Filling the should/nice gaps — 49 verified cards shipped, 40 held at the spend limit
Generated 89 cards for the remaining should/nice coverage gaps, but the monthly API spend limit was hit partway through the fact-verify pass. Shipped only the 49 that passed verification; the 40 unverified (incl. 8 personal-narrative cards) are held rather than shipped unchecked.
The should/nice gaps from the coverage audit (the lower-priority depth, after the MUST set) came to 89 genuinely-new topics after a dedup pass against the existing deck. Generated all 89 as full cards across four shapes (concepts, /ai, /phrases, dbperf), each followed by a second-agent fact-verify — the same generate-then-verify pipeline as the earlier batches.
The monthly API spend limit (claude.ai) hit partway through, so ~40 of the verify steps failed. The pipeline falls back to the unverified generated card when verify can't run, which is exactly the case where I must not call something "done." So this ships only the 49 cards that actually passed verification (47 /concepts — Java/JVM internals, DSA, design patterns, distributed systems, DB; 2 dbperf). CONCEPTS 190 → 237, DBPERF 7 → 9. Confirmed live in the deployed container build, external HTTPS 200.
The 40 unverified are held, not shipped. A check confirmed none of them leaked into the decks (their ids appear only as text mentions inside other cards, not as card entries). Two reasons they stay held: they haven't been fact-checked, and 8 of them are personal-narrative cards (why-Opendoor, resilience/motivation, honest-disclosure-of-weakness, etc.) that depend on the owner's real story — those shouldn't be auto-generated at all, since inventing a personal narrative is fabrication. They need the owner's actual material, or they get dropped.
Next: when the spend limit is raised/reset, re-verify the 40 (minus the personal ones, which need real input) and ship the ones that pass. Deploys are now quick on the NCP box (rebuild slim image → ssh load → recreate container).