awb 0.5.2

Benchmark harness measuring AI coding tool+workflow performance, not just model capability

AI Workflow Benchmark (AWB)

Measure AI coding tool+workflow performance, not just model capability.

_{Install from PyPI, validate 80 tasks, run vanilla vs custom, get capability profiles and improvement suggestions.}

---

Why This Exists

SWE-bench tests models. AWB tests workflows. The same model running vanilla Claude Code vs. a purpose-built setup with a tuned CLAUDE.md, hooks, and structured agents produces meaningfully different results on real engineering tasks. No existing benchmark captures that gap — they all evaluate the model in isolation.

AWB benchmarks the full stack: tool + configuration + workflow + model, together, on 100 tasks drawn from real open-source repositories.

Quick Start

pip install awb

awb quickstart                              # verify your setup
awb run --runs 3 --parallel --adaptive      # full 100-task benchmark (parallel, smart re-runs)
awb run --category workflow --runs 1        # workflow tasks only (quick test)
awb gap results/runs/<run_dir>/             # analyze capability gaps

How It Works

Clone repo at pinned SHA
  → Run setup commands
  → Capture baseline lint/security counts
  → Execute tool with task prompt
  → Run test suite + partial credit rubric
  → Sigmoid-normalize 7 metrics
  → Produce weighted composite + capability profile

Each task starts from a fresh git clone at a pinned commit. Every tool gets the same prompt, the same timeout, and the same verification suite. Results are scored with sigmoid normalization so scores are never negative and never collapse at the boundary.

Scoring System

Seven dimensions, sigmoid-normalized with per-task baselines derived from difficulty:

Dimension	Weight	What It Measures
Correctness	55%	Pass/fail (60%) + partial credit rubric (40%)
Cost efficiency	15%	Estimated USD per task
Speed	10%	Wall-clock seconds vs. estimated task time
Code quality	10%	Lint warning delta (pre vs. post)
Reliability	5%	Pre-existing tests broken by the change
Security	3%	New security issues introduced
Efficiency	2%	Tool turns used vs. task max

Sigmoid curve: score = 100 / (1 + exp(k * (value - baseline)))

• Optimal performance (excellent) → ~95
• Baseline performance (adequate) → ~50
• Above baseline → smooth decay, never negative

Difficulty-weighted aggregation: hard tasks count 2.5×, medium 1.5×, easy 1.0×. A tool that solves hard tasks beats one that only solves easy ones even if the easy-task count is higher.

Per-task baselines by difficulty:

Metric	Easy	Medium	Hard
Cost optimal / baseline	$0.05 / $0.30	$0.20 / $1.00	$1.00 / $3.00
Speed	50% / 100% of estimated_minutes	same	same
Iterations	3 / max_iters	8 / max_iters	15 / max_iters

The 80 Tasks

Real open-source repos, pinned to release tag SHAs. Setup runs in under 15 seconds via venv + pip (Python) or npm (TypeScript).

Category	Count	Easy / Med / Hard	What It Tests
bug-fix	12	7 / 1 / 4	Root cause analysis, test-first diagnosis, N+1 queries
feature-addition	9	3 / 0 / 6	Convention adherence, ambiguous requirements, Dockerfiles, TypeScript typing
refactoring	11	5 / 2 / 4	Multi-file consistency, O(n^2) optimization, CI/CD config, async migration
code-review	9	4 / 2 / 3	Security review (report-only), concurrency analysis, migration guides, OWASP
debugging	10	7 / 0 / 3	Performance profiling, regression bisection, stack trace diagnosis
multi-file	7	4 / 0 / 3	Merge conflicts, plugin systems, auth chains
legacy-code	12	9 / 0 / 3	SQLAlchemy 2.0 migration, 20-file codebase navigation, dead code removal
workflow	30	9 / 12 / 9	Completeness tracking, convention discovery, security methodology, context utilization, async safety, config extraction, test-driven implementation

Repos used: FastAPI, httpx, Flask, Starlette, Click, Pydantic, SQLAlchemy 2.0, Hono

Task IDs: BF-001–014 · FA-001–010 · RF-001–012 · CR-001–010 · DB-001–011 · MF-001–009 · LC-001–012 · WF-001–030

Capability Profiles

Each task maps to 1–3 capabilities, producing a radar chart of tool strengths:

Capability	Tasks	What It Measures
code_comprehension	41	Understanding existing code before modifying
framework_knowledge	35	Knowing API patterns (Pydantic v2, async SQLAlchemy, etc.)
bug_diagnosis	26	Structured root cause analysis, test-first diagnosis
refactoring_discipline	26	Changing code without breaking behavior
multi_file_reasoning	23	Coordinating changes across multiple files
completeness_tracking	10	Following all requirements, not stopping at 80%
convention_adherence	10	Discovering and following project conventions
security_methodology	10	Applying security checklists systematically
context_discovery	10	Reading project docs and config before editing
test_writing	10	Writing correct, meaningful tests
security_awareness	10	Identifying and fixing vulnerabilities
cost_discipline	derived	Token efficiency across all tasks

Example awb gap output:

Capability Profile
------------------
code_comprehension    ████████████████████  82.4  (n=27, conf=high)
framework_knowledge   ████████████████░░░░  68.1  (n=26, conf=high)
refactoring_discipline████████████████░░░░  65.3  (n=23, conf=high)
multi_file_reasoning  ████████████░░░░░░░░  51.2  (n=20, conf=high)
bug_diagnosis         ███████████████░░░░░  63.7  (n=17, conf=med)
test_writing          ██████████░░░░░░░░░░  44.1  (n=8,  conf=low)
security_awareness    █████████████░░░░░░░  55.8  (n=8,  conf=low)

Systematic Patterns
-------------------
- Fails 70%+ of multi_file_reasoning tasks → consider multi-agent workflows
- Token spend on failed hard tasks: $4.20 → add early-exit heuristics
- No failures on easy tasks → baseline is solid

Top Suggestions
---------------
1. Enable subagent mode for tasks spanning >3 files (impact: high)
2. Add repo-level CLAUDE.md with architecture overview (impact: medium)
3. Use --think flag for debugging tasks (impact: medium)

Workflow Lift Score

When awb run executes both vanilla and custom (the default), it produces a Workflow Lift — a single number measuring how much your workflow configuration improves over the raw model:

Workflow Lift: +4.2 pts  (p=0.031, significant)
  Pass rate: vanilla 62% vs custom 68%
  Wins: custom 8 / vanilla 3 / ties 69

  Where your workflow helps:
    bug diagnosis             +12.3 pts  (17 tasks)
    multi file reasoning       +8.1 pts  (20 tasks)
    security awareness         +5.4 pts  (10 tasks)

  Where it hurts:
    cost discipline            -4.2 pts  (80 tasks)

  Biggest task-level differences:
    BF-014   +40  (V=35 C=75)
    LC-012   +15  (V=65 C=80)

The lift is computed per-task (configured score minus vanilla score), averaged across all tasks, and tested for statistical significance. Capability-level breakdowns show where your workflow configuration actually helps vs. adds overhead.

CLI Reference

Command	Description
awb run [tool] [options]	Run benchmark tasks
awb gap <run_dir>	Analyze capability gaps and generate improvement suggestions
awb compare <run1> <run2>	Compare two runs with significance testing
awb export <run_dir> -o file.json	Export results in external submission format
awb submit <file.json>	Validate and display an external submission
awb compare-submissions <a.json> <b.json>	Cross-tool comparison with statistics
awb quickstart	Verify setup: tools available, tasks load, validation passes
awb info <task_id>	Show task details
awb tools	List registered adapters and availability
awb validate	Validate all task YAMLs against schema
awb leaderboard	Generate HTML leaderboard from run results
awb workflow <subcommand>	Export, validate, diff, or init workflow descriptors
awb stability <run_dirs>...	Per-task score stability report
awb calibrate-difficulty <run_dirs>... [--apply]	Recalibrate difficulty labels from empirical pass rates
awb calibrate-timeouts <run_dirs>... [--apply]	Tighten timeouts from empirical p95 data

Common options for awb run:

awb run                            # all tools, all tasks, 3 runs
awb run claude-code-custom         # single tool
awb run -t BF-001                  # single task
awb run --category legacy-code     # filter by category
awb run --difficulty hard          # filter by difficulty
awb run --capability bug_diagnosis # filter by capability
awb run --runs 1 --dry-run        # preview without executing
awb run --resume                   # skip tasks with existing results
awb run --parallel -j 4            # run 4 tasks concurrently
awb run --adaptive                 # re-run near-miss tasks (60-99%) after initial pass

Adding Tasks

Tasks live in awb/tasks/<category>/. Copy awb/tasks/_template.yaml:

id: BF-012
category: bug-fix
title: "Fix response_model silently dropping extra fields in FastAPI"
difficulty: easy
estimated_minutes: 15
languages: [python]
capabilities: [framework_knowledge, test_writing]

repo:
  url: "https://github.com/tiangolo/fastapi"
  commit: "628c34e0"
  setup_commands:
    - "python3 -m venv .venv && source .venv/bin/activate && pip install -e '.[all]'"

issue:
  description: |
    The endpoint's response_model silently strips extra fields...
  files_to_examine:
    - "fastapi/routing.py"

verification:
  test_commands:
    - "source .venv/bin/activate && python3 -m pytest tests/test_extra_fields.py -v"
  partial_credit:
    - criterion: "Uses Pydantic v2 ConfigDict"
      points: 50
      check: "grep -q 'ConfigDict' tests/test_extra_fields.py"
    - criterion: "Tests pass"
      points: 50
      check: "source .venv/bin/activate && python3 -m pytest tests/test_extra_fields.py -v"

constraints:
  max_iterations: 20
  timeout_seconds: 1800

Run awb validate to check your task before opening a PR. Full guide: CONTRIBUTING.md

Adding Tools

Implement the ToolAdapter ABC in awb/adapters/:

from awb.adapters.base import ToolAdapter, ToolResult
from pathlib import Path

class MyToolAdapter(ToolAdapter):
    name = "my-tool"
    display_name = "My Tool"

    async def execute(self, prompt: str, workspace: Path,
                      max_turns: int = 20, timeout_seconds: int = 1800) -> ToolResult:
        ...

    def check_available(self) -> bool:
        ...

    def get_config_hash(self) -> str:
        ...

Register in awb/adapters/registry.py and add an entry point in pyproject.toml.

External Submissions

Anyone can share results using the submission format defined in results/submission-schema.json:

awb run --runs 3
awb export results/runs/<run_dir>/ -o my-results.json
awb submit my-results.json                        # validate locally
awb compare-submissions a.json b.json             # compare with significance testing

The format captures tool version, model, hardware class, and per-task run results. Hardware classes (e.g., apple_m5_24gb, linux_x86_16gb) enable fair speed comparisons — only compared within the same tier.

Statistical Framework

• Confidence intervals via t-distribution (no scipy required for core scoring)
• Significance testing via sign test for paired tool comparison
• Integrity checks: contamination detection (completions <10s flagged), variance anomalies (identical times/tokens across runs)
• Weight profiles: default, correctness_focused, production (see awb/scoring/weights.yaml)
• Stability metric: per-task TaskStability (std_dev, score_range, is_unstable); high-variance tasks can be down-weighted in composite scoring

Links

• Methodology — Fair comparison principles, metric definitions, known limitations
• Architecture — Module graph, data models, pipeline diagrams
• Contributing — Adding tasks, tools, and submitting results
• PyPI — pip install awb

License

MIT