Plan: tab F1 per-tier targets and 10-phase work breakdown#10
Merged
Conversation
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
Plan for closing the Tab F1 gap on GuitarSet validation (currently 0.61 aggregate). Replaces the single-aggregate SPEC §1.4 target with per-tier targets (clean acoustic single-line 0.85, strummed 0.90, clean electric 0.87, distorted electric 0.80) and locks in 10 decisions on the eval composite, compute, and scope. 10-phase plan: foundation setup → pitch ceiling lift → SynthTab pretrain + highres fine-tune → style priors → UI-field audit → learned fusion v2 → video qualitative integration → solo-gated melodic prior → tier shortfall recovery → final eval. Plan-only, no code changes. Phase 0 user actions enumerated in §8.
Address PR #10 review (2026-05-13): - Drop SynthTab (CC-BY-NC-4.0, taints derived weights per SPEC §1.5) and GOAT (request-only research-only) from the default-pipeline path. - Frame the design doc as a SPEC §1.4 amendment proposal; commit the SPEC.md update (§1.4.1) in the same change set, keeping the original SPEC numbers as v1.1 / portfolio stretch reference. - Split the doc: strategy / decision-record kept under the original filename; new Phase 0 implementation plan (2026-05-13-tab-f1-phase-0-implementation.md) with exact files, tests, commands, acceptance outputs. - Add explicit License Gate section (§0) verifying every resource before any compute spend. - Define composite eval policy: ≥ 20 clips / 500 notes per tier, player-split, 95% bootstrap CIs with lower-bound acceptance test, parser-per-source, no synthetic-source clips in eval splits. - Update CLAUDE.md 'Active branch' to reflect main (Modal production deploy landed there; refactor/v1 is 23 commits behind). Plan-only commit; no production code changes.
pgil256
force-pushed
the
plan/tab-f1-per-tier-targets
branch
from
67d9017 to
b7f139d
Compare
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.
Summary
Revised 2026-05-13 per review. Strategy + Phase 0 implementation plan to close the Tab F1 gap (currently 0.61 aggregate on GuitarSet) via per-tier acceptance targets, a public-corpus composite eval set, and the existing v1 pipeline. Plan-only — no code changes.
Two-doc split
docs/plans/2026-05-12-tab-f1-to-spec-design.md. License gate, decisions, framing, resource inventory, composite eval policy, high-level phase outline.docs/plans/2026-05-13-tab-f1-phase-0-implementation.md. Exact files, tests, commands, acceptance outputs for the next ~5 days of work. Later phase impl plans land only after Phase 0 evidence justifies them.Companion changes
SPEC.md§1.4.1 — per-tier amendment table committed in the same change set. Original SPEC numbers kept as v1.1 / portfolio stretch reference.CLAUDE.md— active-branch note updated:mainis active;refactor/v1is 23 commits behind and historical.Key decisions (strategy doc §1)
What changed vs v2 (this PR's first push)
maininstead ofrefactor/v1(the latter is 23 commits behind and lacks 5 of the files v2 referenced)Test plan
impl/tab-f1-phase-0from this base; ship Phase 0 in ~5 days; produce baseline + error-decomposition reports