feat: compaction quality benchmark suite - #4
Draft
TheArchitectit wants to merge 1 commit into
Draft
Conversation
- scripts/benchmark/{facts,extract,compactors,corpus,score,run}.ts
- docs/BENCHMARKS.md with methodology, reproduce steps, and synthetic results
- Pluggable Compactor interface: raw-truncate, mega-compact, pi-vcc-baseline, pi-vcc-ranked
- Symmetric scoring via shared format-agnostic fact extractor (pi-vcc §3.1)
- Paired deltas vs baseline, not marginal medians (pi-vcc §3.4)
- Synthetic corpus: deterministic mulberry32 PRNG, configurable dup fraction
- Real corpus: reads ~/.pi/agent/sessions/*.jsonl (privacy: no transcript text in out/)
- All fully local, zero network (PREVENT-PI-004), deterministic
200-session synthetic results (seed=42):
mega-compact: 56.4% recall, 4.11 density, 6.1k size
pi-vcc-baseline: 30.8% recall, 4.69 density, 2.9k size
pi-vcc-ranked: 37.9% recall, 4.08 density, 4.1k size
raw-truncate: 41.8% recall, 3.80 density, 4.6k size
mega-compact leads on recall (+17.7% over baseline).
pi-vcc-baseline leads on density (value per char).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
Caution The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased. |
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
What
Adds a deterministic benchmark suite that measures compaction quality across multiple compactors using a shared, format-agnostic fact extractor — no compactor gets a parsing advantage.
Structure
scripts/benchmark/facts.tsscripts/benchmark/extract.tsscripts/benchmark/compactors.tsscripts/benchmark/corpus.tsscripts/benchmark/score.tsscripts/benchmark/run.tsdocs/BENCHMARKS.md200-session synthetic results (seed=42)
mega-compact leads on recall (+17.7% over raw-truncate baseline). pi-vcc-baseline leads on density (value per char).
Honesty notes
Reproduce
🤖 Generated with Claude Code