AINA Data Engine Room - 2026-06-12 - local checkpoint ready

AIN-506 Anti-Loop Source Authority And Embedding Repair Handoff

The data was not missing. The missing link was an executable start-here surface that forces agents to reuse the source maps, ledgers, exports, and vector receipts before spending tokens rediscovering them.

Total Gemini vectors5,534

Serviceable title vectors3,440

Top worked vectors1,000

Top 500 coverage500

This pass added source-authority-start-here, tightened title eligibility and repair, pruned stale vectors, restored top-band completeness, and left the repo with passing P0 and validation gates.

Hidden Gems

Anchor	Why it matters
`artifacts/reports/production_source_authority_registry_v1.md`	Build-time source authority and title-cleaning route map.
`docs/TITLE-LEDGER.md`	Title precedence, count reconciliation, and export contract.
`docs/MAPPING-CHAIN-LEDGER.md`	Title to role to workflow to capability join topology.
`/home/ali/conductor/aina-consolidated/20-references/linear/doc__cross-repo-salvage-map.md`	Cross-repo prior-work map; adapt before rebuilding.
`/home/ali/conductor/repos/aina-jobs-research/project-summary-package/file-index/SOURCE_MAP.md`	Jobs-research lineage and output map.
`/home/ali/conductor/repos/aina-core/evidence/canonical`	Stitched evidence atlas parquets.
`/home/ali/ALIPE/data/jobs/linkedin_indeed_clusters_v1/chunks_250k`	Market substrate, only through clean derived chunks.

What Changed

Added source-authority-start-here plus JSON, markdown, and HTML receipts.
Added stricter repair gates for posting IDs, salaries, schedules, school years, location suffixes, comma context suffixes, store/brand fragments, and generic posting phrases.
Preserved real roles like support associate, seasonal sales associate, and anti-money laundering subject matter expert (sme).
Repaired wrappers like teller part time 30 hours, sales 16 and 17 years old, and merchandising part time days.
Embedded 500 serviceable-title vectors and 29 top-worked refill vectors with zero failed rows.
Pruned stale vectors and restored complete top-1,000 and top-500 coverage.

Proof

uv run pytest tests/test_production_embeddings.py tests/test_source_authority_start_here.py -q
34 passed

uv run aina-data-engine --root /srv/aina/aina-data-engine-room ain-506-p0-gate
status: pass, valid: true, model: gemini-embedding-2, dimensions: 768

uv run aina-data-engine --root /srv/aina/aina-data-engine-room validate
status: pass, occupations: 110184, linkedin_jobs: 129165, HF files verified: 15

Resume

cd /srv/aina/aina-data-engine-room
git status --short --branch
uv run aina-data-engine --root /srv/aina/aina-data-engine-room source-authority-start-here
uv run aina-data-engine --root /srv/aina/aina-data-engine-room gemini-embedding-run --source-family serviceable_title --include-repaired --max-new 500 --dry-run

Before the next live call, sample 50 actual candidate titles and only proceed if they are semantically clean.

Ali Mehdi Mukadam - co-authored with Codex - 2026-06-12