Repository navigation
math (clojure replay harness): the final full-stream cut consumes moderation after the last vote - #2861
Merged
Conversation
…eration after the last vote This changes Clojure source: math/dev/replay.clj (the offline replay harness that drives the legacy engine for certification) and its tests. Nothing under math/src, the production engine, is touched. The certification battery flagged moderation-heavy entries at the final cut with moderation-list shape and meta-watermark mismatches and zero numeric outliers. The harness bounded moderation events by the cut time of the last vote, so events arriving after a conversation's last vote were never delivered; production Clojure polls moderation on its own modified cursor and does see them, and the Python full-stream replay already consumes them. On the last slot under full-stream coverage the upper bound is lifted; the lower bound, vote batches, prefix cuts and the source-final-state computation are unchanged. Tests: a Clojure tail test fails on the old harness (three failures, six errors) and passes on the new; a Python parity test checks both engines now see the same events at the final cut. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018NVGBuYCk4EZmiz4csUv9s
Contributor
Delphi Coverage Report
|
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
This changes Clojure source
math/dev/replay.clj(the offline replay harness that drives the legacy engine for certification) and its tests. Nothing undermath/src, the production engine, is touched. Merge awaits the owner's explicit ruling.Why
The certification battery's run 18 receipt flagged the moderation-heavy recipe at its only checkpoint with moderation-list shape and meta-watermark mismatches and zero numeric outliers. Reproduced on a public one-cut stream: the harness bounded moderation events by the cut time of the last vote, so events arriving after a conversation's last vote were never delivered to the legacy engine. Production Clojure polls moderation on its own
modifiedcursor (poller.clj:19-27,postgres.clj:158) and does see them; the Python full-stream replay already consumes them (schedule.py:330-334). The harness was the wrong side.What
A fourth arity on the moderation partition takes the schedule coverage; on the last slot under
full-streamcoverage the upper bound is lifted. The lower bound (no duplicate delivery, tied cut times included), vote batches, prefix cuts, cut-slot resolution and the source-final-state computation are unchanged. Unknown coverage strings are rejected upstream by the Python certifier.Checks
math/dev/replay.clj,math/test/replay_tail_test.clj(new),math/test/test_runner.clj(registration),delphi/tests/replay_harness/test_clj_moderation_tail.py(new). Rejected-word scans empty.🤖 Generated with Claude Code
https://claude.ai/code/session_018NVGBuYCk4EZmiz4csUv9s