Repository navigation
Metrics and Observability
Stephane Pareilleux edited this page Mar 22, 2026
·
2 revisions
The D_c formula measures real governance value — weighted toward user interactions and executed pipelines, not artifact counts:
D_c = user_interactions × 0.40 + executed_pipelines × 0.25 +
validated_findings × 0.20 + external_citations × 0.15
Why D_c matters: The old formula (citations × 0.35 + PDCA × 0.25 + ...) was gameable by agents citing each other. The new formula requires real-world proof of value.
See docs/articles/ergol-vs-lolli-lessons.md for full analysis.
From Jean-Pierre Petit's Economicon (1984):
| Dimension | ERGOL | LOLLI |
|---|---|---|
| Definition | Productive capacity — real value delivered | Inflated metrics — activity without outcome |
| Example | "Cool!" from a real guitarist using the chatbot | 50 design specs with zero execution |
| Detection | Did a real human find this useful today? | Could this be deleted without anyone noticing? |
| Weight in D_c | user_interactions × 0.40 | artifact_count × 0 |
The Anti-LOLLI Checklist — before creating ANY artifact:
- Who will use this today? (not "eventually")
- Can I validate it right now?
- Does something similar already exist?
- What is the ERGOL metric?
- Would deleting this hurt anyone?
Related policies: Anti-LOLLI inflation, Compounding metrics
| Metric | Formula | Interpretation | Target |
|---|---|---|---|
| Moral Sensitivity | signals_fired / governance_actions | Rising = more attentive; Stable = mature; Falling = possible desensitization | 5-30% |
| False Positive Rate | false_positive_signals / total_signals | Falling = more accurate; Rising = too sensitive | < 20% |
| Regret Resolution Velocity | avg_days_to_resolve (30-day rolling) | Falling = learning faster; Rising = systemic issues | < 7 days |
| Recurrence Rate | regrets_with_recurrence > 1 / total_regrets | Falling = preventing repeats; Rising = not learning | < 10% |
| Anticipation Accuracy | pre_mortems_predicting_problems / total_pre_mortems | Rising = better prediction | 15-30% |
| Blind Spot Discovery Rate | new_silence_signals / governed_domains | Low = good coverage; Zero = complacency risk | Healthy low |
| Moral Depth | avg_discomfort_weight_when_fired | Rising = deeper concerns; Falling = filtering important signals | Stable |
| Metric | Source | Alert Condition |
|---|---|---|
| Belief count | state/beliefs/*.belief.json |
Fewer beliefs than governed domains |
| Staleness |
updated_at field |
> 5 days warning, > 7 days alert |
| Contradiction rate | Count of C values | Any C value needs investigation |
| Unknown rate | Count of U values | High U count = insufficient reconnaissance |
| Confidence calibration | Predicted vs actual outcomes | Systematic over/under-confidence |
| Metric | Source | Meaning |
|---|---|---|
| Citation count | state/evolution/*.evolution.json |
How often an artifact is referenced |
| Violation count | Evolution events of type "violated" | How often an artifact is violated |
| Compliance rate | citations / (citations + violations) | Overall artifact effectiveness |
| Promotion candidates | High compliance + high citation | Artifacts ready for elevation |
| Deprecation candidates | Low citation + no violations | Artifacts that may be unnecessary |
| Repo | Submodule | State Dir | CLAUDE.md | Personas | Tests |
|---|---|---|---|---|---|
| Demerzel | N/A | Full | Full | 14 | 60 suites |
| ix | Current | Created | Current | Via MCP | ix-governance |
| tars | Current | Created | Current | TBD | TBD |
| ga | Current | Created | Updated | 7 agents | TBD |
The ix-dashboard crate provides a TUI view of all governance metrics:
cargo run -p ix-dashboard [state_dir]
Four panels: Beliefs, Evolution, PDCA, Conscience Signals. Color-coded tetravalent values: T=green, F=red, U=yellow, C=magenta.
A Discord-connected governance report is also available via /demerzel-report — posts formatted summaries to #governance channel.
| Report | Frequency | Audience | Channel |
|---|---|---|---|
| Conscience digest | Daily | Demerzel (self-reflection) | state/conscience/digests/ |
| Conscience report | Weekly | Human | GitHub Discussions |
| Governance report | Monthly | Ecosystem | All repo discussions |
| Conscience alert | Real-time | Human | GitHub Issues (weight >= 0.9) |
| Discord report | On demand | Community |
#governance via bot |
- Signals fired: 5 (4 from inception + 1 remediation)
- Active regrets: 1 (stale directive to consumer repos)
- Patterns detected: 2 (governance-drift, blind-spot-consumer)
- Intuitions seeded: 4 (candidate status)
- Moral sensitivity: 1.33 (heightened awareness phase)
- Conscience health: Growing
- Milestone: Nascent
- Policies active: 30
- Behavioral test suites: 60
- Departments running: 21
- IxQL pipelines: 7
- Policies — ERGOL/LOLLI and D_c policies
- Conscience-System — conscience signal architecture
- IxQL — all observability pipelines expressed as IxQL