.png&w=256&q=75)
🤓 Latest Submissions
RedFirst: Proof-First Debugging
Problem Debugging waste is usually not the patch. It is shipping a change with no proof the bug was real. Developers and AI agents read a ticket, edit a handler, and stop when the happy-path suite stays green. That suite never covered the failing case. The ticket closes. The defect returns. Rework shows up in the next review or in production. This is visible in ordinary API work. GET /items/:id returned 200 with an empty body for unknown ids. PATCH /items/:id dropped every field the client did not resend, so price became undefined after a name-only patch. Visible Jest tests stayed green because they only sent full bodies and known ids. Solution RedFirst is a proof-first workflow on IBM Bob 2.0. The product is the .bob pack and the gated sequence, not a new SaaS UI. 1. Triage. Bob reads the ticket and maps handler, store, and test gap. Output: reports/<BUG-ID>/TRIAGE.md. 2. Reproduce. Edit lock to tests. Bob must write a failing repro and record STATUS: RED. 3. Fix. Only after RED. Smallest change. Re-run repro + full suite. Output: PROOF.md and PR.md. Target users Backend developers and reviewers who already use an AI IDE and keep merging unproven patches. How they interact Open the repo in Bob IDE. Pick a ticket. Run Triage → Reproduce → Fix. Do not ask Bob to “just fix it.” What is unique Most entries use Bob to generate a separate app. RedFirst uses Bob as the workflow engine. We measured that on the same defect: branch baseline-bug04 applied the correct one-line store merge and wrote zero tests. The RedFirst pass wrote sample-app/test/repro/BUG-04.test.js first (Expected 9.99, Received undefined), then applied the same one-liner. The suite now fails if the bug returns. Impact we actually measured - BUG-02: RED was 404 expected / 200 received; after the GET guard, 9/9 then 10/10 green with BUG-04 included. - BUG-04: raw Agent = fix, no test; RedFirst = RED test + same fix. - BUG-01 and BUG-03 remain seeded on purpose so the sample app is still a benchmark, not a cleaned demo. Repo: https://github.com/webneco/redfirst-ibm-bob
27 Sep 2026
