The record
behind the beta.

A review followed by code and test improvements.

Daniel Reyes reported tightening Base64 secret validation, adding two edge-case tests and noticing a 400/422 contract change after a webhook review. His written feedback explains the changes he made.

Working software, inspectable engineering.

Release Engineer is live in early beta. Paste a public GitHub PR for a structured second review of release risks, testing gaps and breaking changes, with explicit limits and human verification. The application source and CI history are public.

Reports can be downloaded with the reviewed PR head and coverage limitations. A local verification worksheet lets reviewers record what they checked and did. These self-reported records become evidence only after separate source, permission and human review.

Public changes, checked against pinned sources.

An assistant-executed technical walkthrough compares two upstream PRs before and after their changes. It records TypeScript compiler diagnostics, Bun request behavior and 106 passing tests from two selected Hono request-test files, with commands and source revisions.

This is source and runtime verification. No Claude analysis or external developer session was performed; no finding, customer or model-quality claim is added. Failures and coverage limits remain visible in the record.

Approved developer feedback.

Developer quotes supplied by the founder and published with permission. Names, aliases and roles are publication-approved; contact details and PR links are not disclosed.

This is informal qualitative feedback. It does not independently verify findings, establish unique developer counts or demonstrate customer traction.

rails/rails #58968

A one-file, +1/−1 Rails documentation correction, classified as low risk with no findings and explicit coverage limits in a founder-run check. The reviewed head is pinned and the PR later merged upstream.

Founder test — not external validation. The original report and exact product commit are not in the public record. The upstream merge does not prove model accuracy; no Rails endorsement is claimed.

Claude reasons. Application code sets the scope.

Claude is the core reasoning engine for cross-file review, evidence versus inference, and structured findings with explicit uncertainty. Deterministic application code handles GitHub retrieval, bounded context selection, schema validation and coverage enforcement.

Evaluation with failures in view.

Two manually dispatched complete 32-case synthetic Claude runs are documented. The October 7, 2026 candidate completed 32/32 cases with 0 infrastructure errors, with 22 PASS / 10 FAIL under the strict evaluator and one critical violation remaining.

These are dated synthetic observations, not real-world accuracy, customer traction, production reliability proof or external validation.

How live usage is measured.

The server records content-free inference usage and emits operational events when an analysis starts, succeeds or fails. A success means the server prepared a schema-valid report. It does not prove correctness, browser delivery, a unique developer or a customer.

No verified production count is published. Publishing a count requires a preserved, complete production log window, deduplication and explicit exclusion or disclosure of operator testing.

A founder, a public record, a way to contact us.

Launched by Ozan Kenan Güngör in October 2026, in Ankara, Türkiye. Independently built and operated, self-funded and in early beta. No external investment has been raised. No legal company has been incorporated or registered. These project facts are founder-supplied.