Repository navigation
Make Dream sampling, evidence and candidate exploration honestly measurable - #13
Merged
Merged
Conversation
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Dream currently records requested command sampling as if applied, counts supplied-ledger restatements as SUPPORTED, and embeds full source candidates in every exploration candidate. This change records requested/supported/applied controls with measurement caveats, introduces WAKE ECHO and versioned separate counters, and retains one canonical source with ID/hash references.
It also adds
explore --from-candidate, schema-awarevalidate, andaudit-attributionwith chronology/source evidence and nonzero conflict/unresolved status. Failed command/HTTP telemetry is retained with sanitized categories. Default exploration allowlist is mock only; remote authorization and advisory authority remain explicit.Validation: 204 tests pass (baseline 191); Ruff/format/Flake8, mypy, canonical schema drift, wheel/sdist, Bandit, dependency audit, authored DreamBench/DreamValue, example validate/run and three browser viewport checks pass. The six-candidate fixture shrank from 96,888 to 59,556 bytes. A four-call remote Create→Dream smoke reported $0.1587248 total, 43.14 seconds, and separate ECHO counts; command temperature/seed remained unapplied. No local inference or product-building dogfood was run.
Dependency order: provider-core howlcipher/howl-provider-core#3 → Create howlcipher/howlcreate#7 → this PR. Core and ecosystem Create pins reference exact repair commits, including failed-resume finalist retention. No automatic merge requested.
Limits: echo/attribution text matching is conservative lexical analysis; supplied audit timestamps are not authenticated; HTTP sampling application means transmitted; duration remains per-call rather than a total deadline; historical measurement v1 is not directly comparable. Source snapshot/prompt retention needed for replay is still present.