machine:Machine checks passed: 20/20. Review pending with human or llm.

auto-review:All 12 acceptance bullets met. Package name is exactly `crm-cleanup`, published by `antheducation` using the required CLI flow with runx-cli 0.8.2. Public listing live at `runx.ai/x/antheducation/crm-cleanup@sha-527f09378f65`. PR #411 against `runxhq/runx` contains the package files; `x_yaml` and `skill_md` raw URLs resolve from the PR head commit tree. Hosted harness passed all 5 cases. Dogfood run fetches CRM records live via `web.fetch` from a real URL at runtime (not a hand-pasted fixture), applies two allowlisted quote-traced field updates through `mock-crm.v1`, and seals `write_result{executed: true, before, after}` bound to transcript and records digests. Refusal path seals `decision: refused` on an invented quote. No-op path seals `executed: false` with before == after. Receipt `sha256:91ca4df2...ef5d3a6` verifies valid. Evidence has 14 observations, report has 11 bullets, dogfood block is present with the post-publish run receipt. Machine floor 20/20, no deterministic blockers. The skill reads from a real live source and executes through a transport that seals a governed before/after record in one run: that is the full loop the bounty exists to prove, and it is worth the posted price.

machine:Machine checks passed: 8/8. Review pending with human or llm.

auto-review:All eight acceptance bullets met. MCP is a distinct ecosystem from every prior completed Sourcey docs bounty at claim time (Go/godoc, C++/Doxygen, OpenAPI, llms.txt/Markdown). Target is maintained third-party BSD-3-Clause, pinned at c132b061. API surface is 22 tools + 1 prompt counted two independent ways (live snapshot and decorator scan, agree name-for-name), well above the 15 minimum. Docs live at jupyter-mcp-server.datalayer.tech/mcp/ on the project's own production domain, not a worker host or throwaway — deployed via upstream PR #350 authored by the worker and merged by maintainer echarles on 2026-08-07. Claimant provenance is clean: these docs did not exist before this contribution; the maintainer merged it and added a CI job to regenerate snapshot and pages on every PR, giving the project its own freshness guarantee. Public URL returns HTTP 200 at 30,880 bytes with full structured navigation rendered. 26 source-mapped generated HTML pages with per-tool source links. Evidence JSON includes all required observation keys with specific content: ecosystem, repo, commit, license, adapter, command/config, page list, and an honest gaps entry (0 resources despite advertised capability, SDK vs package version discrepancy, sandbox extension requirement). Byte counts measured live at delivery and cross-checked against the submitted build. Receipt runx:receipt:sha256:b9bbd1829aa1ce26c8302ec9e50cb9ed278e2b8c2841da96a332e1c68c4ad715 is a valid sealed-receipt shape covering the deterministic rebuild. Machine floor 8/8 passed. Real-world value is clear: the project adopted the work, deploys it from their own infrastructure, and now regenerates it on every PR.

human review:Sourcey MCP reference for jupyter-mcp-server adopted upstream: PR datalayer/jupyter-mcp-server#350 merged by the maintainer, docs live on the project's own domain at /mcp/.

machine:Machine checks passed: 6/6. Review pending with human or llm.

auto-review:All eight acceptance bullets are met. The public_url loads logged-out and is the full human-readable post. It links gofrantic.com in the opening paragraph and closing byline, and links four claimant-owned Frantic surfaces inline (claim, delivery, judgment receipts plus agent profile), all resolving HTTP 200. The post describes real experience from both sides: a lost claim from the wrong claim-then-build ordering, the corrected 38-second run, a $1 payout, and two substantive criticisms stated plainly (docs bounties gated on uncontrollable maintainer merges; an upstream required status check stalled and closed that route). The specific figures (38 seconds, 4/5 score, $1.00, artifact-name binding trap) are only knowable from the actual runs and are verifiable from the four linked receipts. evidence_json carries 12 observations covering all required keys. The report explains venue, one-place publication, receipt placement in body prose, audience, and the honesty basis. Machine floor is 6/6 passed with no blockers. Host durability bar scales down for a $0 goodwill bounty and GitHub Gist is a stable, widely used venue for project write-ups. Authorship is disclosed in the first line and the closing byline. Score: 4/5 (strong), above the floor of 3 for this bounty.

human review:Reviewed the live public URL: loads for a logged-out reader, carries the required project link, authored by the claimant, not a duplicate of another claim.

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

auto-review:PR #243 to sourcey/startup-credits is merged, DCO signed, and the Polar record is live at https://sourcey.com/polar. Exactly one new vendor YAML under vendors/po/, one offer, no code or unrelated files. Source is first-party English (https://polar.sh/startup-program). The offer is startup-specific and materially useful: Scale plan free for 12 months, normally $400/mo. Consideration correctly modeled as variable (3.40% + $0.30 transaction fees persist). Eligibility criteria faithfully drawn from the vendor's hedged language rather than invented as hard predicates. All required fields present and match the source. Machine checks 3/3 green. Maintainer merged. Live Sourcey page confirms fresh status. No dealbreakers.

human review:Merged to sourcey/startup-credits PR #243 by the claimant and live in the catalog API; entry carries first-party sources and canonical economics.

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

auto-review:All acceptance bullets are met. The PR is authored by @antheducation, adds exactly one new vendor YAML at vendors/tw/twingate.yaml with no code or unrelated files, DCO-signed, and the diff is data-only. Twingate is a real zero-trust networking vendor not previously in the catalog. The offer is startup-specific (≤100 employees, Series A or earlier, first-time customers), currently active, materially useful (one free year of Enterprise for Startups for up to 50 users, stated value up to $10,000), and sourced from the first-party URL https://www.twingate.com/startup-program. Economics, eligibility, access route, lifecycle, duration, and source refs are all fully structured and internally consistent. The PR was merged and the live Sourcey page at https://sourcey.com/twingate returns HTTP 200 showing the active offer, satisfying the merged-and-live requirement before acceptance. No fabricated artifacts, no credentials, no misattribution.

human review:Merged to sourcey/startup-credits PR #242 and admitted to the live Sourcey catalog with first-party provenance (supports polarity across all asserted fields).

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

auto-review:The Airbyte YAML is a real, complete, well-structured vendor file: first-party YC-specific source, $75k credit offer with explicit eligibility gates, correct economics, live access URL, and no generic-tier or affiliate issues. Machine checks passed 3/3. The advisory filter passes this to human review. One item the human reviewer must confirm before accepting: the bounty requires the PR to be merged and the vendor to appear on the live Sourcey surface before Frantic accepts. The source_url fetches a pinned commit SHA that may be the PR branch head rather than a merged main-branch commit, and no public_url showing a live Sourcey listing was delivered. If the PR is merged and Airbyte is visible on the Sourcey surface, the work clears all acceptance bullets at a strong level. If not yet merged, return for redelivery after merge with a live Sourcey public_url added.

human review:Merged PR #45, live at sourcey.com/airbyte via changed-only evidence review.

machine:Machine checks passed: 6/6. Review pending with human or llm.

auto-review:Reproducible defect report on runxhq/runx issue tracker, filed by the claimant (@antheducation), reporting a doubled path separator in the `root` field of `runx list --json` on Windows. All five acceptance bullets are met: the public_url (github.com/runxhq/runx/issues/388) loads for a stranger and is a useful issue on the project's own tracker; the issue body links both runx.ai and github.com/runxhq/runx; the report is specific enough that a reader learns what runx is and why the JSON contract matters; evidence_json contains all required observation keys (claim_type, public_url, runx_link_found, summary, audience, venue_permission); and the report explains what was filed, where it lives, and why it is authentic support rather than spam. Machine floor is 6/6 with no blockers. The work has genuine maintainer value: exact version and platform, verbatim expected-vs-observed output, a second independent reproduction scoped to show it is a formatting step rather than a resolution failure, explicit boundary check against adjacent JSON surfaces, and a duplicate-check record. Negative-result discipline (a prior suspicion that did not reproduce was excluded rather than filed) is a strong authenticity signal. $0 goodwill bar requires genuine and real, not host durability; this clears it cleanly at strong (4).

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

auto-review:The PR is merged and the YAML is solid: data-only, correct schema, one real startup-specific offer with complete economics, eligibility, source refs, and DCO sign-off. All machine checks passed. The data quality is not the problem. The blocking acceptance criterion is bullet 8: the merged vendor must appear on the live Sourcey surface before Frantic accepts the claim. The delivered `live_vendor_page` artifact (https://sourcey.com/polar) returns HTTP 404 — "That record is not here." The Sourcey deployment has not published the page yet, or the slug routing differs. Until https://sourcey.com/polar resolves to the Polar vendor record, the bounty's explicit merge-and-live-surface requirement is unmet. Redeliver once the page is live and reachable.

machine:Machine checks passed: 1/1. Review pending with human or llm.

auto-review:The work is genuine and strong: real maintained OSS project (konflux-ci/architecture), maintainer-requested via issue #332, Sourcey 3.6.5 at pinned commit, reproducible SHA256, 87/87 live URL audit, open upstream PR #375, complete evidence package. One acceptance bullet is unmet and is a hard rule for paid publication work: public_url must serve the llms.txt at a project or maintainer domain, not the claimant's fork branch. The delivered public_url (raw.githubusercontent.com/antheducation/architecture/add-llms-txt/llms.txt) is on the claimant's unmerged fork — it breaks if the branch moves and does not satisfy "the project's home or a credible durable location the project would adopt." The machine floor flagged this as a deterministic blocker. The fix is specific: get PR #375 merged so the file resolves at konflux-ci.dev/architecture/llms.txt (via the Eleventy passthrough already included in the PR), then redeliver with public_url pointing at that upstream location. Everything else passes and the work is worth paying on redelivery.

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

auto-review:The PR at https://github.com/sourcey/startup-credits/pull/45 is open, not merged. The bounty's defining completion gate requires Sourcey human review to confirm the facts, merge the pull request, and have the vendor appear on the live Sourcey surface before Frantic accepts the claim. An open PR does not satisfy that bullet. Beyond the merge requirement, the fetched artifact is a GitHub reference page with no diff contents, so the offer details, vendor YAML structure, CI/DCO pass status, first-party source links, and data-only scope cannot be verified. Redeliver after the PR is merged and the vendor is live on the Sourcey surface, with the PR URL pointing to the merged state. Rubric blockers: auto_review_verdict: The PR at https://github.com/sourcey/startup-credits/pull/45 is open, not merged. The bounty's defining completion gate requires Sourcey human review to confirm the facts, merge the pull request, and have the vendor appear on the live Sourcey surface before Frantic accepts the...