ONE TASK. EXPLICIT TRUST.
From work
to verified receipts.
Task identity follows the report, reviewer decisions, contribution allocation and worker invoices. A historical bridge payment cannot be relabeled as this task’s settlement.
The live workspace requires a private access token. Viewing this page makes no model calls, wallet requests or payments.
New · A real reviewer can start here.
Each invited reviewer runs the local browser console on their own computer, creates an off-chain review key, confirms the exact buyer and plan, fills blank judgments and downloads their signed response. Real participants and external identity certificates are still required.
How to participate · 中文步骤 ↗Free ACI connection check · recorded
Hardware, fresh nonce/keyset binding, expiry, measured Compose and the actual TLS channel passed. Key custody was skipped. No private input or inference request was sent.
Observed 2026-11-04T02:20:04.760Z. This record is not a portable attestation, reviewed release or trusted model-inference proof.
Read the recorded connection observation ↗What is verified today
| Area | Current evidence | Boundary |
|---|---|---|
| Current task checklist | Implemented · authenticated API and workspace | Fresh checks show the first missing release requirement. Real reviewers can use a local browser console, sign their own blank judgments and return responses. Mainnet and testnet payment facts remain distinct from current identity/release readiness. |
| One task | Implemented · public acceptance passed | Signed report → evaluation plan → quality → allocation → invoices → both-chain receipts. Historic payments cannot be reused. The public acceptance used synthetic judgments; no task-bound mainnet payment has been made. |
| Quality and contribution | Implemented · evaluator run pending | Every claim requires a reason and exact source context. Full and worker-removal variants are evaluated separately. An external identity authority and actual independent reviewers are required before payment. |
| Automatic task splits | Keyless LOCAL SANDBOX treasury · five transfers passed | The actual disposable treasury key was removed before one buyer budget funded three workers, a fee and unused-budget return. The native mainnet adapter additionally verifies keyless state at every transaction block. Mainnet is disabled by default and has not been funded. |
| Confidential inference | Real free ACI connection observed · private trial pending | Five hardware/channel discovery checks passed; key custody was skipped. An exact single-release quotation runner is also implemented. Private inference still needs independent OS/KMS/model appraisal, a real ACI credential and bounded spend. The dedicated Phala instance remains off. |
| Recovery and operations | Real cloud and offsite recovery passed | An encrypted live backup was restored into an isolated D1 database. Every business table was compared. A GitHub backup artifact was also downloaded, decrypted and restored. Background push keys are configured; organization login, email setup and real device delivery checks remain pending. |
| Wallet and refunds | Task invoice handoff · refund trial pending | Download exact invoice previews and per-source limits. Already submitted orders cannot produce another payment preview. Refund checks reverify the original deposit and reachable native receipt actions. No real same-task mainnet payment or cross-chain refund has been validated. |
New · A locked common budget.
A second LOCAL SANDBOX run removed every access key before funding and completed five exact transfers. The mainnet adapter requires the same locked-contract policy, an exact buyer authorization and a separate gas-sponsor approval. It is disabled by default. No mainnet escrow deployment or payment is claimed.
Inspect keyless escrow sandbox proof ↗Real reviewers start with a blank page.
Export the exact public report, source context and every full/removal sample from the live workspace. Scores, claim verdicts and decisions begin empty. Large criterion disagreements and opposite contribution directions block release. Distinct signing keys alone are not independent people.
The updated public Workers/D1 acceptance verified blank handoffs, fresh read-only checks, blocked release without external identities, retained evidence and revoked test access. It used synthetic judgments and made no payments or model calls.
Inspect public release workflow acceptance ↗One task. Five local transfers.
Three workers, one protocol fee and one unused-budget return were paid by the unchanged treasury contract in LOCAL SANDBOX. Synthetic review identities are explicitly recorded. Final callbacks and zero remaining credits were checked.
Inspect same-task sandbox evidence ↗Recovery, actually exercised.
The GitHub monitor and backup workflow passed. Its encrypted artifact was downloaded back from GitHub, decrypted and restored locally.
An encrypted production snapshot was decrypted locally and restored to a separate cloud database. All 21 business tables match the sanitized local restoration. The production database was not replaced.
Old sessions were revoked, queued jobs cancelled, invitations consumed, email disabled and payment flows held for reconciliation. The isolated drill database is retained. No restored worker was activated.
Read cloud recovery evidence ↗Checked 2026-11-03T10:30:51.595Z. This proves the recorded snapshot can be recovered; it is not a continuous availability guarantee.
Keep these receipts separate.
The successful 0.5 NEAR → 1.861840 native USDC transfer is a real independent payment demonstration. The original AI report, same-operator protocol acceptance, historical public TDX task and testnet treasury each retain their own scope. They do not add up to a completed independent, confidential, mainnet-settled AI task.