Burn Reports & reconciliation

Scan your existing agent logs

Burn Report reads Claude Code JSONL and Codex session logs. It reports usage, model mix, API-equivalent estimates and repeated-tool signals. Unknown models remain visible instead of receiving an invented price.

leash report
leash report --json

# Choose copies of your logs and save a private report:
leash report --claude-dir /path/to/claude-logs \
  --codex-dir /path/to/codex-logs --output ./leash-report

Share deliberately

The dashboard report gallery opens saved reports and lets you create or revoke a share link. CLI --share is an explicit opt-in to publish aggregate numbers through configured sync; ordinary scans are private.

Shared reports contain aggregates rather than prompts. Review the result before sending the link. History visibility and watermarking follow your configured billing plan.

leash report --share --output ./leash-report

Compare against the provider's records

Settings → Invoice reconciliation accepts a provider CSV and compares it with your workspace's stored usage. It does not ask for a provider key or store the uploaded CSV. Normalize the export to these columns, with UTC dates, exact model names and USD costs before tax or credits.

date,provider,model,cost_usd,currency
2026-10-01,openai,gpt-4o,1.2345,USD
2026-10-01,anthropic,claude-sonnet-4-20250514,2.3456,USD

Test whether an optimization actually saved money

The Compare runs section in the optimization workshop accepts baseline and candidate CSVs for the same task cases. Match every case ID, record actual cost and mark whether each run passed your checks. The comparison keeps task coverage and quality flags visible beside the cost ratio.

Passed flags are operator-reported; Leash does not independently evaluate whether the model solved the task. A ratio for those matched runs is evidence for that experiment, not a universal customer-saving guarantee. Keep prompts, tools, model versions and evaluation criteria controlled when comparing.

node scripts/compare-runs.mjs --baseline before.csv --candidate after.csv --output comparison.json

Export for your own analysis

The authenticated export endpoint provides JSON or CSV under the workspace's permitted history window. The source checkout includes a comparison harness that fails on unmatched groups, duplicate event IDs or differences outside the chosen tolerance.

curl --fail -H "Authorization: Bearer $LEASH_USER_TOKEN" \
  "https://YOUR_DOMAIN/v1/export?format=csv" -o leash-usage.csv
node scripts/reconcile-invoice.mjs --leash leash-usage.csv \
  --provider provider-usage.csv --threshold 2 --output comparison.json