audit-2026-09: realistic cross-framework comparison - #83
Merged
Conversation
Centralize immutable framework versions, artifact identities, environment mappings, and scenario membership. Add a validated cross-repository claim inventory for the library, current chapter notebooks, and publication drafts. Closes ml4t/backtest-dev#2
Serialize both engine outputs, structured exact checks, runtimes, record counts, framework and Python identity, and content-derived provenance. Validate full candidates, reject stale or dirty evidence, and promote accepted results atomically only after the complete matrix passes. Closes ml4t/backtest-dev#3
…al-strategy-expansion
Align every comparison environment with the v0.1.4 ml4t-specs minimum so locked environment builds import the current FeedSpec contract.
Rebuild the complete isolated framework matrix after updating the comparison environments. All 66 required pairs pass and both declared unsupported pairs remain explicit.
Re-execute all 12 required case-study and framework pairs from the frozen bundles against v0.1.4. Every fill, equity, and terminal surface passes the retained comparison policy.
Measure all 12 correctness-passing pairs with one warmup and ten isolated samples per side, retaining engine-only times and deterministic output identities.
stefan-jansen
added
compatibility: affected
Re-run the 250-asset, 5,040-session comparison in all five engines against the current source tree. Every retained order, fill, trade, and terminal-state comparison passes.
…al-strategy-expansion
stefan-jansen
marked this pull request as ready for review
September 3, 2026 12:10
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.
Result\n\n- 17/17 required real-strategy pairs pass across five book-derived workloads\n- eight native-capability exclusions receive no parity credit\n- all 17 passing pairs have one warm-up plus ten isolated engine-call measurements per side\n- synthetic conformance remains separate at 66 required passes and two exclusions\n- the 250-asset synthetic scale workload passes all five engines\n\n## Corrections\n\n- retain complete values, raw differences, digests, versions, inputs, and negative controls\n- compare money after round-half-even to cents, quantity at 1e-5, and prices at 1e-8\n- correct VectorBT automatic call sequencing and LEAN queued-order validation semantics\n- generate README, methodology, profile, guide, and landing-page claims from accepted evidence\n- refresh compatibility and native Zipline evidence\n\n## Verification\n\n- 2,146 tests passed, 14 declared skips, 87.17% coverage\n- Ruff, format, ty, pre-commit, strict MkDocs, and package build passed locally\n- all CI, public parity, compatibility, security, package, and Python 3.12-3.15 qualification checks passed\n- current Chapter 16 notebooks are in stefan-jansen/machine-learning-for-trading#739\n- downstream publication audit is in ml4t/content-marketing#15\n\nThe separate idiomatic performance track was closed as superseded after the user limited performance scope to equivalent engine execution. Public merge and performance-publication treatment remain user decisions.