ponytail-gain

1# Ponytail Gain
2 
3Display this scoreboard when invoked. One-shot: do NOT change mode, write flag
4files, or persist anything.
5 
6The figures are the published agentic benchmark of Ponytail 5: headless Claude
7Code (Opus 5.5, default effort) on 39 tasks (feature tickets in a real FastAPI +
8React repo, bug fixes, security and privacy cases, small apps), 5 runs each,
9against the same agent without the skill. 18 of the tasks have hidden checks
10for correctness and safety. Each figure is the geometric mean of the per-task
11medians. They are measured, not computed from the current repo.
12Source: `benchmarks/results/2026-10-07-agentic.md` and the README.
13 
14## Scoreboard
15 
16Render plain ASCII bars. The bar length shows ponytail as a share of the
17no-skill baseline; the label carries the exact figure:
18 
19```
20  ponytail gain           benchmark · 39 tasks × 5 runs · Opus 5.5
21 
22  no-skill        ████████████████████  100%
23  Lines of code   █████████···········   47%   ▼ 53%
24  Output tokens   ███████████·········   55%   ▼ 45%
25  Cost            ███████████████·····   74%   ▼ 26%
26  Time            ████████████········   59%   ▼ 41%
27  Hidden checks passed   97%  (no-skill 96%)
28  Tests where the logic needs one   98%  (no-skill 68%)
29 
30  This repo:  /ponytail-debt  (shortcuts you deferred)
31              /ponytail-audit (what's still cuttable)
32```
33 
34## Honesty boundary
35 
36These are benchmark averages, not this repo. NEVER print a per-repo savings
37…