Repository navigation
Release 1.6.4: run worth-it check - #30
Merged
Merged
Conversation
carry-launch writes its events in one append and refuses a missing ledger dir; ledger status overrides names each overridden ruling's role and lane; write-owners prints its path with forward slashes; batch tests for carry-launch and undeclared.
checks.waits gains unwaited and left_running ({agent, file}): a background check start no await-checks wait followed, and one whose last wait said still running. The safeguard for Claude Code 2.1.300, which discourages waiting on background commands.
Timed on the real archive (d1a's plan and layout, 32 runs indexed, project Thistledown): 0.12 s.
On a copy of the archive (31 runs): in-sample median -7% (-27%..+10%), out-of-sample -21% (-74%..+1788%), (F - R) / T quartiles 0.92 1.23 1.67; threshold 1 agrees with each run's own comparison on 28 of 31 runs.
…l pair not comparable
… and compares by case
…t the agent never got its result
…pected at the lane cap on small machines
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Release 1.6.4: the run worth-it check.
A run pays fixed phases whatever the plan's size. The Offer now counts how many task-times the session's lane layout saves by running in parallel (
scripts/worth-it, modelled on the engine's lane queue) and recommends a direct build when that is 1 task-time or less (always for a single lane), saying what a direct build gives up. Parallel lanes stays on offer; any failure of the check leaves the Offer as before. The threshold comes from a back-test of 31 archived runs (tools/worth-it-backtest.py): minute estimates failed out of sample (-74% to +1788%), while the fixed phases cost a median 1.23 task-times; at threshold 1 the rule agreed with each run's own comparison on 28 of 31 runs.Also in this release:
checks.commands, by check id) and a widened-checks table; version 7.history --runs A,B: compare two runs and say whether they are comparable.host_budget(lanes at once above the CPU count).--widendirectory without/, capped task-checks print,run-report --outsummary, reviewers' evidence line, setup refuses SHA-256 repositories, autopsy's unwaited checks, the 1.6.3 deferred minors, a reference.md contents list and a fuller skill description.Testing: built directly, test first; a builder check and an adversarial claim check of the design; a final review by two fresh reviewers (no critical findings; three important and one raised from minor fixed test-first; ten of twelve minors fixed). Full suite on Windows: 1547 tests, 1531 pass, 16 skipped, 0 fail. Python 3.8 is no longer tested (end of life 2024). Not run: a validation run on a real project.