Repository navigation
task(atlas): first-pass corpus ingestion — all code repos, all findable docs, dezren39/cairn #6025
Description
Parent: #6018 (Atlas ingestion program). Statement S15 of the 2026-11-04 brief.
Task (Drew, in substance)
Once the ingestion design is somewhat in place, run a first pass adding all these data points into Atlas:
- All the code repos (the full estate: developing-today/code, the forks, darwin-nix, hatch-personal, hatch, zettel, and the rest) — current content, ingested delta-safely so later re-runs are cheap.
- All the docs we can find — including the untracked material enumerated by the docs-migration Atlas sweep (#5981): research notes, goal workspaces, memory topics, scratch findings that survived.
- The cairn repo (dezren39/cairn, 473 files — Drew's dictation says "the can repo"; interpretation recorded in the program inventory).
- Plus the outputs of the sibling pieces as they land: our-repos GitHub comments (sibling extraction task under #6018) and the operational sources (session/chat logs, statistics, reports).
Explicitly OUT of this pass
- External-repo comments — gated on the licensing research (sibling issue under #6018); they join a later pass if the gate clears.
- Anything the classification revalidation (sibling task) flags as mis-tiered gets its tier fixed at ingest, not copied forward.
Execution requirements
- Provenance from the start: every object lands with the program's provenance marks (
repo@hash:pathfor git content; source-system @ machine @ date marks for the rest) and occurrences recorded per #3125 — the first pass must not create a second provenance-audit debt (cf. the existing-embed audit task). - Delta-safe: the pass runs on the delta-only machinery (sibling design) or, if that design is not yet built, on content-hash skipping at minimum, so re-running the pass is idempotent and cheap (the 21 GB rule: 20.5 GB already present must not be re-landed or retagged).
- Tier discipline: secret-class sources route only to the secret tier; the pass reports per-source tier assignments for spot-check against the revalidation findings.
- Cost/disk discipline: the pass reports its storage delta and runtime against the plan ceiling and the disk/IO guardrails from the operations design; a pass that would breach either stops and reports instead.
Done looks like
A completed, reported run: per-source counts (objects in, chunks landed, skipped-as-duplicates, tier split), storage delta vs plan headroom, and a defect list (unassociated/unprovenanced items logged per the provenance standard, not silently dropped). Follow-on passes become the recurring sweeps of the operations design.
Activity
developing-today commentedon Nov 5, 2026
CODE-1691
dezren39 commentedon Nov 5, 2026
🤖 Agent note · picked up
dezren39 commentedon Nov 5, 2026
🤖 Agent note · triage complete
Triage of issue #6025 at current main (verified fresh 2026-11-05):
area/infra,kind/enhancement), an execution task under the Atlas ingestion programme atlas ingestion program: semi-automated updates, versioning, provenance, feedback — long-term design + tasks (2026-11-04 brief) #6018 (OPEN,status/parkedtracker). No hard-stop labels. It is not runnable yet — by its own opening condition ("Once the ingestion design is somewhat in place, run a first pass…") — and triage here records exactly what "in place" means, with fresh states:priority/high. Standard, not gate: provenance marks and occurrence recording follow it from the start, per task(atlas): first-pass corpus ingestion — all code repos, all findable docs, dezren39/cairn #6025's own requirement not to create a second provenance-audit debt (cf. task(atlas): audit existing embeds for provenance/path correctness against the marking standard #6029, the existing-embed audit — related, not blocking).queue/queued— the pass needs its enumeration, so it should land before or alongside the docs portion, but the code-repos portion does not wait on it.Scoped execution order recorded:
Blockers recorded: #6019, #6021, and #6028-or-fallback — all open sibling work under #6018, none escalated, none needing a Drew decision at triage. Issue #6025 stays open as the execution task until its prerequisites land.