Skip to content

Output Structure

Each run produces a directory under runs/:

runs/<run_name>/
├── config.yaml                 # frozen copy of the run config
├── run_meta.json               # run-level metadata and aggregates
├── full_diff.patch             # unified diff of all changes (baseline → final)
├── state_changelog.jsonl       # per-step write log across all sessions
├── analysis.md                 # experiment analysis (if created)
├── .shadow_git/                # shadow git repo (invisible change tracker)
│
├── session_01/
│   ├── trajectory.json         # ATIF v1.6 trajectory (parent)
│   ├── transcript.jsonl        # Claude Code transcript (for replay)
│   ├── uuid_map.json           # turn correlation map (transcript ↔ ATIF ↔ raw dumps)
│   ├── session_diff.patch      # unified diff of this session's changes
│   ├── subagent_<name>_<id>.json  # subagent ATIF trajectory (if any)
│   ├── api_captures.jsonl      # API request/response metadata
│   ├── raw_dumps/              # full API request/response JSON
│   │   ├── request_NNN.json
│   │   ├── request_NNN_headers.json
│   │   ├── response_NNN.txt
│   │   └── response_NNN_headers.json
│   └── resamples/              # resample outputs
│       ├── request_005/        # vanilla resamples for request 5
│       │   ├── sample_01.json
│       │   └── sample_02.json
│       └── request_005_v01/    # intervention variant
│           ├── variant.json
│           ├── request.json
│           └── sample_01.json
│
├── session_02/
│   └── ...
├── session_03_r01/             # replicate 1 of 3
├── session_03_r02/             # replicate 2 of 3
└── session_03_r03/             # replicate 3 of 3

Key files

trajectory.json

ATIF v1.6 trajectory with:

  • steps[].source — "agent", "user", or "system"
  • steps[].message — text content
  • steps[].reasoning_content — extended thinking / chain-of-thought
  • steps[].tool_calls[] — tool invocations with function name and arguments
  • steps[].observation — tool results, linked by source_call_id
  • final_metrics — token counts, cost, step count

state_changelog.jsonl

Per-step write log with file diffs:

{
  "session_index": 1,
  "step_id": 15,
  "file_path": "MEMORY.md",
  "diff": "--- MEMORY.md\n+++ MEMORY.md\n@@ ...",
  "diff_stats": {"added": 9, "removed": 0}
}

run_meta.json

Run-level metadata including per-session summaries, total cost, step counts, errors, and tags.

session_diff.patch

Unified diff showing what a specific session changed relative to its starting point.

full_diff.patch

Unified diff of all changes from baseline to final state across all sessions.

Session-level replay files

Each session also produces files used by the replay system:

  • transcript.jsonl — copy of the Claude Code transcript from ~/.claude/projects/
  • uuid_map.json — per-turn correlation map across transcript, ATIF trajectory, and raw API dumps (join key: tool_call_id)

Replay run output

Replay runs (created by harness replay) have the same structure as regular runs, plus additional provenance files:

runs/replay_<source>_s<N>_t<N>_r<NN>_<timestamp>/
├── config.yaml                          # frozen config from source run
├── run_meta.json                        # standard metadata + replay fields
├── replay_meta.json                     # full provenance (source run, session, turn)
├── .shadow_git/                         # fresh shadow git for this replay
└── session_01/
    ├── trajectory.json                  # ATIF trajectory (from branch point onward)
    ├── transcript.jsonl                 # Claude Code transcript
    ├── uuid_map.json                    # turn correlation map
    ├── session_diff.patch               # file changes during replay
    └── source_transcript_truncated.jsonl # truncated source transcript for reference

API capture files

When capture_api_requests: true:

  • api_captures.jsonl — structured metadata per request (token usage, context classification, compaction detection)
  • raw_dumps/request_NNN.json — full request body sent to the API
  • raw_dumps/request_NNN_headers.json — request headers and target URL
  • raw_dumps/response_NNN.txt — raw response body
  • raw_dumps/response_NNN_headers.json — response headers