WARNING: Could not load knowledge index: {e}", file=sys.stderr)
-        return {"version": 1, "last_updated": "", "total_facts": 0, "facts": []}
-
-
-def fact_fingerprint(fact: dict) -> str:
-    """Generate a deduplication fingerprint for a fact.
-    
-    Uses the fact text normalized (lowercase, stripped) as the key.
-    Similar facts will have similar fingerprints.
-    """
-    text = fact.get('fact', '').lower().strip()
-    # Normalize whitespace
-    text = ' '.join(text.split())
-    return hashlib.md5(text.encode('utf-8')).hexdigest()
-
-
-def deduplicate(new_facts: list[dict], existing: list[dict], similarity_threshold: float = 0.8) -> list[dict]:
-    """Remove duplicate facts from new_facts that already exist in the knowledge store.
-    
-    Uses fingerprint matching for exact dedup and simple overlap check for near-dupes.
-    """
-    existing_fingerprints = set()
-    existing_texts = []
-    for f in existing:
-        fp = fact_fingerprint(f)
-        existing_fingerprints.add(fp)
-        existing_texts.append(f.get('fact', '').lower().strip())
-    
-    unique = []
-    for fact in new_facts:
-        fp = fact_fingerprint(fact)
-        if fp in existing_fingerprints:
-            continue
-        
-        # Check for near-duplicates using simple word overlap
-        fact_words = set(fact.get('fact', '').lower().split())
-        is_dup = False
-        for existing_text in existing_texts:
-            existing_words = set(existing_text.split())
-            if not fact_words or not existing_words:
-                continue
-            overlap = len(fact_words & existing_words) / max(len(fact_words | existing_words), 1)
-            if overlap >= similarity_threshold:
-                is_dup = True
-                break
-        
-        if not is_dup:
-            unique.append(fact)
-            existing_fingerprints.add(fp)
-            existing_texts.append(fact.get('fact', '').lower().strip())
-    
-    return unique
-
-
-def validate_fact(fact: dict) -> bool:
-    """Validate a single knowledge item has required fields."""
-    required = ['fact', 'category', 'repo', 'confidence']
-    for field in required:
-        if field not in fact:
-            return False
-    
-    if not isinstance(fact['fact'], str) or not fact['fact'].strip():
-        return False
-    
-    valid_categories = ['fact', 'pitfall', 'pattern', 'tool-quirk', 'question']
-    if fact['category'] not in valid_categories:
-        return False
-    
-    if not isinstance(fact.get('confidence', 0), (int, float)):
-        return False
-    
-    if not (0.0 <= fact['confidence'] <= 1.0):
-        return False
-    
-    return True
-
-
-def write_knowledge(index: dict, new_facts: list[dict], knowledge_dir: str, source_session: str = ""):
-    """Write new facts to the knowledge store."""
-    kdir = Path(knowledge_dir)
-    kdir.mkdir(parents=True, exist_ok=True)
-    
-    # Add source tracking to each fact
-    for fact in new_facts:
-        fact['source_session'] = source_session
-        fact['harvested_at'] = datetime.now(timezone.utc).isoformat()
-    
-    # Update index
-    index['facts'].extend(new_facts)
-    index['total_facts'] = len(index['facts'])
-    index['last_updated'] = datetime.now(timezone.utc).isoformat()
-    
-    # Write index
-    index_path = kdir / "index.json"
-    with open(index_path, 'w', encoding='utf-8') as f:
-        json.dump(index, f, indent=2, ensure_ascii=False)
-    
-    # Also write per-repo markdown files for human reading
-    repos = {}
-    for fact in new_facts:
-        repo = fact.get('repo', 'global')
-        repos.setdefault(repo, []).append(fact)
-    
-    for repo, facts in repos.items():
-        if repo == 'global':
-            md_path = kdir / "global" / "harvested.md"
-        else:
-            md_path = kdir / "repos" / f"{repo}.md"
-        
-        md_path.parent.mkdir(parents=True, exist_ok=True)
-        
-        # Append to existing or create new
-        mode = 'a' if md_path.exists() else 'w'
-        with open(md_path, mode, encoding='utf-8') as f:
-            if mode == 'w':
-                f.write(f"# Knowledge: {repo}\n\n")
-            f.write(f"## Harvested {datetime.now(timezone.utc).strftime('%Y-%m-%d %H:%M')}\n\n")
-            for fact in facts:
-                icon = {'fact': '📋', 'pitfall': '⚠️', 'pattern': '🔄', 'tool-quirk': '🔧', 'question': '❓'}.get(fact['category'], '•')
-                f.write(f"- {icon} **{fact['category']}** (conf: {fact['confidence']:.1f}): {fact['fact']}\n")
-            f.write("\n")
-
-
-def harvest_session(session_path: str, knowledge_dir: str, api_base: str, api_key: str, 
-                    model: str, dry_run: bool = False, min_confidence: float = 0.3) -> dict:
-    """Harvest knowledge from a single session.
-    
-    Returns: dict with stats (facts_found, facts_new, facts_dup, elapsed_seconds, error)
-    """
-    start_time = time.time()
-    stats = {
-        'session': session_path,
-        'facts_found': 0,
-        'facts_new': 0,
-        'facts_dup': 0,
-        'elapsed_seconds': 0,
-        'error': None
-    }
-    
-    try:
-        # 1. Read session
-        messages = read_session(session_path)
-        if not messages:
-            stats['error'] = "Empty session file"
-            return stats
-        
-        # 2. Extract conversation
-        conv = extract_conversation(messages)
-        if not conv:
-            stats['error'] = "No conversation turns found"
-            return stats
-        
-        # 3. Truncate for context window
-        truncated = truncate_for_context(conv, head=50, tail=50)
-        transcript = messages_to_text(truncated)
-        
-        # 4. Load extraction prompt
-        prompt = load_extraction_prompt()
-        
-        # 5. Call LLM
-        raw_facts = call_llm(prompt, transcript, api_base, api_key, model)
-        if raw_facts is None:
-            stats['error'] = "LLM extraction failed"
-            return stats
-        
-        # 6. Validate
-        valid_facts = [f for f in raw_facts if validate_fact(f) and f.get('confidence', 0) >= min_confidence]
-        stats['facts_found'] = len(valid_facts)
-        
-        # 7. Deduplicate
-        existing_index = load_existing_knowledge(knowledge_dir)
-        existing_facts = existing_index.get('facts', [])
-        new_facts = deduplicate(valid_facts, existing_facts)
-        stats['facts_new'] = len(new_facts)
-        stats['facts_dup'] = len(valid_facts) - len(new_facts)
-        
-        # 8. Write (unless dry run)
-        if new_facts and not dry_run:
-            write_knowledge(existing_index, new_facts, knowledge_dir, source_session=session_path)
-        
-        stats['elapsed_seconds'] = round(time.time() - start_time, 2)
-        return stats
-        
-    except Exception as e:
-        stats['error'] = str(e)
-        stats['elapsed_seconds'] = round(time.time() - start_time, 2)
-        return stats
-
-
-def batch_harvest(sessions_dir: str, knowledge_dir: str, api_base: str, api_key: str,
-                  model: str, since: str = "", limit: int = 0, dry_run: bool = False) -> list[dict]:
-    """Harvest knowledge from multiple sessions in batch."""
-    sessions_path = Path(sessions_dir)
-    if not sessions_path.is_dir():
-        print(f"ERROR: Sessions directory not found: {sessions_dir}", file=sys.stderr)
-        return []
-    
-    # Find session files
-    session_files = sorted(sessions_path.glob("*.jsonl"), reverse=True)  # Newest first
-    
-    # Filter by date if --since provided
-    if since:
-        since_dt = datetime.fromisoformat(since.replace('Z', '+00:00'))
-        filtered = []
-        for sf in session_files:
-            # Try to parse timestamp from filename (common format: session_YYYYMMDD_HHMMSS_hash.jsonl)
-            try:
-                parts = sf.stem.split('_')
-                if len(parts) >= 3:
-                    date_str = parts[1]
-                    file_dt = datetime.strptime(date_str, '%Y%m%d').replace(tzinfo=timezone.utc)
-                    if file_dt >= since_dt:
-                        filtered.append(sf)
-            except (ValueError, IndexError):
-                # If we can't parse the date, include the file (be permissive)
-                filtered.append(sf)
-        session_files = filtered
-    
-    # Apply limit
-    if limit > 0:
-        session_files = session_files[:limit]
-    
-    print(f"Harvesting {len(session_files)} sessions...")
-    
-    results = []
-    for i, sf in enumerate(session_files, 1):
-        print(f"[{i}/{len(session_files)}] {sf.name}...", end=" ", flush=True)
-        stats = harvest_session(str(sf), knowledge_dir, api_base, api_key, model, dry_run)
-        if stats['error']:
-            print(f"ERROR: {stats['error']}")
-        else:
-            print(f"{stats['facts_new']} new, {stats['facts_dup']} dup ({stats['elapsed_seconds']}s)")
-        results.append(stats)
-    
-    return results
-
-
-def main():
-    parser = argparse.ArgumentParser(description="Harvest knowledge from session transcripts")
-    parser.add_argument('--session', help='Path to a single session JSONL file')
-    parser.add_argument('--batch', action='store_true', help='Batch mode: process multiple sessions')
-    parser.add_argument('--sessions-dir', default=os.path.expanduser('~/.hermes/sessions'), 
-                        help='Directory containing session files (default: ~/.hermes/sessions)')
-    parser.add_argument('--output', default='knowledge', help='Output directory for knowledge store')
-    parser.add_argument('--since', default='', help='Only process sessions after this date (YYYY-MM-DD)')
-    parser.add_argument('--limit', type=int, default=0, help='Max sessions to process (0=unlimited)')
-    parser.add_argument('--api-base', default=DEFAULT_API_BASE, help='LLM API base URL')
-    parser.add_argument('--api-key', default='', help='LLM API key (or set HARVESTER_API_KEY)')
-    parser.add_argument('--model', default=DEFAULT_MODEL, help='Model to use for extraction')
-    parser.add_argument('--dry-run', action='store_true', help='Preview without writing to knowledge store')
-    parser.add_argument('--min-confidence', type=float, default=0.3, help='Minimum confidence threshold')
-    
-    args = parser.parse_args()
-    
-    # Resolve API key
-    api_key = args.api_key or DEFAULT_API_KEY or find_api_key()
-    if not api_key:
-        print("ERROR: No API key found. Set HARVESTER_API_KEY or store in one of:", file=sys.stderr)
-        for p in API_KEY_PATHS:
-            print(f"  {p}", file=sys.stderr)
-        sys.exit(1)
-    
-    # Resolve knowledge directory
-    knowledge_dir = args.output
-    if not os.path.isabs(knowledge_dir):
-        knowledge_dir = os.path.join(SCRIPT_DIR.parent, knowledge_dir)
-    
-    if args.session:
-        # Single session mode
-        stats = harvest_session(
-            args.session, knowledge_dir, args.api_base, api_key, args.model,
-            dry_run=args.dry_run, min_confidence=args.min_confidence
-        )
-        print(json.dumps(stats, indent=2))
-        if stats['error']:
-            sys.exit(1)
-    elif args.batch:
-        # Batch mode
-        results = batch_harvest(
-            args.sessions_dir, knowledge_dir, args.api_base, api_key, args.model,
-            since=args.since, limit=args.limit, dry_run=args.dry_run
-        )
-        total_new = sum(r['facts_new'] for r in results)
-        total_dup = sum(r['facts_dup'] for r in results)
-        errors = sum(1 for r in results if r['error'])
-        print(f"\nDone: {total_new} new facts, {total_dup} duplicates, {errors} errors")
-    else:
-        parser.print_help()
-        sys.exit(1)
-
-
-if __name__ == '__main__':
-    main()
--- a/scripts/measurer.py
+++ b/scripts/measurer.py
@@ -0,0 +1,403 @@
+#!/usr/bin/env python3
+"""
+Compounding Intelligence Metrics Engine.
+
+Computes 7 metrics that prove whether the knowledge compounding loop is working:
+  1. Knowledge velocity -- new facts per day
+  2. Knowledge coverage -- % of domains with >10 facts
+  3. Hit rate -- % of sessions referencing bootstrap knowledge
+  4. Error recurrence -- same errors across sessions (should decrease)
+  5. Task completion -- % of sessions ending successfully
+  6. First-try success -- actions without backtracking
+  7. Knowledge age -- staleness of facts
+
+Usage:
+  python3 measurer.py                        # All metrics, all time
+  python3 measurer.py --since 2026-04-01     # Time range
+  python3 measurer.py --repo the-nexus       # Per-repo metrics
+  python3 measurer.py --format json          # JSON output (default)
+  python3 measurer.py --format markdown      # Human-readable
+  python3 measurer.py --knowledge-dir ./knowledge  # Custom knowledge path
+  python3 measurer.py --db ~/.hermes/state.db      # Custom DB path
+
+Data sources:
+  - knowledge/index.json  -- fact index
+  - knowledge/            -- YAML fact files for coverage
+  - ~/.hermes/state.db    -- session/message metadata
+"""
+
+import argparse
+import json
+import os
+import re
+import sqlite3
+import sys
+from collections import Counter, defaultdict
+from datetime import datetime, timedelta, timezone
+from pathlib import Path
+from typing import Any
+
+
+# --- Defaults ---
+
+DEFAULT_KNOWLEDGE_DIR = Path(__file__).parent.parent / "knowledge"
+DEFAULT_DB_PATH = Path.home() / ".hermes" / "state.db"
+SEVEN_DAYS = timedelta(days=7)
+
+
+# --- Knowledge Store ---
+
+def load_facts(knowledge_dir):
+    index_path = knowledge_dir / "index.json"
+    if not index_path.exists():
+        return []
+    with open(index_path) as f:
+        data = json.load(f)
+    return data.get("facts", [])
+
+
+def count_yaml_facts(knowledge_dir):
+    domain_counts = {}
+    for subdir in ["repos", "global", "agents"]:
+        dirpath = knowledge_dir / subdir
+        if not dirpath.exists():
+            continue
+        for yaml_file in dirpath.glob("*.yaml"):
+            count = 0
+            try:
+                content = yaml_file.read_text()
+                count = len(re.findall(r"^\s*-\s*id:", content, re.MULTILINE))
+            except Exception:
+                pass
+            domain = yaml_file.stem
+            domain_counts[domain] = domain_counts.get(domain, 0) + count
+    return domain_counts
+
+
+# --- Session Database ---
+
+def open_db(db_path):
+    if not db_path.exists():
+        print(f"WARNING: Database not found at {db_path}", file=sys.stderr)
+        return None
+    conn = sqlite3.connect(str(db_path))
+    conn.row_factory = sqlite3.Row
+    return conn
+
+
+def query_sessions(conn, since=None, repo=None):
+    if conn is None:
+        return []
+    query = "SELECT id, started_at, ended_at, end_reason, message_count, tool_call_count, model FROM sessions WHERE 1=1"
+    params = []
+    if since:
+        since_ts = datetime.fromisoformat(since).replace(tzinfo=timezone.utc).timestamp()
+        query += " AND started_at >= ?"
+        params.append(since_ts)
+    query += " ORDER BY started_at ASC"
+    cur = conn.execute(query, params)
+    return [dict(row) for row in cur.fetchall()]
+
+
+def query_messages(conn, session_ids=None, since_ts=None):
+    if conn is None:
+        return []
+    query = "SELECT m.session_id, m.role, m.content, m.tool_name, m.timestamp FROM messages m WHERE 1=1"
+    params = []
+    if since_ts:
+        query += " AND m.timestamp >= ?"
+        params.append(since_ts)
+    if session_ids:
+        placeholders = ",".join("?" for _ in session_ids)
+        query += f" AND m.session_id IN ({placeholders})"
+        params.extend(session_ids)
+    cur = conn.execute(query, params)
+    return [dict(row) for row in cur.fetchall()]
+
+
+# --- Metric Computations ---
+
+def compute_knowledge_velocity(facts, since=None):
+    if not facts:
+        return {"value": 0.0, "total_facts": 0, "period_days": 0, "new_facts": 0}
+    dates = []
+    for f in facts:
+        d = f.get("first_seen") or f.get("created")
+        if d:
+            try:
+                dt = datetime.fromisoformat(d.replace("Z", "+00:00"))
+                if dt.tzinfo is None:
+                    dt = dt.replace(tzinfo=timezone.utc)
+                dates.append(dt)
+            except (ValueError, AttributeError):
+                pass
+    if not dates:
+        return {"value": 0.0, "total_facts": len(facts), "period_days": 0, "new_facts": 0}
+    if since:
+        cutoff = datetime.fromisoformat(since).replace(tzinfo=timezone.utc)
+        dates = [d for d in dates if d >= cutoff]
+    if not dates:
+        return {"value": 0.0, "total_facts": len(facts), "period_days": 0, "new_facts": 0}
+    earliest = min(dates)
+    latest = max(dates)
+    period_days = max((latest - earliest).days, 1)
+    return {"value": round(len(dates) / period_days, 2), "total_facts": len(facts), "period_days": period_days, "new_facts": len(dates)}
+
+
+def compute_knowledge_coverage(facts, yaml_counts):
+    domain_fact_counts = defaultdict(int)
+    for f in facts:
+        domain = f.get("domain", "unknown")
+        domain_fact_counts[domain] += 1
+    for domain, count in yaml_counts.items():
+        domain_fact_counts[domain] = max(domain_fact_counts[domain], count)
+    total_domains = len(domain_fact_counts)
+    if total_domains == 0:
+        return {"value": 0.0, "covered_domains": 0, "total_domains": 0, "domain_details": {}}
+    covered = sum(1 for c in domain_fact_counts.values() if c >= 10)
+    return {"value": round(covered / total_domains, 3), "covered_domains": covered, "total_domains": total_domains, "domain_details": dict(sorted(domain_fact_counts.items(), key=lambda x: -x[1])[:20])}
+
+
+def compute_hit_rate(sessions, messages, facts):
+    if not sessions or not facts:
+        return {"value": 0.0, "hit_sessions": 0, "total_sessions": len(sessions)}
+    fact_fragments = set()
+    for f in facts:
+        text = f.get("fact", "").lower().strip()
+        if len(text) > 10:
+            fact_fragments.add(text)
+        words = re.findall(r'\w{4,}', text)
+        for w in words:
+            fact_fragments.add(w)
+    if not fact_fragments:
+        return {"value": 0.0, "hit_sessions": 0, "total_sessions": len(sessions)}
+    session_messages = defaultdict(list)
+    for m in messages:
+        content = (m.get("content") or "").lower()
+        if content:
+            session_messages[m["session_id"]].append(content)
+    hit_sessions = 0
+    for session in sessions:
+        sid = session["id"]
+        all_content = " ".join(session_messages.get(sid, []))
+        if any(frag in all_content for frag in fact_fragments):
+            hit_sessions += 1
+    return {"value": round(hit_sessions / len(sessions), 3) if sessions else 0.0, "hit_sessions": hit_sessions, "total_sessions": len(sessions)}
+
+
+def compute_error_recurrence(messages):
+    if not messages:
+        return {"value": 0.0, "unique_errors": 0, "recurring_errors": 0, "top_errors": []}
+    error_pattern = re.compile(r'(?:error|Error|ERROR|failed|FAIL|exception|Exception)[:\s]*(.{10,80})', re.IGNORECASE)
+    error_to_sessions = defaultdict(set)
+    for m in messages:
+        content = m.get("content") or ""
+        if not content:
+            continue
+        for match in error_pattern.finditer(content):
+            sig = match.group(1).strip().lower()
+            sig = re.sub(r'\s+', ' ', sig)
+            if len(sig) > 5:
+                error_to_sessions[sig].add(m["session_id"])
+    if not error_to_sessions:
+        return {"value": 0.0, "unique_errors": 0, "recurring_errors": 0, "top_errors": []}
+    recurring = {e: s for e, s in error_to_sessions.items() if len(s) > 1}
+    total_errors = len(error_to_sessions)
+    recurring_count = len(recurring)
+    top = sorted(recurring.items(), key=lambda x: -len(x[1]))[:10]
+    return {"value": round(recurring_count / total_errors, 3) if total_errors else 0.0, "unique_errors": total_errors, "recurring_errors": recurring_count, "top_errors": [{"error": e, "sessions": len(s)} for e, s in top]}
+
+
+def compute_task_completion(sessions):
+    if not sessions:
+        return {"value": 0.0, "completed": 0, "total": 0, "breakdown": {}}
+    breakdown = Counter()
+    for s in sessions:
+        reason = s.get("end_reason") or "unknown"
+        breakdown[reason] += 1
+    completed = breakdown.get("cron_complete", 0) + breakdown.get("session_reset", 0)
+    normal_endings = completed + breakdown.get("cli_close", 0) + breakdown.get("compression", 0)
+    return {"value": round(completed / len(sessions), 3) if sessions else 0.0, "normal_end_rate": round(normal_endings / len(sessions), 3) if sessions else 0.0, "completed": completed, "total": len(sessions), "breakdown": dict(breakdown.most_common())}
+
+
+def compute_first_try_success(sessions):
+    if not sessions:
+        return {"value": 0.0, "avg_tool_msg_ratio": 0.0, "sampled": 0}
+    ratios = []
+    for s in sessions:
+        msgs = s.get("message_count", 0) or 0
+        tools = s.get("tool_call_count", 0) or 0
+        if msgs > 2:
+            ratios.append(tools / msgs if msgs > 0 else 0)
+    if not ratios:
+        return {"value": 0.0, "avg_tool_msg_ratio": 0.0, "sampled": 0}
+    avg_ratio = sum(ratios) / len(ratios)
+    first_try = sum(1 for r in ratios if r < 0.5)
+    return {"value": round(first_try / len(ratios), 3), "avg_tool_msg_ratio": round(avg_ratio, 3), "sampled": len(ratios), "interpretation": "Higher value = fewer backtracks = better first-try success"}
+
+
+def compute_knowledge_age(facts):
+    if not facts:
+        return {"value": 0.0, "avg_age_days": 0, "stale_facts": 0, "total_facts": 0}
+    now = datetime.now(timezone.utc)
+    ages = []
+    stale_count = 0
+    for f in facts:
+        confirmed = f.get("last_confirmed") or f.get("first_seen")
+        if confirmed:
+            try:
+                dt = datetime.fromisoformat(confirmed.replace("Z", "+00:00"))
+                if dt.tzinfo is None:
+                    dt = dt.replace(tzinfo=timezone.utc)
+                age = (now - dt).days
+                ages.append(age)
+                if age > 30:
+                    stale_count += 1
+            except (ValueError, AttributeError):
+                pass
+    if not ages:
+        return {"value": 0.0, "avg_age_days": 0, "stale_facts": 0, "total_facts": len(facts)}
+    avg_age = sum(ages) / len(ages)
+    freshness = max(0.0, 1.0 - (avg_age / 90))
+    return {"value": round(freshness, 3), "avg_age_days": round(avg_age, 1), "stale_facts": stale_count, "total_facts": len(facts), "interpretation": "1.0 = all facts fresh. 0.0 = all facts 90+ days old"}
+
+
+# --- Trend Computation ---
+
+def compute_trend(current, previous, metric_key="value"):
+    if not previous:
+        return {"delta": "N/A", "direction": "unknown"}
+    curr_val = current.get(metric_key, 0)
+    prev_val = previous.get(metric_key, 0)
+    if prev_val == 0:
+        return {"delta": "N/A (no baseline)", "direction": "unknown"}
+    pct = ((curr_val - prev_val) / abs(prev_val)) * 100
+    direction = "up" if pct > 0 else "down" if pct < 0 else "flat"
+    if metric_key == "error_recurrence" or metric_key == "knowledge_age":
+        direction_label = "good" if pct < 0 else "bad" if pct > 0 else "neutral"
+    else:
+        direction_label = "good" if pct > 0 else "bad" if pct < 0 else "neutral"
+    return {"delta": f"{'+' if pct > 0 else ''}{pct:.1f}%", "direction": direction, "assessment": direction_label}
+
+
+# --- Output Formatters ---
+
+def format_json(metrics):
+    return json.dumps(metrics, indent=2)
+
+
+def format_markdown(metrics):
+    lines = ["# Compounding Intelligence Metrics", f"**Generated:** {metrics.get('generated_at', 'unknown')}", ""]
+    trend = metrics.get("trend_7d", {})
+
+    def metric_block(name, data, desc, good_direction="up"):
+        val = data.get("value", 0)
+        t = trend.get(name, {})
+        delta = t.get("delta", "N/A")
+        assessment = t.get("assessment", "unknown")
+        arrow = "up" if assessment == "good" else "down" if assessment == "bad" else "---"
+        lines.extend([f"## {name}", desc, "", f"**Value:** {val}  |  **7d trend:** {delta} {arrow} ({assessment})", ""])
+        for k, v in data.items():
+            if k != "value" and k != "interpretation":
+                if isinstance(v, (int, float, str)):
+                    lines.append(f"- {k}: {v}")
+        lines.append("")
+
+    metric_block("knowledge_velocity", metrics.get("knowledge_velocity", {}), "New facts extracted per day. Higher = compounding loop working.")
+    metric_block("knowledge_coverage", metrics.get("knowledge_coverage", {}), "Percentage of domains/repos with 10+ facts. Measures breadth.")
+    metric_block("hit_rate", metrics.get("hit_rate", {}), "Percentage of sessions referencing bootstrapped knowledge.")
+    metric_block("error_recurrence", metrics.get("error_recurrence", {}), "Ratio of recurring errors. Lower = fleet learning from mistakes.", good_direction="down")
+    metric_block("task_completion", metrics.get("task_completion", {}), "Percentage of sessions ending with successful completion.")
+    metric_block("first_try_success", metrics.get("first_try_success", {}), "Percentage of sessions completed without backtracking.")
+    metric_block("knowledge_age", metrics.get("knowledge_age", {}), "Freshness of knowledge store. 1.0 = all fresh, 0.0 = all stale.", good_direction="up")
+    return "\n".join(lines)
+
+
+# --- Snapshot Persistence ---
+
+def load_snapshot(metrics_dir):
+    snapshot_path = metrics_dir / "latest_snapshot.json"
+    if snapshot_path.exists():
+        with open(snapshot_path) as f:
+            return json.load(f)
+    return {}
+
+
+def save_snapshot(metrics_dir, metrics):
+    metrics_dir.mkdir(parents=True, exist_ok=True)
+    snapshot_path = metrics_dir / "latest_snapshot.json"
+    with open(snapshot_path, "w") as f:
+        json.dump(metrics, f, indent=2)
+
+
+# --- Main ---
+
+def main():
+    parser = argparse.ArgumentParser(description="Compounding Intelligence Metrics")
+    parser.add_argument("--since", help="Start date (YYYY-MM-DD)")
+    parser.add_argument("--repo", help="Filter by repo/domain")
+    parser.add_argument("--format", choices=["json", "markdown"], default="json")
+    parser.add_argument("--knowledge-dir", type=Path, default=DEFAULT_KNOWLEDGE_DIR)
+    parser.add_argument("--db", type=Path, default=DEFAULT_DB_PATH)
+    parser.add_argument("--save-snapshot", action="store_true", help="Save current metrics as snapshot for trend tracking")
+    parser.add_argument("--metrics-dir", type=Path, default=Path(__file__).parent.parent / "metrics", help="Directory for snapshots and dashboard")
+    args = parser.parse_args()
+
+    facts = load_facts(args.knowledge_dir)
+    yaml_counts = count_yaml_facts(args.knowledge_dir)
+    if args.repo:
+        facts = [f for f in facts if f.get("domain") == args.repo]
+    conn = open_db(args.db)
+    sessions = query_sessions(conn, since=args.since)
+    messages = query_messages(conn) if conn else []
+    if conn:
+        conn.close()
+
+    velocity = compute_knowledge_velocity(facts, since=args.since)
+    coverage = compute_knowledge_coverage(facts, yaml_counts)
+    hit_rate = compute_hit_rate(sessions, messages, facts)
+    error_recurrence = compute_error_recurrence(messages)
+    task_completion = compute_task_completion(sessions)
+    first_try = compute_first_try_success(sessions)
+    age = compute_knowledge_age(facts)
+
+    previous = load_snapshot(args.metrics_dir)
+    trend = {
+        "knowledge_velocity": compute_trend(velocity, previous.get("knowledge_velocity", {})),
+        "knowledge_coverage": compute_trend(coverage, previous.get("knowledge_coverage", {})),
+        "hit_rate": compute_trend(hit_rate, previous.get("hit_rate", {})),
+        "error_recurrence": compute_trend(error_recurrence, previous.get("error_recurrence", {}), "value"),
+        "task_completion": compute_trend(task_completion, previous.get("task_completion", {})),
+        "first_try_success": compute_trend(first_try, previous.get("first_try_success", {})),
+        "knowledge_age": compute_trend(age, previous.get("knowledge_age", {})),
+    }
+
+    now = datetime.now(timezone.utc).isoformat()
+    metrics = {
+        "generated_at": now,
+        "knowledge_velocity": velocity,
+        "knowledge_coverage": coverage,
+        "hit_rate": hit_rate,
+        "error_recurrence": error_recurrence,
+        "task_completion": task_completion,
+        "first_try_success": first_try,
+        "knowledge_age": age,
+        "trend_7d": trend,
+    }
+    if args.since:
+        metrics["since"] = args.since
+
+    if args.save_snapshot:
+        save_snapshot(args.metrics_dir, metrics)
+        dashboard_path = args.metrics_dir / "dashboard.md"
+        with open(dashboard_path, "w") as f:
+            f.write(format_markdown(metrics))
+
+    if args.format == "json":
+        print(format_json(metrics))
+    else:
+        print(format_markdown(metrics))
+
+
+if __name__ == "__main__":
+    main()
--- a/scripts/session_reader.py
+++ b/scripts/session_reader.py
@@ -1,142 +0,0 @@
-#!/usr/bin/env python3
-"""
-session_reader.py — Parse Hermes session JSONL transcripts.
-
-Each line in a session file is a JSON object representing a message.
-Standard fields: role (user|assistant|system), content (str), timestamp (str).
-Tool calls and tool results are also captured.
-"""
-
-import json
-import sys
-from pathlib import Path
-from typing import Iterator, Optional
-
-
-def read_session(path: str) -> list[dict]:
-    """Read a session JSONL file and return all messages as a list."""
-    messages = []
-    with open(path, 'r', encoding='utf-8') as f:
-        for line_num, line in enumerate(f, 1):
-            line = line.strip()
-            if not line:
-                continue
-            try:
-                msg = json.loads(line)
-                messages.append(msg)
-            except json.JSONDecodeError as e:
-                print(f"WARNING: Skipping malformed JSON at line {line_num}: {e}", file=sys.stderr)
-    return messages
-
-
-def read_session_iter(path: str) -> Iterator[dict]:
-    """Iterate over session messages without loading all into memory."""
-    with open(path, 'r', encoding='utf-8') as f:
-        for line_num, line in enumerate(f, 1):
-            line = line.strip()
-            if not line:
-                continue
-            try:
-                yield json.loads(line)
-            except json.JSONDecodeError as e:
-                print(f"WARNING: Skipping malformed JSON at line {line_num}: {e}", file=sys.stderr)
-
-
-def extract_conversation(messages: list[dict]) -> list[dict]:
-    """Extract user/assistant conversation turns, skipping tool-only messages."""
-    conversation = []
-    for msg in messages:
-        role = msg.get('role', '')
-        content = msg.get('content', '')
-        
-        # Skip empty messages and pure tool calls
-        if role in ('user', 'assistant', 'system'):
-            if isinstance(content, str) and content.strip():
-                conversation.append({
-                    'role': role,
-                    'content': content.strip(),
-                    'timestamp': msg.get('timestamp', '')
-                })
-            elif isinstance(content, list):
-                # Multimodal content — extract text parts
-                text_parts = []
-                for part in content:
-                    if isinstance(part, dict) and part.get('type') == 'text':
-                        text_parts.append(part.get('text', ''))
-                if text_parts:
-                    conversation.append({
-                        'role': role,
-                        'content': '\n'.join(text_parts),
-                        'timestamp': msg.get('timestamp', '')
-                    })
-    return conversation
-
-
-def truncate_for_context(messages: list[dict], head: int = 50, tail: int = 50) -> list[dict]:
-    """Truncate long sessions: keep first N + last N messages.
-    
-    This preserves session start (initial context) and end (final results),
-    skipping the messy middle of long debugging sessions.
-    """
-    if len(messages) <= head + tail:
-        return messages
-    
-    truncated = messages[:head]
-    truncated.append({
-        'role': 'system',
-        'content': f'[{len(messages) - head - tail} messages truncated]',
-        'timestamp': ''
-    })
-    truncated.extend(messages[-tail:])
-    return truncated
-
-
-def messages_to_text(messages: list[dict]) -> str:
-    """Convert message list to plain text for LLM consumption."""
-    lines = []
-    for msg in messages:
-        role = msg.get('role', 'unknown').upper()
-        content = msg.get('content', '')
-        if msg.get('role') == 'system' and 'truncated' in content:
-            lines.append(f'--- {content} ---')
-        else:
-            lines.append(f'{role}: {content}')
-    return '\n\n'.join(lines)
-
-
-def get_session_metadata(path: str) -> dict:
-    """Extract metadata from a session file (first message often has config info)."""
-    messages = read_session(path)
-    if not messages:
-        return {'path': path, 'message_count': 0}
-    
-    first = messages[0]
-    last = messages[-1]
-    
-    return {
-        'path': path,
-        'message_count': len(messages),
-        'first_timestamp': first.get('timestamp', ''),
-        'last_timestamp': last.get('timestamp', ''),
-        'first_role': first.get('role', ''),
-        'has_tool_calls': any(m.get('tool_calls') for m in messages),
-    }
-
-
-if __name__ == '__main__':
-    if len(sys.argv) < 2:
-        print(f"Usage: {sys.argv[0]} <session.jsonl>")
-        sys.exit(1)
-    
-    path = sys.argv[1]
-    meta = get_session_metadata(path)
-    print(json.dumps(meta, indent=2))
-    
-    messages = read_session(path)
-    conv = extract_conversation(messages)
-    print(f"\nConversation: {len(conv)} turns")
-    
-    truncated = truncate_for_context(conv)
-    print(f"After truncation: {len(truncated)} turns")
-    print(f"\nPreview (first 500 chars):")
-    print(messages_to_text(truncated[:5])[:500])
--- a/scripts/test_harvester_pipeline.py
+++ b/scripts/test_harvester_pipeline.py
@@ -1,162 +0,0 @@
-#!/usr/bin/env python3
-"""
-Smoke test for harvester pipeline — verifies the full chain:
-session_reader -> prompt -> LLM (mocked) -> validate -> deduplicate -> store
-
-Does NOT call the real LLM. Tests plumbing only.
-"""
-
-import json
-import sys
-import tempfile
-import os
-from pathlib import Path
-
-# Setup path
-SCRIPT_DIR = Path(__file__).parent.absolute()
-sys.path.insert(0, str(SCRIPT_DIR))
-
-from session_reader import read_session, extract_conversation, truncate_for_context, messages_to_text
-from harvester import validate_fact, deduplicate, load_existing_knowledge, fact_fingerprint
-
-
-def test_session_reader():
-    """Test that session_reader parses JSONL correctly."""
-    with tempfile.NamedTemporaryFile(mode='w', suffix='.jsonl', delete=False) as f:
-        f.write('{"role": "user", "content": "Hello", "timestamp": "2026-04-13T10:00:00Z"}\n')
-        f.write('{"role": "assistant", "content": "Hi there", "timestamp": "2026-04-13T10:00:01Z"}\n')
-        f.write('{"role": "user", "content": "Clone the repo", "timestamp": "2026-04-13T10:00:02Z"}\n')
-        f.write('{"role": "assistant", "content": "Cloned successfully", "timestamp": "2026-04-13T10:00:05Z"}\n')
-        path = f.name
-
-    messages = read_session(path)
-    assert len(messages) == 4, f"Expected 4 messages, got {len(messages)}"
-
-    conv = extract_conversation(messages)
-    assert len(conv) == 4, f"Expected 4 conversation turns, got {len(conv)}"
-
-    text = messages_to_text(conv)
-    assert "USER: Hello" in text
-    assert "ASSISTANT: Hi there" in text
-
-    truncated = truncate_for_context(conv, head=2, tail=2)
-    assert len(truncated) == 4  # 4 <= head+tail, so no truncation
-
-    os.unlink(path)
-    print("  [PASS] session_reader pipeline works")
-
-
-def test_validate_fact():
-    """Test fact validation."""
-    good = {"fact": "Gitea token is at ~/.config/gitea/token", "category": "tool-quirk", "repo": "global", "confidence": 0.9}
-    assert validate_fact(good), "Valid fact should pass"
-
-    bad_missing = {"fact": "Something", "category": "fact"}
-    assert not validate_fact(bad_missing), "Missing fields should fail"
-
-    bad_category = {"fact": "Something", "category": "nonsense", "repo": "x", "confidence": 0.5}
-    assert not validate_fact(bad_category), "Bad category should fail"
-
-    bad_conf = {"fact": "Something", "category": "fact", "repo": "x", "confidence": 1.5}
-    assert not validate_fact(bad_conf), "Confidence > 1.0 should fail"
-
-    print("  [PASS] fact validation works")
-
-
-def test_deduplicate():
-    """Test deduplication."""
-    existing = [
-        {"fact": "Token is at ~/.config/gitea/token", "category": "tool-quirk", "repo": "global", "confidence": 0.9}
-    ]
-    new = [
-        {"fact": "Token is at ~/.config/gitea/token", "category": "tool-quirk", "repo": "global", "confidence": 0.9},  # exact dup
-        {"fact": "Deploy uses Ansible on port 22", "category": "pattern", "repo": "fleet", "confidence": 0.8},  # unique
-    ]
-    result = deduplicate(new, existing)
-    assert len(result) == 1, f"Expected 1 unique, got {len(result)}"
-    assert result[0]["fact"] == "Deploy uses Ansible on port 22"
-    print("  [PASS] deduplication works")
-
-
-def test_knowledge_store_roundtrip():
-    """Test loading and writing knowledge index."""
-    with tempfile.TemporaryDirectory() as tmpdir:
-        # Load empty index
-        index = load_existing_knowledge(tmpdir)
-        assert index["total_facts"] == 0
-
-        # Write a fact
-        new_facts = [{"fact": "Test fact", "category": "fact", "repo": "test", "confidence": 0.9}]
-
-        # Use harvester's write function
-        from harvester import write_knowledge
-        write_knowledge(index, new_facts, tmpdir, source_session="test.jsonl")
-
-        # Reload and verify
-        index2 = load_existing_knowledge(tmpdir)
-        assert index2["total_facts"] == 1
-        assert index2["facts"][0]["fact"] == "Test fact"
-        assert index2["facts"][0]["source_session"] == "test.jsonl"
-
-        # Check markdown was written
-        md_path = Path(tmpdir) / "repos" / "test.md"
-        assert md_path.exists(), "Markdown file should be created"
-
-    print("  [PASS] knowledge store roundtrip works")
-
-
-def test_full_chain_no_llm():
-    """Test the full pipeline minus the LLM call."""
-    with tempfile.NamedTemporaryFile(mode='w', suffix='.jsonl', delete=False) as f:
-        f.write('{"role": "user", "content": "Clone compounding-intelligence", "timestamp": "2026-04-13T10:00:00Z"}\n')
-        f.write('{"role": "assistant", "content": "Cloned successfully", "timestamp": "2026-04-13T10:00:05Z"}\n')
-        session_path = f.name
-
-    with tempfile.TemporaryDirectory() as knowledge_dir:
-        # Step 1: Read
-        messages = read_session(session_path)
-        assert len(messages) == 2
-
-        # Step 2: Extract conversation
-        conv = extract_conversation(messages)
-        assert len(conv) == 2
-
-        # Step 3: Truncate
-        truncated = truncate_for_context(conv, head=50, tail=50)
-
-        # Step 4: Convert to text (this goes to the LLM)
-        transcript = messages_to_text(truncated)
-        assert "Clone compounding-intelligence" in transcript
-
-        # Step 5-7: Would be LLM call, validate, deduplicate
-        # We simulate LLM output here
-        mock_facts = [
-            {"fact": "compounding-intelligence repo was cloned", "category": "fact", "repo": "compounding-intelligence", "confidence": 0.9}
-        ]
-        valid = [f for f in mock_facts if validate_fact(f)]
-
-        # Step 6: Deduplicate
-        index = load_existing_knowledge(knowledge_dir)
-        new_facts = deduplicate(valid, index.get("facts", []))
-        assert len(new_facts) == 1
-
-        # Step 7: Store
-        from harvester import write_knowledge
-        write_knowledge(index, new_facts, knowledge_dir, source_session=session_path)
-
-        # Verify
-        index2 = load_existing_knowledge(knowledge_dir)
-        assert index2["total_facts"] == 1
-
-    os.unlink(session_path)
-    print("  [PASS] full chain (reader -> validate -> dedup -> store) works")
-
-
-if __name__ == "__main__":
-    print("Running harvester pipeline smoke tests...")
-    test_session_reader()
-    test_validate_fact()
-    test_deduplicate()
-    test_knowledge_store_roundtrip()
-    test_full_chain_no_llm()
-    print("\nAll tests passed.")
--- a/scripts/validate_knowledge.py
+++ b/scripts/validate_knowledge.py
@@ -0,0 +1,80 @@
+#!/usr/bin/env python3
+"""
+Validate knowledge files and index.json against the schema.
+
+Usage:
+    python scripts/validate_knowledge.py
+"""
+
+import json
+import sys
+from pathlib import Path
+
+VALID_CATEGORIES = {"fact", "pitfall", "pattern", "tool-quirk", "question"}
+REQUIRED_FACT_FIELDS = {"id", "fact", "category", "domain", "confidence"}
+MAX_FACT_LENGTH = 280
+
+
+def validate_fact(fact, source=""):
+    errors = []
+    for field in REQUIRED_FACT_FIELDS:
+        if field not in fact:
+            errors.append(f"{source}: missing required field '{field}'")
+    if "fact" in fact:
+        if not isinstance(fact["fact"], str) or len(fact["fact"].strip()) == 0:
+            errors.append(f"{source}: 'fact' must be non-empty string")
+        elif len(fact["fact"]) > MAX_FACT_LENGTH:
+            errors.append(f"{source}: 'fact' exceeds {MAX_FACT_LENGTH} chars")
+    if "category" in fact and fact["category"] not in VALID_CATEGORIES:
+        errors.append(f"{source}: invalid category '{fact['category']}'")
+    if "confidence" in fact:
+        if not isinstance(fact["confidence"], (int, float)):
+            errors.append(f"{source}: 'confidence' must be a number")
+        elif not (0.0 <= fact["confidence"] <= 1.0):
+            errors.append(f"{source}: 'confidence' must be 0.0-1.0")
+    if "id" in fact:
+        parts = fact["id"].split(":")
+        if len(parts) != 3:
+            errors.append(f"{source}: 'id' must be domain:category:sequence")
+        elif parts[1] not in VALID_CATEGORIES:
+            errors.append(f"{source}: id category '{parts[1]}' invalid")
+    return errors
+
+
+def main():
+    repo_root = Path(__file__).parent.parent
+    index_path = repo_root / "knowledge" / "index.json"
+    all_errors = []
+
+    if not index_path.exists():
+        print(f"VALIDATION FAILED: index.json not found at {index_path}")
+        sys.exit(1)
+
+    with open(index_path) as f:
+        data = json.load(f)
+
+    if "version" not in data:
+        all_errors.append("index.json: missing 'version'")
+    if "facts" not in data or not isinstance(data["facts"], list):
+        all_errors.append("index.json: missing or invalid 'facts'")
+
+    seen_ids = set()
+    for i, fact in enumerate(data.get("facts", [])):
+        all_errors.extend(validate_fact(fact, f"facts[{i}]"))
+        if "id" in fact:
+            if fact["id"] in seen_ids:
+                all_errors.append(f"index.json: duplicate id '{fact['id']}'")
+            seen_ids.add(fact["id"])
+
+    if all_errors:
+        print(f"VALIDATION FAILED - {len(all_errors)} error(s):\n")
+        for e in all_errors:
+            print(f"  x {e}")
+        sys.exit(1)
+    else:
+        print(f"VALIDATION PASSED - {len(data.get('facts', []))} facts, schema v{data.get('version', '?')}")
+        sys.exit(0)
+
+
+if __name__ == "__main__":
+    main()