AINA Data Engine Room Production Reconciliation Sweep
Current truth, branch reality, pending decisions, and the next safe execution plan before more embedding work.
The Single Idea
The production truth is the local branch codex/aina-prod-readiness-2026-06-14, not main yet. This reconciliation/board-refresh lane started from preserved execution base 11bf3c0, 26 commits ahead of local main, with no remote configured. Embeddings remain parked until the Fusion/core import ledger and engine_room_export_manifest_v1 define product-consumed source families.
01 · Current repo truth
The repo is locally clean and validated.
| Surface | Current state |
|---|---|
| Branch | codex/aina-prod-readiness-2026-06-14 |
| Execution base head | 11bf3c0 Add data engine academy alignment board before this board-refresh checkpoint commit |
| Remote | No remote configured in this checkout |
| Main status | Current branch is 26 commits ahead of local main; main is not yet final truth. |
| Pre-edit preservation | Archive tag archive/2026-06-15/prod-spine-board-refresh-11bf3c0; verified bundle /srv/aina/checkpoints/aina-data-engine-room/2026-06-15-production-spine-board-refresh/aina-data-engine-room-11bf3c0-prod-spine-board-refresh.bundle; SHA256 df9677cbf79ecb68730168a085c029b4c209c3f578124cea44216d2b0759f35b. |
| Runtime boundary | Public runtime, real-user data, external writes, production telemetry, and runtime embedding authority all remain false |
| Fresh check | Result |
|---|---|
ain-506-p0-gate | pass, Vertex ADC, project aina-495702 |
docs-frontmatter-check | pass, 0 invalid, 0 missing |
artifact-exposure-scan | pass, 0 active findings |
validate | pass |
| AIN-510 | promotion_ready |
02 · What is done
The current branch has the strongest production data-engine state.
Runtime boundaryPlatform-live, donor-retirement, and sensitive bridge receipts exist.
Embeddings151,983 vectors, exact-cosine proof, 0 stale vectors.
RollbackSemantic-review 5k failed quality and was pruned safely.
Clean familiesO*NET task, IWA repaired authority, workflow families, named tools, and top cohorts have vectors.
| Metric | Value |
|---|---|
| Total Gemini vectors | 151,983 |
| Combined chunks | 467,436 |
| Unvectorized chunks | 315,453 |
| Known-pair cosine gap | 0.190303 |
| Stale vectors | 0 |
| Top 500 / Top 1,000 coverage | Complete |
For repaired-overlay families, reconciliation counts base plus repaired overlays. “Complete” means the repaired/current-authority chunks have vectors, not that every historical base row has become embedding authority.
03 · Branch sweep
Main can fast-forward later, but Fusion needs decisions first.
main is an ancestor of the Codex branch. The current branch is 26 commits ahead. Only main, fusion/fn-003, and fusion/fn-015 are merged into the current branch. Twenty-four Fusion branches remain unmerged.
| Branch set | Meaning |
|---|---|
| Merged/ancestor | main, fusion/fn-003, fusion/fn-015 |
| High-value manual-port candidates | fusion/fn-016, fusion/fn-021, fusion/fn-050, fusion/fn-064 |
| Likely docs/advisory import candidates | fusion/fn-005, fusion/fn-039, fusion/fn-068, fusion/fn-073, fusion/fn-075 |
| Likely advisory-only | fusion/fn-008, fusion/fn-020, fusion/fn-022, fusion/fn-024, fusion/fn-025, fusion/fn-029, fusion/fn-037, fusion/fn-066, fusion/fn-081, fusion/fn-094 |
| Do not merge wholesale | fusion/fn-018, because it is a huge artifact-exposure remediation branch with 24 commits and massive generated churn |
All listed worktrees are clean. There is one stash from the pre-Fusion integration root. Nothing should be deleted until archive proof and founder approval exist.
04 · What is pending
The next blocker is not more API spend. It is import truth.
| Pending area | Next decision |
|---|---|
| Git/main consolidation | Port or decline unmerged Fusion branches, then fast-forward local main. |
semantic_review | Do not retry 5k as one mixed family; partition or improve QA suite first. |
jobs_research_role | Do not embed title-only repaired rows; enrich with JD/responsibility/workflow/function/seniority context. |
jobs_research_responsibility | High-value pending family; needs eligibility, repair, and semantic QA before embeddings. |
serviceable_title | Needs cleanup/source-authority pass because prior sampling exposed marketplace artifacts. |
| Public production | Still blocked until release receipts, auth/session/tenant, privacy, telemetry, and founder approval exist. |
05 · Recommended plan
Reconcile branches, define the export, then continue clean embedding.
| Step | Action |
|---|---|
| 1 | Freeze embeddings until branch/import truth is logged. |
| 2 | Create an import decision ledger for every unmerged Fusion branch. |
| 3 | Manual-port high-value Fusion work in small batches with gates after each batch. |
| 4 | Fast-forward local main only after accepted imports are reconciled. |
| 5 | Define engine_room_export_manifest_v1 and prove top-500/top-1,000 Academy-safe export consumption. |
| 6 | Resume embeddings only for product-consumed families with repaired-input QA pass. |
06 · Resume commands
Start from the clean Codex branch.
cd /srv/aina/aina-data-engine-room git status --short --branch git branch --no-merged HEAD --format='%(refname:short)' 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 production-runtime-readiness uv run aina-data-engine --root /srv/aina/aina-data-engine-room platform-live-boundary uv run aina-data-engine --root /srv/aina/aina-data-engine-room validate git bundle verify /srv/aina/checkpoints/aina-data-engine-room/2026-06-15-production-spine-board-refresh/aina-data-engine-room-11bf3c0-prod-spine-board-refresh.bundle sha256sum /srv/aina/checkpoints/aina-data-engine-room/2026-06-15-production-spine-board-refresh/aina-data-engine-room-11bf3c0-prod-spine-board-refresh.bundle
Where to start
Start by classifying the unmerged Fusion branches, then define the export contract. The data engine is healthy enough to pause; repo truth and product-consumable payloads are the things to make crisp before more embedding work.