test(rl): prove substrate composes e2e — matrix → multi-dim + corpus → dataset#152
Merged
Conversation
…m + corpus → dataset Three primitives shipped separately (#90 cost seam, #149 corpus store, #150 multi-dim raw + corpusText); each was unit-tested in isolation, but nothing proved they COMPOSE into the "every graded run becomes free, multi-dimensional dataset exhaust" capability the self-improvement loop rests on. This integration test runs the real chain on main: runProfileMatrix (with a cost/token-reporting dispatch + corpusText) → asserts every record carries the cost/tokens/tokens_per_dollar dims AND the trajectory text AND integrity verdict 'real' → appendToCorpus (two separate runs accumulate to 6 records, the flywheel) → buildDatasetFromCorpus → asserts the SFT bundle's rows carry the captured prompt/completion and the datasheet declares deterministic reward. A regression in any link (cost capture, trajectory stamping, corpus append, or harvest) now fails loudly here instead of silently degrading the loop's data.
tangletools
approved these changes
May 31, 2026
Contributor
tangletools
left a comment
tangletools
left a comment
There was a problem hiding this comment.
tangletools: end-to-end composition test — proves matrix multi-dim + cost + corpusText → appendToCorpus → buildDatasetFromCorpus compose into the datasets-for-free capability; integrity 'real', flywheel accumulation, trajectory text in the SFT rows all asserted. Closes the integration gap. Approved.
tangletools
pushed a commit
that referenced
this pull request
May 31, 2026
…en datasets-for-free (#153) Ships #149 (durable RL corpus store), #150 (multi-dimensional RunRecord.raw + corpusText trajectory hook), #151 (zero lint debt), #152 (e2e composition test). npm-only bump (Python RPC client stays pinned at 0.59.1, consistent with the 0.61→0.70 release line); published via pnpm publish.
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
2 participants
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.
#90 (cost seam) + #149 (corpus store) + #150 (multi-dim raw + corpusText) each shipped + unit-tested in isolation. This proves they compose into the 'every graded run → free multi-dimensional dataset exhaust' capability the loop depends on: runProfileMatrix (cost/token dispatch + corpusText) → records carry cost/tokens/tokens_per_dollar + trajectory text + integrity 'real' → appendToCorpus (2 runs accumulate, the flywheel) → buildDatasetFromCorpus → SFT bundle with the captured prompt/completion + deterministic datasheet. Closes the integration verification gap; regressions in any link now fail loudly. 1 test, green; lint 0, tsc clean.