Demo

Run it yourself. It takes under 10 seconds.

Python 3.11 or newer, standard library only, no network. Sections 1–4 below come from a lab-prototype run on 2026-10-10. Section 5 was re-run at public aox-protocol@526916c on 2026-10-14.

Commands

git clone https://github.com/arigatoexpress/aox-protocol.git aox && cd aox
git checkout 526916c
python3 -m unittest discover -s tests      # Ran 95 tests ... OK
python3 demo.py                            # section 5 below

Sections 1–4 were captured from the lab prototype on 2026-10-10 and were not re-run against the public snapshot for these pages.

1 · Referee

aox.uppercase correct              -> OK
aox.uppercase claims passed, lazy  -> FAILED | output is not the uppercased input
aox.uppercase spent=True           -> FAILED | spent must be False
hostile non-object output          -> FAILED | output is not an object
task with no contract              -> FAILED | no contract for 'aox.unknown_task'
founder.research invented source   -> FAILED | sources must be exactly the supplied sources

2 · Synthetic market

synthetic_only=True real_agents=0 real_demand_evidence=False
d1 match: selected o-2-up (rank 1 of 2) under reputation policy
ledger events: 27 replay_matches: True

Score breakdown, credit balances and oracle totals are omitted here; this page reports percentages only. The independent oracle matched the replay.

3 · Tamper test

edited event #14 (accept) price -> 7
simulator replay   REJECTED: accept must name the rank-1 offer and its price
accounting oracle  REJECTED: hash mismatch at 14

This shows internal consistency, not tamper-proofing. Anyone could recompute an unkeyed chain.

4 · Matcher experiment (synthetic)

Same synthetic task set for both	Reputation-aware	Lowest price
Synthetic OK rate	78.3%	44.5%

Read this before quoting it. We invented the providers and their hidden quality. The setup favors reputation by design, because the cheapest provider is also the worst. Lowest price spends fewer simulated credits per OK task because FAILED tasks are refunded. This is a software test of a toy hypothesis. It is not evidence of demand or of real agent behaviour.

5 · Verdict, receipt and OK-only gate

python3 demo.py
3. Verdict OK. is_success=True. Only OK counts as success.
4. Receipt aox-verdict-receipt@1 field verdict=OK replayed hash=018eb601069b8f5215fe96eeaeda349e0a082a9a45b40b8e5c269de607b0215b.
5. Simulated release on the OK receipt: decision RELEASE settled False liveSettlementAllowed false mode simulated network eip155:84532.
6. FAILED junk: verdict FAILED decision DENY reason verdict_not_ok settled False
   UNKNOWN missing: verdict UNKNOWN decision DENY reason verdict_not_ok settled False
   UNKNOWN timed out: verdict UNKNOWN decision DENY reason verdict_not_ok settled False

This is a real run of demo.py at aox-protocol@526916c, re-checked 2026-10-14. Base Sepolia testnet only. The gate releases only for an aox-verdict-receipt@1 that replays with verdict OK. A bare “OK” is refused, and each receipt releases once. The default demo is simulated and makes no network call. A Base Sepolia transaction is off by default. It is an opt-in testnet-only path (python3 demo.py --live-testnet). Mainnet chain ids are denied. No live testnet settlement is shown here.