v2: evaluate vibe→acoustic translation — disconfirmed, ships flag-off - #24
Merged
Conversation
The vibe→acoustic-terms translation flagship was disconfirmed by live measurement (translation lowers the CLAP text cosine instead of raising it), so the translator ships flag-off (VIBE_TRANSLATION_ENABLED=False) as dormant, fully-degradable infrastructure — pipeline behaviour is byte-identical to pre-v2 when off. Also couples the fusion weights: β is the env knob, α is derived as 1−β, so combined_score stays in [0, 1] (a Codex review caught that independent knobs let β=0.5 + default α=0.7 emit combined_score>1 and break the rendered [0,1] contract).
Label-free A/B + β-sensitivity instrument, built to measure the flagship before shipping it. Findings: the acoustic translation lowers the CLAP text cosine rather than raising it (0/3 vibe seeds pass the gate), and at β=0.3 the vibe leg is near-inert (β≈0.5 needed to move the output). Measurement-first disconfirmed the flagship. Pure metric core (spearman / vibe_ab_metrics / beta_sensitivity) is unit-tested offline; the driver runs against a live warm corpus.
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
The first v2 "deepen the engine" arc — hypothesize, build flag-off, measure, decide. The flagship was an LLM step that rewrites a natural-language vibe into literal CLAP-trained acoustic terms before text-encoding, to clear the text encoder's weak ~0.15–0.37 cultural-descriptor scores. A label-free A/B + β-sensitivity harness (eval/vibe_ab.py) ran it live against the warm corpus and disconfirmed the premise: the acoustic translation lowers the CLAP text cosine instead of raising it (0/3 vibe seeds pass the gate), and at the production β=0.3 the vibe leg barely moves the output anyway. So the ClaudeVibeTranslator ships flag-off (VIBE_TRANSLATION_ENABLED defaults false) as dormant, fully-degradable infrastructure — the pipeline is byte-identical to pre-v2 when off — and the measurement is the deliverable, not the feature. Also couples the fusion weights (β is the knob, α=1−β) so combined_score stays in [0,1], and extracts a shared _open_query_log helper from the eval harness. 291 offline tests green.