docs(verify): keep resting orders across a restart and cover the safe exit refusals

[MarcosACH]

The round trip recipe read FAIL on main. It still expected paper to ghost a resting order after a restart. Since #500, paper rebuilds its book from the store, so the order stays live. The engine is right and the map was stale.

What changed

• paper-round-trip and ghost-reconcile now prove the order stays live after a restart, on replay and on the wall clock. A real ghost is no longer reachable from the helper, and the file says so.
• config-refusals adds the store lock and the reserved __operator__ id.
• safe-exit adds a paper section: cancel-all is refused while the engine runs, and flatten works with no key.
• crash-recovery and postgres-store check seq_records.
• hyperliquid-testnet adds the 10 second wait its gotcha asked for, and notes that live margin reads null until the venue posts it.
• Small fixes: the strategy count and the store table list in SKILL.md.

Proof

Every feature was driven live on main @ 496ca90. Every changed recipe was re-driven as written and reads PASS (rt-m2 8/8, ghost-m2 2/2, ghostlive-m 3/3, refuse-m2 8/8, se-paper-m 6/6, testnet-m 10/10, crash-m 9/9, pg-m 7/7). No product code changed.

Closes #525

🤖 Generated with Claude Code

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[MarcosACH]

Review of docs/verify-map-upkeep @ d5f59fa

Reviewer: code-review skill
Reviewed: 2026-10-17
Optimization axis: correctness (the recipes must match what the engine does)

Mechanical

• ruff format: pass
• ruff check: pass
• mypy: pass
• lint-imports: pass (7 kept, 0 broken)
• pytest: pass (2185 passed, 75 skipped, coverage gate met)

No product code changed, so the mechanical pass only confirms the tree is still green.

I checked the recipe claims against the code. These all hold:

• _resting_orders in app/build.py keeps only LIVE or partly filled GTC limits with a price.
• dump prints five tables (STORE_TABLES in verify.py).
• run_strategy.py registers five strategies. REST_BELOW defaults to 0.8 and WAIT_SECONDS to 0.
• StoreLockHeld refuses an engine with exit 1, and an exit run with exit.refused and reason="lock_held" (engine/runner.py:279).
• The __operator__ message matches domain/position.py:61.
• ADR-0052, ADR-0054, and ADR-0059 say what the recipes cite them for.
• #500, #517, #519, and #524 are all merged on main.

Findings

[WARN] R001 — .claude/skills/verify-tickwright/features/safe-exit.md:32 — Paper section hides behind testnet preconditions

Category: docs
Why: The preconditions need a funded testnet key and list only run id se-paper's sibling safe-exit. An operator with no key stops there and never runs the paper part. That is the one part of this file that runs on the hermetic path. The paper steps also start after the testnet Cleanup, so a top-to-bottom reader thinks the recipe ended.
Fix: Add a precondition line: "The paper section needs no key. Run id se-paper is unused." Also add a short lead-in at line 75 that says the paper section runs on its own, with or without the testnet part.
Status: open

[NIT] R002 — .claude/skills/verify-tickwright/features/ghost-reconcile.md:17 — Reason string is asserted in a different test file

Category: docs
Why: tests/engine/test_reconcile.py proves the window and the reject, but it never asserts the reason. The string reconciliation: ghost: vanished from the venue is asserted in tests/engine/test_execution_manager.py:1460.
Fix: Cite test_execution_manager.py for the reason. Optionally cite tests/engine/test_reconcile_runtime_e2e.py:179 for the window on the composed engine.
Status: open

[NIT] R003 — .claude/skills/verify-tickwright/features/ghost-reconcile.md:4 — One new sentence is over 25 words

Category: docs
Why: "It arms a 90 second grace window, the open-order cadence re-reads ..." is about 27 words. CLAUDE.md caps a sentence at 25.
Fix: Split it into two sentences after "window".
Status: open

Summary

• BLOCKING: 0 open
• WARN: 1 open
• NIT: 2 open
• Merge: BLOCKED

Ralph-State: changes-requested

[MarcosACH]

Correction to R001 Why: the preconditions list only run id safe-exit. They do not list se-paper. The rest of R001 stands.

[MarcosACH]

• R001 Status: resolved — ba263ec
• R002 Status: resolved — ba263ec
• R003 Status: resolved — ba263ec

[MarcosACH]

Review of docs/verify-map-upkeep @ ba263ec

Reviewer: code-review skill
Reviewed: 2026-10-17
Optimization axis: correctness (the recipes must match what the engine does)

Mechanical

• ruff format: pass
• ruff check: pass
• mypy: pass
• lint-imports: pass (7 kept, 0 broken)
• pytest: pass (2185 passed, 75 skipped, coverage gate met)

The fix commit ba263ec touches two Markdown files only.

Findings

[WARN] R001 — .claude/skills/verify-tickwright/features/safe-exit.md:32 — Paper section hides behind testnet preconditions

Status: resolved — ba263ec (verified)
The preconditions now say the paper section needs no key and names run id se-paper. The paper set up step says it runs on its own.

[NIT] R002 — .claude/skills/verify-tickwright/features/ghost-reconcile.md:17 — Reason string is asserted in a different test file

Status: resolved — ba263ec (verified)
The recipe now cites tests/engine/test_execution_manager.py for the reason. That file asserts it at line 1460.

[NIT] R003 — .claude/skills/verify-tickwright/features/ghost-reconcile.md:4 — One new sentence is over 25 words

Status: resolved — ba263ec (verified)
The sentence is split after "window".

No new findings. The recipes cite #484 (the issue) and the PR body cites #500 (its fix). Both are correct.

Summary

• BLOCKING: 0 open
• WARN: 0 open
• NIT: 0 open
• Merge: READY

Ralph-State: ready