AI Fluency Public O*NET Proof-Tail Repair
A conservative source-authority repair slice that lifts clean proof-tail rows while leaving risky title guesses blocked.
This checkpoint repairs the AI Fluency proof-tail only where public O*NET evidence and existing AINA/JD evidence are strong enough. It raises top-1,000 local judgment/proof coverage from 984 to 991 while keeping 9 rows blocked instead of laundering broad title guesses into runtime authority.
O*NET Is Authority Only When The Match Is Clean
The repair lane now has a curated public O*NET authority path. Exact public anchors can repair; special-education family repairs need JD context; electrical technician needs related AINA title-alias evidence; subject/domain guesses stay blocked as context-only.
| File | Purpose |
|---|---|
src/aina_data_engine/ai_fluency_proof_tail_authority_repair.py | Loads O*NET, indexes AINA title aliases, separates authority from context-only evidence, and keeps embeddings/runtime disabled. |
tests/test_ai_fluency_proof_tail_authority_repair.py | Adds exact O*NET success, context-only block, and electrical alias-gated repair coverage. |
docs/planning/aina-production-readiness-board-2026-06-12.md | Updates the production board to the 11-ready/9-blocked proof-tail state. |
991 Of The Top 1,000 Now Have Local Proof Proxies
| Metric | Value |
|---|---|
| Cumulative ready repairs | 11 |
| Public O*NET ready repairs | 7 |
| Still blocked rows | 9 |
| Top-1,000 local judgment/proof proxies | 991 |
| Top-500 local judgment/proof proxies | 497 |
| Embedding, batch, production unlocks | 0 / 0 / 0 |
The repaired rows are still local evaluator/proof fixtures, not learner-observed proof and not embedding authority. Public runtime, real-user data, external writes, production telemetry, and runtime embedding authority remain disabled.
The Remaining 9 Are Blocked On Purpose
| Ready examples | Function |
|---|---|
| special education teacher, elementary school teacher | education |
| cassandra dba, geographic information system specialist | data_analytics |
| dog trainer | physical_frontline |
| electrical technician | engineering_hardware |
| Blocked row | Reason |
|---|---|
| teacher-science / teacher-social studies / teacher-computer | Subject and grade authority are underdetermined. |
| lab assistant | Domain could be medical, research, veterinary, dental, chemical, or academic. |
| landscape designer | Landscape Architect context exists, but designer vs licensed architect is too risky. |
| AML SME, ecommerce manager, food and beverage lead auditor | Exact audit/source authority is still unmapped or too broad. |
Multi-Agent Review Kept The Scope Tight
Three read-only Codex subagents recommended the 7-current-repair/9-block split. Claude CLI passed the diff and flagged two refinements: dog trainer should not be human education, and electrical should use title-alias support only. Both were implemented before validation.
Claude dynamic workflows were not used here because shared schema/artifact writes were safer through one orchestrator. They remain useful later for large read-only or disjoint source-mining sweeps.
Linear Auth Was Revoked During Closeout
The MCP returned auth_revoked, so the issue comment was not posted. Use this exact payload for AIN-510, with AIN-520 and AIN-527 as related proof references.
AI Fluency proof-tail public O*NET repair checkpoint is locally green on `/srv/aina/aina-data-engine-room`, branch `ali/ain-506-p0-gate-2026-06-12`. Proof: - Proof-tail authority repair: 11 cumulative ready repairs, 9 still blocked, 7 public O*NET-backed repairs, 3 top-500 ready repairs. - Top-band coverage: top 1,000 local judgment/proof proxies increased to 991/1000; top 500 increased to 497/500; vector coverage remains 1000/1000 and 500/500. - Runtime boundary: no live Gemini calls, no embedding manifests, no batch manifests, no public runtime, no production telemetry, no external writes, no real-user data, runtime embedding authority remains false. - Final validators passed: focused pytest, ruff, proof-tail repair, proof-tail fixtures, capability coverage, AIN-506, AIN-510, production-runtime-readiness, and full `validate`. Handoff: `/srv/aina/aina-data-engine-room/docs/handoff/2026-06-13-ai-fluency-public-onet-proof-tail-repair-handoff.md`
Cold-Start From Here
cd /srv/aina/aina-data-engine-room git status --short --branch git log -5 --oneline uv run aina-data-engine --root /srv/aina/aina-data-engine-room ain-506-p0-gate uv run aina-data-engine --root /srv/aina/aina-data-engine-room ain-510-retrieval-promotion-gate uv run aina-data-engine --root /srv/aina/aina-data-engine-room ai-fluency-proof-tail-authority-repair uv run aina-data-engine --root /srv/aina/aina-data-engine-room ai-fluency-proof-tail-fixtures uv run aina-data-engine --root /srv/aina/aina-data-engine-room ai-fluency-capability-coverage uv run aina-data-engine --root /srv/aina/aina-data-engine-room production-runtime-readiness uv run aina-data-engine --root /srv/aina/aina-data-engine-room validate
Harvest stronger source authority for the remaining 9 blocked rows before embedding or runtime reliance.