Repository navigation
Native paper capacity evidence for the order-throughput harness (180 actions/min passed) + --stream-timeout - #184
Merged
Conversation
Add blueprints/us-equities/order-throughput/evidence/ with the four 2026-09-24 native_paper receipts (pre-market quote_stale refusal, the frozen opening-auction run, its recover receipt, and the 13:50Z run that met the frozen 200/min capacity criteria), a sanitization index with the SHA-256 of each private original, and the coordinator's independent alpaca-py SDK listing (496 orders with the prefix, all canceled, 0 filled, 0 positions, 0 open). These are capacity measurements, not strategy trades; 1000/min is not measured (x-ratelimit-limit 200). Finding: paper cancel confirmations took about 16 s on trade_updates in the opening auction, beyond the 10 s stream timeout. Add --stream-timeout (stream_timeout_seconds, 0.5-120, default 10), document avoiding roughly the first 15 minutes after the open unless it is raised, and add offline StreamTimeoutTests that reproduce the 16 s lag. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The coordinator's SDK listing for capacity-200-20260924-rth2 kept no listing output, SDK calls or library version, privately or as a hash. State that in the README evidence-status row, the independent-observation section and the index outcome so the match is not read as a retained artifact. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The first coordinator SDK listing was not retained (review medium). It was re-run at 14:16Z as evidence/observe-capacity-20260924.py (alpaca-py 0.44.0, Python 3.12.3; reads the credential file without printing it) and reproduced the same counts: 496 orders with the run prefix, all canceled, filled 0, 0 positions, 0 open orders. stdout, sha256s, redacted argv and exit code are recorded; README and index no longer call it unretained. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
|
You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard. |
seathatflowsinourveins
deleted the
claude/capacity-evidence-20260924
branch
September 24, 2026 14:37
This was referenced Sep 24, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
First native_paper evidence for the execution-capacity harness (
blueprints/us-equities/order-throughput, #165), recorded on 2026-09-24 on the WSL2 workstation. The account reportsx-ratelimit-limit200.quote_stale. IEX pre-market quotes were older than 60 s. 0 orders.stream_terminal_missingafter 10 orders. Paper cancel confirmations took about 16 s (p50 16.1 s) against the 10 s stream timeout. Cancels returned 11×204 and 4×422, and 1 order was still open at cleanup. About 20 s later,recoverfound all 10 cancelled and verified 0 open.Independent observation. This is a retained script using alpaca-py, separate from the harness code path. It found 496 orders with the run's prefix, all
canceled, filled 0, with 0 positions and 0 open orders. The script, its stdout, its hashes and its redacted argv are inevidence/.Fix: a
--stream-timeoutflag (0.5-120 s, default 10), plus guidance to avoid about the first 15 minutes after the open. Offline tests show the harness froze at the 10 s default when cancels confirmed 16 s late, and completed with the timeout at 30 s.Receipts. The committed receipts are byte-identical to the private originals, which contain no host paths, secrets, account ids or order UUIDs.
index-20260924.jsonrecords every SHA-256.Claim boundary. These are capacity measurements, not strategy trades. 1000/min is not measured, because the account is on the 200 tier.
Verification
tests.test_order_throughput: 73 OK, run on the pinned runtime.validate.py,validate_catalogs.py,evidence_manifest.py --checkandverdict_review_gate("no verdict rows changed") all pass.Review. An independent review approved it with 1 medium finding: the observation was not retained. That is fixed here by the retained re-run.
🤖 Generated with Claude Code