Sprint 1, week 1 review (18–24 Oct)¶
Summary¶
The MMDEP platform is ahead of plan; the project stories are behind. Most of the week's output went into MMDEP's own tooling, which now works. The ChatGPT projects (API/MM-Cron, MWMC, MCRM) made little visible progress on their sprint 1 stories: open merge requests keep growing, and their memory files are stale. The sprint 2 design harness (queues per agent, Sonnet building their tools, the resumed Lane S agent) is the fix, starting this week.
MMDEP (on plan)¶
- 56 merges into main since 18 Oct, all CI-green; test suite 833 passing.
- Sonnet finished backlog B1–B36, including:
- harness auto-merge and the GitLab source;
- ZDR on OpenRouter;
- the actions sync and knowledge extraction;
- the docs site, ADRs and the C4 model;
- the design check, and design → issues;
- the LiteLLM backend and escalation.
- Integration sprint-1 was merged on Atul's batch approval (20 Oct).
- Added 22–24 Oct:
- the knowledge extraction now runs by itself (40 chats per pass);
- the inbox relay (the SEO audit PDF reaches MWMC on the next jobs pass);
- the docs site fixes and the Cloudflare publishing path;
- the technical decisions register (TD-001…039) with the five-parameter evaluation (rule §22);
- the panel trail, comments on closed items, notes;
- the sprint 2 design-harness plan and queues.
The local code harness (behind)¶
- Starved: 16 runs this month (11 passed), US$0.19 spent. Every queued issue sits in "review", so each pass skips everything.
- gemini-flash-lite passes 100% of its runs (8 of 8); gemini-flash-lite-b passes 38%.
- Fix: Sonnet T0 (close the finished issues, seed 10 ready ones), then B42 (model roles with per-model limits). OpenRouter's US$10 credit now allows 1,000 free-model requests a day.
Projects (from the collector, 24 Oct)¶
| Project | Open MRs (idle > 14 d) | Signals | Sprint 1 stories |
|---|---|---|---|
| API | 23 (9), last pipeline failed | Production /api/health 404. NAV duplicates unchanged (3 dates for scheme 120245, 10 for NIFTY 50). Staging still almost empty: schemes 1 record, index NAV 404, NFOs 500 |
S1-01, 04, 05, 06, 15: no visible completion |
| MM-Cron | 22 (13), last pipeline failed | — | S1-04, 07: no visible completion |
| MWMC / webapp | 6 (6) | memory files stale | S1-02, 03: route inventory done in chats, not linked to stories |
| Website (WPDev) | 6 (2) | memory files stale | — |
| MCRM | 56 (44) | worst WIP | S1-10 category tree: open |
| PE | 5 | workbook awaiting Atul's visual review (#309/#310) | S1-08 waits on Atul |
The sprint status page shows 0% for every project. No issue or MR names a story ID, and GitLab reads hit rate limits (429) on 24 Oct. From sprint 2 the queues name story IDs, so the page can count them.
Other findings¶
- Scheduled check-in paused since 21 Oct ("device absent": it was tied to the linked computer). Under TD-024 Claude stops relaying, so the recommendation is to retire it once the GitHub inbox (B39/B41) is live, or to re-enable it now on Atul's OK.
- Data-source checks S1–S9 time out from the laptop two days running. AMFI and AMC sites are slow or blocked from the office network, as seen before. These checks move to the server when SRV-1 starts, or to a GitHub Actions job (cloud network) sooner.
- Agents aren't writing memory (rule §21): "sessions stale" and "CURRENT.md stale" flags on every project. The queues now require
runs/<step>.mdafter each step.
Atul's side this week¶
- Done: ZDR on OpenRouter, jobs task on, KNOW-1 run, OpenRouter US$10, SEO audit dropped in the inbox, answers on ARCH-1, AGT-1, DOCS-1 (choice).
- Open, in order of effect:
- API-2: resume the Lane S agent.
- S2-GO: one PROCEED each to API Lane P and MWMC.
- SON-GO: Sonnet's PROCEED.
- ASSIST-1: the Assistant as a ChatGPT project.
- DOCS-1: Cloudflare steps.
- SPECS-1, POOL-1, SRV-1, PRG-1.
- The design questions due 1 Nov (DQ-WEB1/2, DQ-ULIP1).
Outlook for week 2 (25–31 Oct)¶
Run the design harness once on API/MM-Cron and MWMC (plan: docs/plans/sprint-2-design-harness-api-mwmc.md).
Exit criteria by 31 Oct: - every sprint 2 story for these projects has an LLD or an issue list; - at least 10 ready issues sit in the harness queue; - Sonnet's four tools (NAV checker, Lighthouse, contract tests, crawler checks) are merged and running daily; - API and MM-Cron are down to at most 5 open MRs each.