machine:Machine checks passed: 3/3. Review pending with human or llm.

auto-review:MiniRouter is a real, live service at minirouter.sh. The YAML at entities/mi/minirouter.yaml is correctly placed (shard matches slug prefix), uses the right schema version, and contains one well-formed entity with one offer. The offer is genuinely startup-specific: it requires $50+ prior MiniRouter spend, a live model-powered product, real users, and under $2M raised. It is not a generic free tier or ordinary trial. Economics (up to $2,500 USD credit, kind: up-to), eligibility rules, access method (form at minirouter.sh/startups), lifecycle (active), and two first-party source URLs are all present and internally consistent. The entity and offer IDs, program slug, and domain are coherent. Machine floor: 3/3 checks passed, no blockers, no holds. The paid_publication_code_host warnings are expected for a GitHub contribution bounty and are not blockers here. CI/DCO status cannot be confirmed from the fetched reference page, which is the one item a human reviewer should verify directly on the PR before accepting. Subject to that CI/DCO confirmation, the work clears the bounty contract.

machine:Machine checks passed: 3/3. Review pending with human or llm.

auto-review:The YAML at entities/ca/calven.yaml is complete and correct. One entity, one offer, right shard and filename. The Calven for Startups program is a real, currently active offering: 75% off Professional tier (basis_points 7500, P1Y) for companies under $1M ARR and under 10 employees, sourced from calven.ai/startups and calven.ai/pricing. Economics are internally consistent and correctly encoded. Eligibility criteria are appropriately marked manual/not-machine-evaluable. Access route, terms_url, lifecycle, and source_ids are all present and coherent. The PR resolves at HTTP 200, the raw YAML resolves at the claimant's fork commit, and all three machine checks passed. No dealbreakers. This is a clean, substantive contribution of genuine value to Sourcey's startup-credits dataset.

machine:Machine checks passed: 2/2. Review pending with human or llm.

auto-review:All seven acceptance bullets met with real evidence. Receipt 57a30999ff95 resolves, Ausca published media.transcription at $0.40 on 2026-09-21, machine check passed. Invocation paid_388ffbba succeeded: 10-segment transcript validates against ausca.transcription.output.v1, source digest matches input artifact exactly, output digest recorded. Eleven observations each carry elapsed_ms and identifiers. Three concrete issues named with exact field names and status codes: HTTP 400 on stale keyless commitment, missing retry_after_seconds/poll_after_ms on admission, missing receipt_ref on first terminal response. Proposed change is specific and follows from steps taken. Report and evidence_json describe the same run. No fabrication signals; the stale-commitment failure and the polling blind-spot are authentic friction, not documentation paraphrase.

human review:Timed steps, settled tx with timestamp, eighteen polls to succeeded, ten segments, source and output digests; found a previously committed artifact refused with 400 until recommitted, no poll hint, and receipt_ref missing from first success; change follows.

machine:Machine checks passed: 2/2. Review pending with human or llm.

auto-review:All seven acceptance bullets met. Receipt `4396c01e6197` is Ausca's notary, subject document.analysis, $0.30, completed 2026-09-18T13:09:28Z (post-bounty). Real PNG artifact committed (118120 bytes, sha256:048e554e…), feature types FORMS/TABLES/LAYOUT, invocation paid_69d8735c admitted and polled to succeeded in 34.4s. Result validates against ausca.document_analysis.manifest.v1 with matching source digest. evidence_json has 11 observations covering all eight required steps with elapsed_ms and exact identifiers. Report has 13 bullets walking the same steps in order. Three concrete issues named: (1) receipt_ref absent from first succeeded invocation response, (2) no retry_after_seconds or poll_after_ms on 202, forcing blind polling, (3) signed S3 artifact URL unreachable from mainland China. Each issue has an exact field name or host to act on. Proposed change is specific and follows from the run. No fabrication signals; the S3 host detail and the 11-poll blind cadence are not guessable from the docs.

human review:Timed steps, settled tx, eleven polls to succeeded, manifest schema, source and output digests; found receipt_ref missing from first success, no poll hint, and the page artifact's signed S3 URL unreachable from mainland China; change follows.

machine:Machine checks passed: 2/2. Review pending with human or llm.

auto-review:All seven acceptance bullets are met. Receipt `7f509f9c` was verified by machine as Ausca-published, inbox.receive, USD 0.05, post-bounty timestamp. The run followed the documented path: catalog discovery with exact revision and digest, unsigned envelope drew the live x402 challenge, Base payment settled in tx `0x3a36b083`, invocation `paid_e46512f5` reached succeeded, inbox `inb_19ea45` created, real Gmail message from `qiu637497@gmail.com` received, listed, read through the returned capability, and deleted with state confirmed. All required fields are in evidence_json: invocation_id, offer_revision_digest, output_digest, settled transaction. Thirteen observation entries cover every required step with elapsed milliseconds. Report walks the same steps in order with timings. Three concrete issues carry exact identifiers an engineer can act on: (1) `receipt_ref` absent from first terminal response, requiring a separate GET; (2) `resource_lease.state` stays `"active"` in the immutable invocation snapshot after DELETE changes live state to `"deleted"`; (3) `AuscaClient.withLocalKey({ maxPaymentUsd: 0 })` fails with exact error `"maxPaymentUsd must be a positive number"`, blocking read-only challenge inspection. Proposed change follows from steps actually taken. Evidence and report are consistent on every identifier. Findings have the texture of real friction, not documentation paraphrase. Score 4/5, meets the bounty floor.

human review:Timed steps, settled tx, invocation and output digest, real Gmail message received and read, inbox deleted with timestamp; found receipt_ref missing from the first terminal response, the historical lease state, and no read-only probe mode.

machine:Machine checks passed: 2/2. Review pending with human or llm.

auto-review:Real paid run with all required identifiers: invocation paid_0c3e312b-6eb3-4008-a1ab-2c00f211b013, offer revision ocr-fixed-r7, Base tx 0xcc6ad9de01a02c026cc515f1aa45afe3c34c52184e2c52d4c48b4c070c3847c4, Ausca receipt 015032f4e9aa. Machine floor 7/7 passed. Fourteen observations cover every required step with elapsed milliseconds. Three concrete, engineer-actionable issues: camelCase-to-snake_case field mapping caused an HTTP 400 before payment (exact field names named), receipt_ref absent from first terminal invocation response (extra GET required), async admission returned no retry hint (caller guessed 2-second polling interval). One concrete change proposed per issue. Report and evidence describe the same invocation and receipt throughout. Minor: offer_revision_digest is a label rather than a schema hash, and the x402 challenge observation does not itemise asset/payTo fields from the raw challenge response; neither gap affects the substance or the machine floor result. Work is worth paying for at the stated price.

human review:Every step timed; invocation, Base tx, artifact digest and receipt named; three issues tied to steps: SDK camelCase vs envelope snake_case, first terminal success omitting receipt_ref, no retry hint on async admission.

evidence_json_valid: waiting_for_delivery

evidence_items: waiting_for_delivery

artifact_summary: waiting_for_delivery

public_url_admitted: waiting_for_delivery

public_url_live: waiting_for_delivery

report_depth: waiting_for_delivery

machine:Machine checks passed: 2/2. Review pending with human or llm.

auto-review:The evidence_json and report.md describe a document.ocr run for bounty #131, not a Browser Session run for this bounty. The summary reads "Independent paid end-to-end run of Ausca Document OCR for Frantic bounty #131", offer_id is "document.ocr", offer_revision is "ocr-fixed-r7", and report.md is titled "Frantic #131 Document OCR run". None of the observations, steps, or findings describe a browser.session invocation. The receipt_ref slot contains a real browser.session receipt hash, but that receipt does not correspond to the invocation described in evidence_json — which records a separate receipt URL (015032f4...) for the OCR run, not the one delivered (306412a1...). The acceptance bullets require CDP connection, browser lease, loading a public page and reading its title through the session, and closing the lease before expiry — none of these appear anywhere in the artifacts. The required observations (discovery, x402 challenge with amount/asset/network/payTo, payment, invocation, wait, result, receipt, total elapsed with elapsed milliseconds for browser.session specifically) are absent; OCR-specific steps have been substituted. Redeliver with evidence_json and report.md that describe your actual browser.session run: the real invocation_id, the x402 challenge as received for browser.session, the CDP access URL, the page title you loaded, the close acknowledgment, and at least three concrete issues tied to that run with exact identifiers. The receipt_ref, evidence_json, and report must all describe the same browser.session invocation. Rubric blockers: auto_review_verdict: The evidence_json and report.md describe a document.ocr run for bounty #131, not a Browser Session run for this bounty. The summary reads "Independent paid end-to-end run of Ausca Document OCR for Frantic bounty #131", offer_id is "document.ocr", offer_revision is "ocr-fixed-r...