neo4j/aip-skillbench-24task-sonnet-aipv0_3a3

AIP-SkillBench — 24-task combined cohort (Sonnet, AIP v0.3a3)

Raw evaluation-run data.

What this is

A head-to-head evaluation of two skill formats on the same tasks:

• human-curated — the task's original human-authored skill (prose).
• aip-from-curated — that same human skill compiled into AIP (a schema-validated execution-graph representation).

The 24-task stratified sample is run as three balanced 8-task cohorts (A, B, C). Each task × mode is run for 5 independent trials.

field	value
tasks	24 (cohorts A/B/C, 8 each)
modes	human-curated, aip-from-curated
trials	5 per task × mode
total runs	24 × 2 × 5 = 240
solver agent	claude-agent-acp
solver model	claude-sonnet-4-6
AIP authoring	claude-opus against AIP spec v0.3a3
sandbox	docker
benchmark	extends SkillsBench

Layout

cohort-a/   cohort-b/   cohort-c/      # one folder per balanced cohort, each:
  campaign.json     # the run matrix (tasks, modes, trials, model, agent)
  status.json       # run totals (done / pass / fail / error)
  summary.csv       # one row per trial — the primary table
  summary.jsonl     # same, JSON Lines
  cells/            # per-trial working dirs: rewards, timing, result.json, agent trajectory
  logs/             # per-trial solver logs

summary.csv columns: task, model, mode, trial, status, reward, n_tool_calls, wall_clock, error, jobs_dir, trial_dir, started_at, finished_at, subprocess_rc.

Source code, skills, and how to reproduce

The benchmark harness, run configs, and the AIP-compiled skills themselves live in the GitHub repo. Check out the matching tag to see the exact skills used for these runs:

• Repo: https://github.com/zach-blumenfeld/aip-skillbench
• Tag: sonnet-aipv0.3a3
• AIP-compiled skills: generated-skills/<task>/aip-from-curated/…
• Human skills: vendor/skillsbench/tasks/<task>/environment/skills/…
• Run configs: configs/eval-cohort-{a,b,c}-sonnet-aipv0_3a3.yaml

git clone https://github.com/zach-blumenfeld/aip-skillbench
cd aip-skillbench && git checkout sonnet-aipv0.3a3

Reading the data

import pandas as pd
a = pd.read_csv("hf://datasets/neo4j/aip-skillbench-24task-sonnet-aipv0_3a3/cohort-a/summary.csv")

Note on trajectory logs

Not all tasks in this dataset include full acp_trajectory.jsonl records.

All other artifacts are complete and unmodified, including every reward and summary record, so all reported metrics remain fully reproducible.

Contact zach.blumenfeld@neo4j.com if you need more information.