honey-gain

Honey Gain

Report the committed benchmark results — never a guessed or per-session number, and never an embedded copy that can drift from the bench.

Do

1. Read the committed scoreboard at use time — don't recite from memory:
   • Per-tier table → bench/results/combined.md (code / user-facing / agent-to-agent split — the split is the finding).
   • The hive's own handoff numbers → bench/hive/RESULTS.md.
2. Report the tier table terse: quality (judge vs baseline %, or lossless recovery for handoffs) and output tokens vs baseline, per variant. Honey leads quality in every tier while cutting tokens where it's safe — deepest on code and handoffs, spending more on user-facing polish (the carve-out).

Rules

• Quote only what bench/results/ currently holds. If combined.md and the README disagree, trust bench/results/ (the harness output) and say they're out of sync.
• Asked for numbers on this repo? The bench measures the skill on a fixed task suite, not the user's codebase — offer to run cd bench && npm run bench, don't extrapolate.
• One honest caveat, once: small suite, judge noise — the objective test-pass column is the trustworthy correctness signal.
• Never resurrect the old unreproducible 92%/78%/73% / −57%/−65%/−70% numbers (see the README honesty note).