ward test

Run the full invariant suite and write a JSON report.

Usage

ward test
# or, from source:
npm run ward -- test

What it does

1. Builds the scenario

   Loads config, constructs the chain adapter and facilitator client, and registers all eight invariants via buildScenario().

2. Runs every invariant sequentially

   runInvariants(invariants, ctx, { concurrent: false }) executes U1, U4, U2, U3, A1, U5, A2, A3 in that order (see D6 for why), catching and recording a failure per-invariant rather than aborting the whole run on one exception.

3. Renders a terminal report

   Prints a table grouped by category (universal / algorand), each row showing a PASS/FAIL badge, invariant id and name, duration, and a one-line evidence summary.

4. Writes the JSON report

   Writes both reports/ward-report-<ISO-timestamp>.json (a permanent, timestamped record) and reports/latest.json (always overwritten with the most recent run, read by ward report).

5. Sets the exit code

   Exits non-zero if any must-have invariant (U1, U2, U3, U4, A1) failed — safe to gate CI on. Stretch invariant failures (U5, A2, A3) are reported but don’t affect the exit code.

Example output

Ward report — self-hosted x402-avm facilitator (GoPlausible) (algorand-testnet)
run 7c1e...-...4f2a · 2026-08-17T09:12:03.441Z

UNIVERSAL
  PASS  U1  verify/settle consistency under post-verify mutation  (1834ms)
        network: OK, asset: OK, recipient: OK, amount: OK
  PASS  U2  retry safety  (2210ms)
        all 3 retries converged on a single transaction (ABCD...) — no double-settlement
  PASS  U3  concurrent double-settle  (1956ms)
        5 concurrent /settle calls resolved to 1 distinct on-chain transaction(s), 1 HTTP-level success(es) — exactly one winner, as expected
  PASS  U4  allowlist enforcement  (642ms)
        unsupported network (listed=false): rejected; unsupported asset: rejected
  PASS  U5  gas abuse resubmission  (3301ms)
        all 5 resubmissions rejected with bounded, consistent latency (max 340ms vs first 298ms)

ALGORAND
  PASS  A1  atomic group integrity  (2143ms)
        fee-payer leg amount > 0 ...: OK; ... : OK
  PASS  A2  ASA opt-in precheck  (511ms)
        verify correctly rejected before settlement would have burned a fee (...)
  PASS  A3  rekey authorization  (2876ms)
        settle correctly rejected the stale pre-rekey authorization (...)

8/8 passed, 0 failed

The written report’s shape

interface WardReport {
  runId: string;
  timestamp: string;
  facilitator: { name: string; commit: string; chain: string };
  results: InvariantResult[];  // one per invariant, with full evidence
  summary: { total: number; passed: number; failed: number; skipped: number };
}

Each InvariantResult.evidence contains the complete, per-sub-case request/response trail for that invariant — the same data ward fuzz prints for a single invariant, just for all eight at once. Numeric values that are JavaScript bigints (e.g. microAlgo amounts) are serialized to strings so the report is valid JSON.

Interpreting must-have failures in CI

npm run build && ward init && ward test
echo "exit code: $?"

A non-zero exit means at least one of U1, U2, U3, U4, or A1 failed — open the written report and check that invariant’s evidenceSummary and evidence fields first; each cites its source so you know exactly which guarantee was violated.

Isolate a single invariant

Once you know which invariant failed, re-run it alone with ward fuzz <id> for the full, verbose evidence trail.