Active plan
26 / 61 US
25 done · 1 under observation
A living dashboard for the MLOps v2 plan. Progress is calculated from user stories; paused work is neither hidden nor used to inflate completion.
Active plan
26 / 61 US
25 done · 1 under observation
We close the platform first, then reduce debt with measured payoff, synchronize all six surfaces, and only then train cohorts.
Now: D7 accumulates two real rebuilds in parallel; the next product change is D9. D5 waits for the causal F2 campaign.
Every bar uses the same weighting: done 100%, observing 75%, active 50%, and all other states 0%.
Isolated baseline, backports, minimum CI and cleanup with proven recovery.
Safe manual ingestion and a cron resilient to the official source being blocked.
Main protection, cron authentication, supply chain and web security.
Release integrity, observability and reproducible operations before training more models.
Reduce duplication and complexity only where value is demonstrated.
One canonical truth across data, paper, web, RAG, model card and production.
Answer the stability hypothesis with preregistered evaluation and positive or negative results.
Turn the platform and its results into a reproducible thesis, paper and defense.
Open an epic to see the expected outcome and available evidence for every story.
Keep main isolated from the preserved R9 work.
Port the diagram and generator with visual equivalence.
Integrate and compile the documents without losing branches.
Keep a strict ci-gate and linear history.
Keep R9 open and in draft pending an independent decision.
Clean only accredited targets and retain recovery receipts.
Record only completed facts and verified SHAs.
Inject the fetcher, apply one retry and type WAF blocking.
Validate, promote atomically and upload to S3 in create-only mode.
Separate source access from gates, CI and publishing.
Version source state and maintain one canonical issue.
Cover fetch, freeze and ingestion without depending on the Internet.
Optional time-boxed spike; the manual ritual remains the reliable path.
Publish 300 snapshots, 298 months and the 2026-09 cut under rule zero.
release/2026-09-158ec972c234Use a least-privilege GitHub App to publish through gates.
Retain locks, contracts, triage and SHA-pinned actions.
Verify toolchain and locks inside live jobs.
Require policy, consistency, lint and modeling success.
Reduce 17 production advisories to zero and document triage.
Distinguish blocked source from broken pipeline without losing alerts.
Measure the real CI critical path on a published SHA.
Block pushes when the lock does not represent the DAG.
f67edf9Reject empty artifacts and turn warnings and advisories into contracts.
0092af0Maintain a closed, atomic and fail-closed state feed.
004e478Break the card–manifest cycle with normalized identity.
f9c22a2Remove it only after the causal F2 campaign; never by hand.
Inventory, preserve and clean with conservative GC and receipts.
Observe two real rebuilds before deciding whether track_run stays.
80b3bfb · 0/2Create ENGINEERING.md with a source matrix and backlinks, without deleting yet.
Publish a one-page system overview and SVG using canonical figures.
Centralize categories and metadata with a golden master.
Remove scraper duplication with pure functions.
Replace mutable globals with a testable AuditReport.
Remove silent failures and log country and month for every skip.
Extract shared theme, language and saving from three generators.
Separate migrations, loading and governance while preserving the fingerprint.
Remove consumerless paths with anti-resurrection guards.
Use one verifiable definition of a dirty tree.
Measure coverage and complexity across the real product.
Prevent tooling from again exceeding half the product.
Remove the web literal and read key_facts fail-closed.
Show source blocking without promising automatic updates.
Guard macros, tables, proposal and documentation.
Answer figures from JSON and require a coherent production pin.
Move literals and caveats into versioned facts.
Add the cohort hook without duplicating data.
Correct stale counts and claims.
Mechanize the data → web → production order.
Centralize the pre-training naïve scale.
Build stability without using hold-out information.
Measure already-scored results by cohort without retraining.
Train a registered ladder and document failure too.
Compete against each cohort's naïve baseline under the canonical gate.
Carry results into thesis, web, RAG and model card under rule zero.
Unify universes and correct stale catalog entries and docstrings.
Update and visually review both documents.
Document Cloudflare and semi-automatic ingestion.
Publish the subsection even if the result is negative.
Propagate the cut and cohorts with derived caveats.
Generate presentation figures from key_facts.
Bring cohorts, multi-horizon results and open decisions to Chente.
This log changes only when verifiable evidence exists: a commit, CI, release or receipt.
Monthly tracking merged with 5/5 CI; observation over two real rebuilds begins (0/2).
The single release_id contract is on main and will activate with the next real bulletin.
Empty artifacts, advisories and warnings are now covered by visible contracts and green CI.
The DVC hook blocks stale locks and the web audit records zero vulnerabilities.
Repositories, platform, evidence and recovery were verified without losing work.
Production became fresh with 300 bulletins and an immutable release.