Paid pilot
Aethel Node proves that the next action is no longer a misdiagnosis.
Available only after the workflow is reliable and a measurable business outcome has been demonstrated on matched cases.
- Price the measured Φ (error reduction) against a predeclared baseline.
- Settlement is verified only after the client exports and accepts the proof record.
- No charge on ties or regressions.
Current status
Evidence collection. Pricing and settlement activate after one confirmed improved Run under safety constraints.
External confirmation
What an outside party can check today, without the team narrating it:
- Three Safety-proof error wins on /workbench/proof: Φ 0.014 (incomplete evidence pack), 0.065 (missing measurements) and 0.065 (provenance gap). Each is a cold Self() run against a reuse run on the same plant, and counts as a win only when the reuse final e is strictly below the cold final e. The page recomputes them in your browser from code in the repo.
- Recorded Level 4 runs can be replayed at /api/level4/verify (add ?run_id=…), which recomputes the stored e, e0, cold e, Φ and eligibility. That endpoint covers recorded runs, not the three proof plants above.
- The MCP commercial case, Φ +0.159 (sealed sequence 0.171 against a greedy cold run at 0.33), on /workbench/peer, compare, independent and independence: tick the cold-run checkbox, then the commercial-case checkbox. This starting plant is constructed by us, not customer data. An exhaustive search of all four-step orders also reaches 0.171, so it beats the greedy cold controller, not an exhaustive one. On B1's default start the same comparison gives Φ −0.051, not a win.
- The evidence-pack export on /workbench/result and the Φ > 0 pilot checklist on /workbench/pilot, which says "ready" only for an eligible run with an exportable pack and a strictly positive Φ.
Not yet externally confirmed:
- No paid pilot has happened, and no customer has validated the value. Nothing is settled or revenue.
- No third-party attestation. The audit summary is a heuristic mock scan, and no outside party has reviewed or replicated these results.
- Any economic benefit is self-reported by the customer, stored in their browser, and not verified by the engine or the ledger.
- The anchor is a dry run: the Merkle root is prepared, nothing is submitted on-chain. The micro-ledger holds test events, not customer settlements.
- The proof plants and the commercial case are our own constructed cases, and the Level 4 runs were made by us.
Evidence goes to an outside party only after a person reviews it: External confirmation workflow (Agent B produces the pack and links, a human reviews, then it is shared).
Request a pilot
Aethel Node proves that the next action is no longer a misdiagnosis. Start a pilot request by running your repeated workflow and reference through the client pipeline and exporting the Run. Pricing is set per pilot against the measured Φ; no charge on ties or regressions.
Request a pilot →One line on price: set per pilot against the measured Φ, no charge on ties or regressions. There is no fixed price and no self-serve checkout. Evidence first: /workbench/status.
Success metric: the run reaches the goal threshold e < 0.16 with a positive Φ vs cold Self(); no charge on ties or regressions.
Evidence you can share: a public proof URL (/workbench/result?run_id=…), a downloadable ledger, and read-only verify and anchor endpoints (/api/level4/verify, /api/level4/anchor); the anchor root is a prepared Merkle root, not on-chain.
Proof you can send: the run's public URL (/workbench/result?run_id=…), which anyone can check against /api/level4/verify, plus the copied metrics string (run, e, Δe, Φ) or the latest eligible ledger line.
Get the proof: open the public proof URL (/workbench/result), check it against /api/level4/verify, then Request a pilot →
Boundary density: the dynamic limit keeps per-element tolerance constant (ε = 0.16·√(N/4)), so tensor size alone does not trigger a misdiagnosis. proof page
Enterprise surfaces (read-only, off-chain micro-ledger): an asset map and an audit summary with counts by domain, signature status and settlement rail, plus the current Merkle root, which is not anchored on-chain. Audit results come from a heuristic mock scan, not a third-party attestation. Not settled, not revenue. Workspace · Asset map (JSON) · Audit summary (JSON)
Proof-to-pilot journey: 1. Declare the problem → 2. Bounded execution → 3. Verify the evidence → 4. Validate the value. Each step uses the existing engine and pages; an economic benefit the customer records is self-reported and not verified.
Proof-to-pilot journey and the two agent roles (acquisition and evidence, both human-approved): docs/agent-roles.md
Record actual economic benefit (optional)
Self-reported by you. By default it is stored only in this browser and is not written to the ledger. It is not verified by the engine. If present, it is added to the exported evidence pack and labelled self-reported. The ledger button below is opt-in.
What you get
- Live: a verified run on your repeated workflow, exported as an inspectable Run (client pipeline).
- Live: the per-step ledger (e, Δe, Φ, status) for that run, downloadable from status and result.
- Live: independent read-only checks: GET /api/level4/verify and GET /api/level4/anchor for the run.
- Directional: on-chain anchoring. The anchor endpoint returns a commitment root today; nothing is written on-chain.
- Pricing: set per pilot against the measured Φ; no charge on ties or regressions.
Client handoff: /workbench/client. Export the JSON from there before any commercial discussion.