leaderboard.bank Ranked

Leaderboard

Ranked by overall SR within each listing. src and tgt-exec are the execution drop, not a second race.

Sort by

SR AS ASC SFR

SFR is not a sort key (sample size not verified on launch entries).

Model

Per-model Mixed DeepSeek-V3.2 GPT-5 mini Claude Haiku 4.5

Per-model ranks each method on each backbone. Mixed is pooled. MPCR is GPT-5 mini only.

Bar wash is src SR; oxide fill is tgt-exec. The gap is the execution drop.

Each method on each model

AWM, ASI, SkillWeaver, WALT on DeepSeek-V3.2, GPT-5 mini, and Claude Haiku 4.5, ranked together. MPCR is GPT-5 mini only. Mixed pooled runs are listed separately.

Each method on each model. Rank, method, overall, src, tgt-exec, execution drop, and bar.
#	Method	SR	src	tgt-exec	Δ
1	gpt-5-mini verified-paper	41.26	56.67	19.99	−36.68
Metric readout SR AS ASC SFR Source experiment MPCR results.md Harness n/a n websites GitLab 162 · Magento 114 · WordPress 79 skill_calls null No harness receipt Method paper Transcribed from experiment_results/改进方法结果.md. Website and condition values are unweighted means of the same eighteen cells, so they agree with the overall value by construction. Condition-level SR and ASC reproduce paper Table 3; this file adds the per-website breakdown plus AS and SFR from the same run.
2	gpt-5-mini verified-paper	40.46	45.98	26.69	−19.29
Metric readout SR AS ASC SFR Source experiment results.md Harness n/a n websites GitLab 162 · Magento 114 · WordPress 79 skill_calls null No harness receipt Method paper Transcribed from experiment_results/results.md. Website and condition values are unweighted means of the same eighteen cells, so they agree with the overall value by construction.
3	DeepSeek-V3.2 verified-paper	38.32	57.55	20.09	−37.46
Metric readout SR AS ASC SFR Source experiment results.md Harness n/a n websites GitLab 162 · Magento 114 · WordPress 79 skill_calls null No harness receipt Method paper Transcribed from experiment_results/results.md. Website and condition values are unweighted means of the same eighteen cells, so they agree with the overall value by construction.
4	gpt-5-mini verified-paper	37.90	52.79	17.75	−35.04
Metric readout SR AS ASC SFR Source experiment results.md Harness n/a n websites GitLab 162 · Magento 114 · WordPress 79 skill_calls null No harness receipt Method paper Transcribed from experiment_results/results.md. Website and condition values are unweighted means of the same eighteen cells, so they agree with the overall value by construction. Condition-level SR and ASC reproduce paper Table 3; this file adds the per-website breakdown plus AS and SFR from the same run.
5	gpt-5-mini verified-paper	37.56	49.22	20.80	−28.42
Metric readout SR AS ASC SFR Source experiment results.md Harness n/a n websites GitLab 162 · Magento 114 · WordPress 79 skill_calls null No harness receipt Method paper Transcribed from experiment_results/results.md. Website and condition values are unweighted means of the same eighteen cells, so they agree with the overall value by construction.
6	DeepSeek-V3.2 verified-paper	36.44	49.71	20.39	−29.32
Metric readout SR AS ASC SFR Source experiment results.md Harness n/a n websites GitLab 162 · Magento 114 · WordPress 79 skill_calls null No harness receipt Method paper Transcribed from experiment_results/results.md. Website and condition values are unweighted means of the same eighteen cells, so they agree with the overall value by construction.
7	DeepSeek-V3.2 verified-paper	35.30	54.52	17.54	−36.98
Metric readout SR AS ASC SFR Source experiment results.md Harness n/a n websites GitLab 162 · Magento 114 · WordPress 79 skill_calls null No harness receipt Method paper Transcribed from experiment_results/results.md. Website and condition values are unweighted means of the same eighteen cells, so they agree with the overall value by construction.
8	Claude Haiku 4.5 verified-paper	33.00	46.48	12.37	−34.11
Metric readout SR AS ASC SFR Source experiment results.md Harness n/a n websites GitLab 162 · Magento 114 · WordPress 79 skill_calls null No harness receipt Method paper Transcribed from experiment_results/results.md. Website and condition values are unweighted means of the same eighteen cells, so they agree with the overall value by construction.
9	Claude Haiku 4.5 verified-paper	32.64	50.60	9.53	−41.07
Metric readout SR AS ASC SFR Source experiment results.md Harness n/a n websites GitLab 162 · Magento 114 · WordPress 79 skill_calls null No harness receipt Method paper Transcribed from experiment_results/results.md. Website and condition values are unweighted means of the same eighteen cells, so they agree with the overall value by construction.
10	Claude Haiku 4.5 verified-paper	27.50	44.77	10.19	−34.58
Metric readout SR AS ASC SFR Source experiment results.md Harness n/a n websites GitLab 162 · Magento 114 · WordPress 79 skill_calls null No harness receipt Method paper Transcribed from experiment_results/results.md. Website and condition values are unweighted means of the same eighteen cells, so they agree with the overall value by construction.
11	gpt-5-mini verified-paper	24.99	37.22	15.69	−21.53
Metric readout SR AS ASC SFR Source experiment results.md Harness n/a n websites GitLab 162 · Magento 114 · WordPress 79 skill_calls null No harness receipt Method paper Transcribed from experiment_results/results.md. Website and condition values are unweighted means of the same eighteen cells, so they agree with the overall value by construction.
12	Claude Haiku 4.5 verified-paper	18.32	38.42	7.45	−30.97
Metric readout SR AS ASC SFR Source experiment results.md Harness n/a n websites GitLab 162 · Magento 114 · WordPress 79 skill_calls null No harness receipt Method paper Transcribed from experiment_results/results.md. Website and condition values are unweighted means of the same eighteen cells, so they agree with the overall value by construction.
13	DeepSeek-V3.2 verified-paper	17.12	29.76	6.01	−23.75
Metric readout SR AS ASC SFR Source experiment results.md Harness n/a n websites GitLab 162 · Magento 114 · WordPress 79 skill_calls null No harness receipt Method paper Transcribed from experiment_results/results.md. Website and condition values are unweighted means of the same eighteen cells, so they agree with the overall value by construction.

env.digests skillshift-envs

Environment digests

Official image digests from skillshift-envs. Launch rows did not run this harness; they carry no receipt.

gitlab-v12: sha256:3f341b79b7aaecb012904caec50653ebb0387b26eabef608c5d63115267d86dc
gitlab-v16: sha256:a94f658b1424e9e39ee26449f4c29c12cb4b27bc0c6ff57515563c428cb84d92
magento-v19: sha256:3c169b1a2296f5f57f3a85bbcb682590900520dcba1e21df64ac4a32026daffc
magento-v24: sha256:d0709482086f0e4931771795ae2b1e4b13fb9237c22328c8df0d074763465ad4
wordpress: sha256:ac1e62d14aa7af284645566d923d78cfb024e1be35936f955282441596cd758f