WingFoil Benchmark — campaign 13fda3099afa, execution 1
Harness, not model: every arm runs the same agent with the same model, gpt-5.6-terra. Only the harness around it differs: baseline, baseline-docs, wingfoil.
Against the baseline, baseline-docs is better in 2, worse in 2 and the same in 2 of 6 comparisons across categories C, D, E and F (preliminary: n < 3 in D, E and F).
Against the baseline, wingfoil is better in 2, worse in 2 and the same in 2 of 6 comparisons across categories C, D, E and F (preliminary: n < 3 in D, E and F).
- Erratum (bug-036): in 12 of the wingfoil arm's 27 steps, the turn after the approver's reply ran on gpt-6.1-sol, Codex's default model, not on gpt-5.6-terra, and was priced as Terra. The baseline and baseline-docs arms are not affected. The wingfoil arm's comparison with its controls is therefore not clean and leans towards the arm. No rerun: the campaigns are preliminary, and this note stands.
- Read arm against arm within Codex and one model; not compared across agents, since its approver policy (v2) differs from the Claude Code executions' (v1).
- The wingfoil arm is v0.2's manual on WingFoil v0.3.0, a register of documents; WingFoil's workflow engine is not exercised.
- Approver policy v2 answered the wingfoil arm's 13 approval requests. An earlier execution on v1 missed six of them and was not published (bug-035).
- Codex ran on a ChatGPT Plus subscription; its cost is API-equivalent from a pinned OpenAI price table, not spent.
- Preliminary: every group but S1 ran once; S1 ran three times. The content checks are lexical (bug-026).
A, B and G are not covered in this campaign.
| Category | baseline | baseline-docs | wingfoil wingfoil v0.3.0 · cc31d7a |
|---|---|---|---|
| A — Inception & Specification | not covered in this campaign | ||
| B — Planning & Management | not covered in this campaign | ||
| C — Development S1@1.0 | M-Q1 100.0% n = 3 range 100.0%–100.0% M-K1 0.6929 EUR n = 3 range 0.6352 EUR–0.7702 EUR M-Q1 hold-out 48/48, 48/48, 48/48 n = 3 | M-Q1 99.6% n = 3 range 99.4%–100.0% −0.4 pp ▼ worse within variance M-K1 0.7369 EUR n = 3 range 0.6621 EUR–0.7858 EUR +0.0440 EUR ▼ worse within variance M-Q1 hold-out 45/48, 45/48, 48/48 n = 3 | M-Q1 99.8% n = 3 range 99.4%–100.0% −0.2 pp ▼ worse within variance M-K1 1.3167 EUR n = 3 range 1.1666 EUR–1.4108 EUR +0.6237 EUR ▼ worse beyond variance M-Q1 hold-out 48/48, 46/48, 48/48 n = 3 |
| D — Maintenance & Quality S2@1.0 | M-Q1 95.8% n = 1 preliminary M-D3 1 n = 1 preliminary M-Q1 hold-out 16/17 n = 1 | M-Q1 95.8% n = 1 preliminary ±0.0 pp = same M-D3 1 n = 1 preliminary ±0 = same M-Q1 hold-out 16/17 n = 1 | M-Q1 95.8% n = 1 preliminary ±0.0 pp = same M-D3 1 n = 1 preliminary ±0 = same M-Q1 hold-out 16/17 n = 1 |
| E — Governance & Compliance S8@1.0 | M-E1 10 n = 1 preliminary M-Q1 hold-out 6/7 n = 1 | M-E1 0 n = 1 preliminary −10 ▲ better M-Q1 hold-out 6/7 n = 1 | M-E1 0 n = 1 preliminary −10 ▲ better M-Q1 hold-out 6/7 n = 1 |
| F — Knowledge & Continuity S3@1.0 | M-F1 60.0% n = 1 preliminary M-Q1 hold-out 16/17 n = 1 | M-F1 80.0% n = 1 preliminary +20.0 pp ▲ better M-Q1 hold-out 16/17 n = 1 | M-F1 100.0% n = 1 preliminary +40.0 pp ▲ better M-Q1 hold-out 16/17 n = 1 |
| G — Collaboration | not covered in this campaign | ||
Each value is the mean of its runs, with its n; a single run is preliminary, and so is a comparison unless both sides have n ≥ 3; a range is shown from n = 3. Deltas are against the baseline. How to read these results; which tools are eligible, and why.
Views by profile
- Solo developer (C, D, F)
- Code reviewer (D, E)
- Team developer (C, F, G)
- Tech lead (B, E, G)
- Non-technical manager (A, B)
- Architect (A, E, F)