Parth TiwariStart a project
MenuClose

Notes / Errata and writing

Things went wrong.
I wrote them down.

What I predicted, what happened, and what I had to change. Writing lives here too when there is something worth publishing.

Corrections
12
General posts
Coming soon

12 records shown

Filed 28 August 2026

What went wrong

Not victory laps. Corrections with the source boundary still attached.

  1. 01The consent box I should not have removedA cleaner upload flow removed a constraint the product was responsible for keeping.1 internally audited sourceBeatMind
  2. 02The explanation needs the same inputSeparate transformations could make the API score and the explanation disagree.1 internally audited sourceFraud Risk Intelligence
  3. 03A refusal is part of the MedRAG resultFluent answers initially received more attention than evidence sufficiency.1 internally audited sourceMedRAG
  4. 04The atlas was broader than its contentThe OncoVerse concept looked complete before the source-backed directory was complete.1 internally audited sourceOncoVerse
  5. 05The auto-disable path was not wiredOracle Auto Provision had a stop helper in code but not the token it needed in the workflow.1 internally audited sourceOracle Auto Provision
  6. 06Authority is not the same as approvalThe portfolio diagram promised a universal human gate that the workflow did not implement.1 internally audited sourceOrder Supervisor
  7. 07The right lift, the wrong denominatorQueryPilot's correction result belonged to the core set, not the combined benchmark.1 internally audited sourceQueryPilot
  8. 08SecondSelf has to waitCareer automation became safer when waiting for a person became a durable state.1 internally audited sourceSecondSelf
  9. 09A live shell is not a working demoSpur Chat's public client could load while its backend interaction remained unverified.1 public sourceSpur Chat
  10. 10Tathya should not decide the verdictThe first product framing pushed a source record toward a conclusion it could not justify.1 internally audited sourceTathya
  11. 11UPI had two precision numbersA held-out model test and a seven-day replay were collapsed into one result.2 internally audited sourcesUPI Fraud Engine
  12. 12The faster Vivid profile was worseA large speed win failed the visual review that mattered.1 internally audited sourceVivid

Writing / Open shelf

There is nothing here yet.

Writing is coming soon. The errata are already open.

I will publish when a thought has survived the work, not to keep a schedule alive.