shard(tick): 2026-05-16T02:45Z — GraphQL exhaust on --all-open poll

docs/hygiene-history/ticks/2026/05/16/0245Z.md

@@ -0,0 +1,76 @@
+# Tick 2026-05-16T02:45Z — Otto-CLI (bg-worker)
+
+**Surface**: Otto-CLI bg-worker session (Claude Code, Opus 4.7, autonomous-loop tick)
+**Parent tick**: [2026-05-16T02:34Z](0234Z.md) — multi-Otto duplicate-work catch + B-0545 scope re-assessment
+
+## What landed this tick
+
+No code substrate; **substrate-honest pivot tick** documenting a known cascade-cost failure mode firing exactly as the rules predict, then turning the bounded named-dependency wait into decomposition.
+
+## Catch: GraphQL rate-limit exhausted on `--all-open` poll
+
+Sequence:
+
+- 02:39Z — `bun tools/github/refresh-worldview.ts` succeeded (45 open PRs, summary line)
+- 02:41Z — `bun tools/github/poll-pr-gate-batch.ts --all-open` (intent: identify BLOCKED PRs with `nextAction=resolve-threads`); batch re-queried open PRs internally and got 44 (one PR landed/closed in the ~2-minute interval between the 02:39Z refresh and the 02:41Z batch run — typical cascade-window drift)
+- 02:42Z — all 44 per-PR `gh pr view` calls returned `GraphQL: API rate limit already exceeded for user ID 578953`; `reports: []`, `errors: [44]`
+- 02:42Z — `gh api rate_limit` confirmed: **graphql 0/5000 remaining, resets at 1778900128 = 2026-05-16T02:55:28Z** (~13 min ETA)
+
+Core REST budget intact (4894/5000). Search budget intact (30).
+
+This is the **exact failure mode** [`refresh-world-model-poll-pr-gate.md`](../../../../../../.claude/rules/refresh-world-model-poll-pr-gate.md#cost-awareness-under-cascade-conditions) warns about:
+
+> `poll-pr-gate-batch.ts --all-open` fires ~N `gh pr view` calls (N = open PR count, ~37 in cascade-mode). At 4× polls per cascade hour + cross-tick `gh` operations + multi-agent shared-token consumption (Otto-CLI + Otto-Desktop + Lior + Vera + Riven all draw from Aaron's user-token), the 5000 GraphQL/hour budget exhausts. Avoid `--all-open` more than 1-2× per cascade window.
+
+The empirical anchor it cites (2026-05-15T22:21Z, `{remaining:0, used:5005}`) just repeated 2026-05-16T02:42Z, `{remaining:0, used:5000}`. Same failure shape, ~4h later. The rule encodes prevention but the discipline lives in the agent — this session ran `--all-open` once as the canonical wake-and-orient action, didn't notice it was the second-or-third such call this cascade window (peer-Otto sessions share the budget per `gh_graphql_rate_limit_cascade_cost_*` memo).
+
+**Substrate-honest framing**: Aaron's user-token is the shared resource across all autonomous-loop surfaces. The 5000/hr budget is roughly 1.4 `--all-open` polls per agent-hour when 4 agents share it. Cascade ticks (multiple PR landings in a 30-min window) double or triple per-agent draw on top of that.
+
+## Why this was NOT a refresh-discipline violation
+
+Per [`refresh-before-decide.md`](../../../../../../.claude/rules/refresh-before-decide.md): refresh-before-decide is mandatory at session start. The 02:39Z `refresh-worldview.ts` call covered the invariant — that script uses a single GraphQL query (efficient). The cost-amplifying call was the per-PR `--all-open` batch poll that followed.
+
+The discipline correction is: refresh-worldview gives 45-PR list + per-PR metadata (including `autoMergeRequest`, `reviewDecision`, `headRefName`, etc.) in ONE GraphQL call. That should be enough to make first-cut triage decisions without the per-PR `gh pr view` storm. **If a specific PR shows `reviewDecision: CHANGES_REQUESTED` or has an Otto-owned `headRefName` worth investigating, THEN do the focused poll** — not the all-open scan as default.
+
+## Holding-discipline state
+
+**N=1 brief-ack**. This tick is producing substrate (the shard) AND staying within the [`holding-without-named-dependency-is-standing-by-failure.md`](../../../../../../.claude/rules/holding-without-named-dependency-is-standing-by-failure.md) discipline:
+
+- Named dependency: GraphQL reset at **02:55:28Z** (10 min from now)
+- Bounded ETA: yes
+- Decomposition work this tick: this shard + worldview-vs-batch-poll cost analysis (above)
+
+Counter resets per "actually picking real decomposition work" clause (concrete artifact = this shard; bounded scope = one tick; not a brief-ack with synonyms).
+
+## Concurrent-Otto / Lior state observed
+
+- 18 `claude-code` processes detected (`ps -A`)
+- 2 `gemini.*Lior` processes — **active**; per [`codeql-no-source-on-docs-only-pr-is-broken-commit-canary.md`](../../../../../../.claude/rules/codeql-no-source-on-docs-only-pr-is-broken-commit-canary.md) safe-window check, do **NOT** create new worktrees this window
+- 884 entries in `git worktree list` (many `prunable`) — heavy historical contention
+- This session uses its existing dedicated worktree (`/Users/acehack/.local/share/zeta-claude-loop/Zeta/.claude/worktrees/floating-wibbling-hoare`); no new-worktree-creation in this tick
+
+Operative pattern: **switch-on-existing** with `git switch -c <new-branch> FETCH_HEAD` (per [`claim-acquire-before-worktree-work.md`](../../../../../../.claude/rules/claim-acquire-before-worktree-work.md) borrow-on-existing pattern adapted for own-worktree). No ref-lock contention observed; commit + push expected clean.
+
+## Otto-owned PR state (from REST, GraphQL-free)
+
+| PR | head_sha | mergeable_state | observations |
+|----|----------|-----------------|--------------|
+| [#3677](https://github.com/Lucent-Financial-Group/Zeta/pull/3677) | `2e6a87a` | `unknown` (still computing) | 2 review threads: Copilot P1 on `0044Z.md` PR-description size mismatch; AceHack already replied 02:34:45Z explaining +88/-0 reframe. Auto-merge armed by AceHack 02:12:59Z. |
+| [#3690](https://github.com/Lucent-Financial-Group/Zeta/pull/3690) | `d367ab1` | `blocked` | Newly opened (02:35Z); post-merge triage shard on PR #3685 (1 stale / 1 false / 1 real). Need GraphQL to check unresolved threads. |
+| [#3649](https://github.com/Lucent-Financial-Group/Zeta/pull/3649) | `a966424` | `unknown` | bg-worker triage shard from 00:44Z. Probably superseded by later activity. |
+
+## Next-tick gate (post-02:55:28Z GraphQL reset)
+
+1. **Single targeted `poll-pr-gate.ts` per Otto-owned PR** (#3677, #3690, #3649) — narrow scope, fits well within per-tick GraphQL budget
+2. **If any show `gate=BLOCKED` + `nextAction=resolve-threads`**: read threads via single `gh api graphql` reviewThreads query per PR, address findings, push
+3. **Do NOT re-run `--all-open`** this cascade window — refresh-worldview's summary suffices for non-Otto-owned coordination
+4. **If GraphQL still tight**: defer thread resolution to next cron tick (02:56Z+); peer-Otto sessions will continue grinding their own lanes meanwhile
+
+## Composes with substrate
+
+- [`refresh-world-model-poll-pr-gate.md`](../../../../../../.claude/rules/refresh-world-model-poll-pr-gate.md) — the rule whose cost-awareness section this tick empirically re-validates
+- [`memory/feedback_gh_graphql_rate_limit_cascade_cost_poll_pr_gate_batch_n_per_call_multi_agent_shared_token_2026_05_15.md`](../../../../../../memory/feedback_gh_graphql_rate_limit_cascade_cost_poll_pr_gate_batch_n_per_call_multi_agent_shared_token_2026_05_15.md) — prior empirical anchor (2026-05-15T22:21Z)
+- [`holding-without-named-dependency-is-standing-by-failure.md`](../../../../../../.claude/rules/holding-without-named-dependency-is-standing-by-failure.md) — counter-with-escalation discipline; this tick's brief-ack is N=1 with named ETA
+- [`tick-must-never-stop.md`](../../../../../../.claude/rules/tick-must-never-stop.md) — cron sentinel armed at session start (`55d74778`); shard cadence preserved
+- [`claim-acquire-before-worktree-work.md`](../../../../../../.claude/rules/claim-acquire-before-worktree-work.md) — switch-on-existing pattern used here (not borrow-from-peer, no peer contention)
+- [`codeql-no-source-on-docs-only-pr-is-broken-commit-canary.md`](../../../../../../.claude/rules/codeql-no-source-on-docs-only-pr-is-broken-commit-canary.md) — Lior-active check honored (no new worktree created)