Decision Health — detection patterns
Proposed signals for human review. Not verdicts.
Module: detection.py → data/detection.json → Intelligence → Know vs assume.
Complement: deep_scan.py → download + parse /proposals/*.md, open issues with bodies, governance workflow YAMLs → stored corpus → Patterns / Predictions (aggregates). Raw grant-execution lists are Admin-reference only.
| Pattern | What we measure | Artifact |
|---|---|---|
| Fast-track past objections | Closed CIP PRs where approval/merge is close to unresolved objection language | CIP PRs |
| Feasibility limbo | CIP Approved/Active >60 days without Final | CIP dirs / status |
| Grant milestone quality | Adoption vs artifact framing | /proposals/*.md (+ linked issues) — not PR checklist |
| SV weight concentration | Weight rows parsed from CIP text (not on-chain votes) | CIP text |
| Featured App tech review | FA-related CIP threads lacking review keywords | CIP / FA threads |
| DA self-funding | Share of funded proposal filenames with DA-* |
/proposals/*.md listing |
| Grant execution activity | Open issues with milestone/delivery language | Dev Fund issues (excl. PRs) |
Patterns (structural signals)
Module: org_intel.py → _patterns() / _predictions() → data/org_snapshot.json → Intelligence → Patterns.
Patterns are grounded structural signals, not Decision-Health card tags. Each one is measured from live GitHub counts or classified-card data and rendered as an evidence-backed card. Contract for every pattern object:
| Field | Rule |
|---|---|
stat |
The headline number/date/ratio — the observation. Tabular, verifiable. |
evidence[] |
1+ rows, each {label, value, url} linking to the source (repo issues/PRs, board, CIP). No evidence → no pattern. |
implication |
The strategic read, impersonal. Why this drives roadmap / review / outreach. |
watch |
The concrete artifact that would confirm or break the signal. |
basis |
One line naming the measured source (counts, timestamps, attribution). |
strength |
1–3, derived from measured magnitude — never a guess. |
kind |
structural | people | surface | governance (drives colour). |
Every pattern must name a measurable artifact (count, date, ratio, or linked PR/issue). Strength and cited evidence carry confidence — emit nothing that cannot point at a source.
Predictions (_predictions) follow the same discipline: from (surface), trigger (measured), watch_for (next artifact), confirm (what closes the loop), url (source).
Posture
Every Decision-Health row is a proposed signal. Soft language (“possible”, “worth a look”). Cards remain the HITL correction path; detection does not auto-inject into grounding.
Governance CI (sig-label / champion / milestone / aipricheck) is executable — cite workflows, don’t paraphrase as optional policy.
Caps
~12 CIP PRs for fast-track · ~35 CIP dirs for limbo · search API capped · call count in detection.json.
Deep scan: proposals directory listing + ≤3 pages of open issues.