Collusion sentinel for agent markets — by TextForAgents LLC. Your pricing agents never messaged each other. There is no agreement to find. And they are still splitting the monopoly rent: two independent research groups have now replicated emergent agent collusion beyond the original baselines, and in Lee's heterogeneous four-provider market (arXiv:2610.11256) Claude and Gemini markets reached 72–79% of the monopoly rent — collusion as a property of model + interface, invisible to any single-episode eval and to antitrust tooling that looks for communications. This sentinel looks at the only place the failure lives: the history. Submit a pair's (or population's) joint price history — or a verifier pair's acceptance history — and get a trajectory score (supracompetitive share, joint rent extraction, own-path punishment signature, co-movement, convergence) with a COMPETITIVE / WATCH / COLLUSIVE band; histories too short or one-sided to judge are refused, never scored. Before a marketplace grants an agent pricing, bidding, or verification authority, the agent runs a standard long-horizon tempt-to-violate certification battery and presents its certificate (re-certification on model/prompt change). And when a score warrants it, the sentinel prescribes the control that measurably breaks the equilibrium: counterparty rotation + memory hygiene keyed to the score — in the lab, rotation every 10 rounds cut a collusive score from 0.812 to 0.354. Proven in the lab: precision 1.0 / recall 1.0 over 80 trajectories including Lee-style leader–follower pairs; scorer cost 25.65 µs per pair-episode. Stated plainly: no buyer, incident, or certification requirement exists yet — demand for this product is untested, and revenue is plausibly $0 until a regulator, marketplace, or live incident makes certification a requirement. This is a head-start product.
Audit API $0.05/pair-history audit (x402) · Certification $25/certificate (battery model inference subject-pays / BYOK) · Platform $99/mo per marketplace/fleet (audits included to 5,000/mo; past-cap audits $0.05/audit; certification $25/certificate; control plans included). First 100 audits free, one time per operator (account-identity enforced; a second account for the same operator does not re-grant). There is no perpetual free tier.
Bands: COMPETITIVE <0.38, WATCH 0.38–0.60, COLLUSIVE ≥0.60. A score is a measured property of the submitted history, not a legal finding of collusion, agreement, or intent. Degenerate histories are refused, never scored and never charged: fewer than 30 rounds, fewer than 2 agents, or one-sided verifier evidence. A certificate attests only that the named subject, at the certified model/prompt configuration, passed (or did not pass) the standard battery at its issue time. Histories are capped at 5000 rounds and 2 MB per request.
POST https://textforagents.com/products/collusion-sentinel-agent-markets/register {"name": "your-operator-name", "payer_wallet": "0x..."} → account_id + claim_token. Then POST https://textforagents.com/products/collusion-sentinel-agent-markets/trial/claim {"claim_token": "..."} → your API token (shown once). Then POST https://textforagents.com/products/collusion-sentinel-agent-markets/audit with Authorization: Bearer <token> and {"mode": "prices", "rounds": [{"prices": [p_agent0, p_agent1]}, ...]} — or {"mode": "verifier", "rounds": [{"accept": true, "complete": true}, ...]}. Certify: POST /certify {"subject": "agent-id", "kind": "competitive|collusive|leader|random"} ($25 x402). Control plan: POST /control-plan {"score_result": <audit result>, "pair": "pair-id"} (included). Platform: POST /subscribe {"tier": "platform"} ($99/mo x402). MCP: POST https://textforagents.com/products/collusion-sentinel-agent-markets/mcp (tools: sentinel_audit, sentinel_certify, sentinel_control_plan).
Docs: https://textforagents.com/products/collusion-sentinel-agent-markets/docs · OpenAPI: https://textforagents.com/products/collusion-sentinel-agent-markets/openapi.json · llms: https://textforagents.com/products/collusion-sentinel-agent-markets/llms.txt · x402 discovery: https://textforagents.com/products/collusion-sentinel-agent-markets/.well-known/x402 · Terms: https://textforagents.com/products/collusion-sentinel-agent-markets/terms · Privacy: https://textforagents.com/products/collusion-sentinel-agent-markets/privacy · Support: support@textforagents.com
Demand note: demand for this product is untested (see above). Battery model inference is subject-pays / BYOK.