[ops-sweep] 2026-10-01

[i-xtsu-sixyou-ken-mei]

Ops snapshot

2026-10-01T08:30Z, degraded only errors (32 stale, 0 active 24h, confirmed via Sentry inspect 08:27Z); jobs all healthy: backtest-refresh 36831209928 success 07:35Z (09-29 hole remains in curve, needs backfill per #671), ops-operator heartbeat fresh 142m (cadence 240m), ops-cost-sync daily success (Oct month-boundary unknowns clear on next run). Main CI green (push #680 success; 4e4113c data-only snapshot: build success, deploy skipped). Open issues 22, open PRs 4 (incl this sweep). PR CLEAN, CI green; origin/main +1 data commit, no merge (not DIRTY).

Done

• a7f0b80 operator runner skips abandoned/non-renderable renders, falls through to priorities/idle. Fixes control-center: ops-operator cycle pinned to abandoned failed render #675
• 1fae973 Today healthy EmptyState requires no unknown domains and fresh data, names unobserved domains, re-evaluates staleness on 60s timer. Fixes control-center: Today shows healthy with unknown domain or aged data #673
• 3fe33c2 Today ready on overview/operations/costs-history/costs-podcast; statements stay pipeline-lazy, queues/journey second-wave. Fixes control-center: Today waits on /api/statements it never renders #674
• ddee671 drop obsolete support.js ignore; file absent, only ignore remained (refs [operator] Move repo-root credential/keystore files out of the repo root; delete stray support.js #477, keys move still needs human).

Decisions to review

• control-center: ops-operator cycle pinned to abandoned failed render #675: reused exported isRetryableRenderFailure from render-signals.ts in runner.ts instead of copying predicate, so signal feed and runner cannot disagree.
• control-center: Today shows healthy with unknown domain or aged data #673: 60s setInterval tick with cleanup re-renders for staleness without refetch; unknown notice joins domain names with 、 and only shows when top is empty to avoid double-alerting.
• control-center: Today waits on /api/statements it never renders #674: kept costHistory in home-ready gate per issue spec though TodayPage does not render it, to avoid blanking cost-adjacent freshness; queues/journey render not-loaded states via existing QueuePanel/JourneyGlance.

Left for the owner

• backtest-refresh 09-29 hole: proximate cause was CoinGecko 429 on the 09-29 ETL run (ALPHA-ETL-2, single event 09-29T16:22Z, daily-refresh green since). Fix is still backfill, not retry: run Alpha ETL token-price backfill for 2026-09-29 (needs WEBHOOK_SECRET) then redispatch backtest-refresh.yml. See [operator] Decide backfill of missing BTC/ETH prices for 2026-09-29 #671. 10-01 run green does not fill the hole; consider CoinGecko backoff policy separately (retrying a 429 can worsen quota).
• Sentry ANALYTICS-ENGINE-1/2 missing ETH price -> [operator] Decide daily-suggestion handling of missing spot prices #643 needs valuation-semantics decision; dust-debt ZodError -> [operator] Decide the borrowing-positions contract for dust debt #644 needs contract decision; Cloudflare 403 token perms -> [operator] Ops Cost Sync red nightly: Cloudflare billable usage returns 403 #646 needs dashboard/token action (code already fails closed with tested 403 message; nothing to fix in repo).
• Product/cost decisions [operator] 18 of 19 priority customer accounts are 30-day inactive but still refresh daily #475, [operator] Choose reliable scheduler for ops heartbeat jobs #546, [operator] Decide Vercel deploy timeout/reconciliation policy #551, [operator] Validate landing-to-CTA measurement and choose experiment #574, [operator] Decide native auth input pre-validation UX #532, [operator] Decide required PR and CI protection for main #515, [operator] Identify test-qa-hourly and repo-truth runtimes #513, [operator] Inspect current iOS and Android EAS failures before redispatch #512, [operator] Decide the CLAUDE.md/GEMINI.md compatibility-entry-point convention #479 (code+CONTRIBUTING already match symlink convention; owner can close), [operator] Move repo-root credential/keystore files out of the repo root; delete stray support.js #477 (keys move still needs human), [operator] Decide OpenRouter key daily limit vs translation token ceiling #670, [operator] Set a latency budget for control-center Supabase reads #676-678 (server rollup/mutation surface need semantics decisions per [operator] Guard control-center mutations against cross-site and local requests #677/[operator] Decide which unobserved states may roll up healthy in control-center #678).
• Sentry stale podcast-pipeline 23 (inspected 08:32Z: image/narration timeouts, fetch failures, R2/playwright transients; 6x SOCIAL_RELEASE_FAILED single/low-count incl two on 09-29; og:image planning x2; LLM_MODEL env x1; title-length x1; ENOENT copy x1; lease-lost x1), web 3, native 1, desktop 2: inspect before classify; fix on main not yet deployed counts as deploy-clears. Podcast 402 credit-exhaustion (PODCAST-PIPELINE-1A, 09-29) belongs with [operator] Decide OpenRouter key daily limit vs translation token ceiling #670 OpenRouter limit decision, not a code fix.
• Cost ledger Oct unknowns (brave/cloudflare/debank/openrouter/supabase): month-boundary, ops-cost-sync runs daily 04:30Z; clears on next success, no action unless it stays unknown past 10-02.

Not verified locally

• e2e / PostgreSQL analytics-engine / Docker / secrets: PR CI runs them. Local knip under .claude/worktrees misreports; trust PR CI code-quality. Repo lint mapping green locally (verify-jobs.sh lint: 20 tasks ok); raw eslint on account-engine without ^build shows phantom type errors, always run via turbo.