Parth Tiwari — Notes and Errata https://parth-tiwari-1.vercel.app/notes/ What went wrong, what changed, and the evidence behind each correction. The consent box I should not have removed A cleaner upload flow removed a constraint the product was responsible for keeping. https://parth-tiwari-1.vercel.app/notes/beatmind-copyright-consent/ https://parth-tiwari-1.vercel.app/notes/beatmind-copyright-consent/ Fri, 28 Aug 2026 00:00:00 GMT The explanation needs the same input Separate transformations could make the API score and the explanation disagree. https://parth-tiwari-1.vercel.app/notes/fraud-risk-frozen-preprocessing/ https://parth-tiwari-1.vercel.app/notes/fraud-risk-frozen-preprocessing/ Fri, 28 Aug 2026 00:00:00 GMT A refusal is part of the MedRAG result Fluent answers initially received more attention than evidence sufficiency. https://parth-tiwari-1.vercel.app/notes/medrag-refusal-is-a-feature/ https://parth-tiwari-1.vercel.app/notes/medrag-refusal-is-a-feature/ Fri, 28 Aug 2026 00:00:00 GMT The atlas was broader than its content The OncoVerse concept looked complete before the source-backed directory was complete. https://parth-tiwari-1.vercel.app/notes/oncoverse-one-complete-entry/ https://parth-tiwari-1.vercel.app/notes/oncoverse-one-complete-entry/ Fri, 28 Aug 2026 00:00:00 GMT The auto-disable path was not wired Oracle Auto Provision had a stop helper in code but not the token it needed in the workflow. https://parth-tiwari-1.vercel.app/notes/oracle-auto-disable-not-wired/ https://parth-tiwari-1.vercel.app/notes/oracle-auto-disable-not-wired/ Fri, 28 Aug 2026 00:00:00 GMT Authority is not the same as approval The portfolio diagram promised a universal human gate that the workflow did not implement. https://parth-tiwari-1.vercel.app/notes/order-supervisor-authority-not-approval/ https://parth-tiwari-1.vercel.app/notes/order-supervisor-authority-not-approval/ Fri, 28 Aug 2026 00:00:00 GMT The right lift, the wrong denominator QueryPilot's correction result belonged to the core set, not the combined benchmark. https://parth-tiwari-1.vercel.app/notes/querypilot-denominator/ https://parth-tiwari-1.vercel.app/notes/querypilot-denominator/ Fri, 28 Aug 2026 00:00:00 GMT SecondSelf has to wait Career automation became safer when waiting for a person became a durable state. https://parth-tiwari-1.vercel.app/notes/secondself-human-gate/ https://parth-tiwari-1.vercel.app/notes/secondself-human-gate/ Fri, 28 Aug 2026 00:00:00 GMT A live shell is not a working demo Spur Chat's public client could load while its backend interaction remained unverified. https://parth-tiwari-1.vercel.app/notes/spur-demo-needs-an-end-to-end-check/ https://parth-tiwari-1.vercel.app/notes/spur-demo-needs-an-end-to-end-check/ Fri, 28 Aug 2026 00:00:00 GMT Tathya should not decide the verdict The first product framing pushed a source record toward a conclusion it could not justify. https://parth-tiwari-1.vercel.app/notes/tathya-no-verdict/ https://parth-tiwari-1.vercel.app/notes/tathya-no-verdict/ Fri, 28 Aug 2026 00:00:00 GMT UPI had two precision numbers A held-out model test and a seven-day replay were collapsed into one result. https://parth-tiwari-1.vercel.app/notes/upi-two-precisions/ https://parth-tiwari-1.vercel.app/notes/upi-two-precisions/ Fri, 28 Aug 2026 00:00:00 GMT The faster Vivid profile was worse A large speed win failed the visual review that mattered. https://parth-tiwari-1.vercel.app/notes/vivid-faster-was-worse/ https://parth-tiwari-1.vercel.app/notes/vivid-faster-was-worse/ Fri, 28 Aug 2026 00:00:00 GMT