PRODUCT SIGNAL
PROVISIONALGood public signals, with important decisions worth reopening.
READ THIS FIRST
Next: action plan / check index / decision evidence / scope & sources
THE FIRST THING WE'D CHALLENGE
THE FOUNDER CHECK
The interesting part isn't your grade. It's what the grade can't prove.
Public polish can hide implementation, runtime, authority, retry, or recovery decisions. The next step is not a prettier score. It's stronger evidence.
Use the action plan below with your team, or request a bounded review. Personal or work email is fine.
02 / ACTION PLAN
What to do next.
One review task you can assign. Suggested roles are not assigned owners, and completion does not authorize deployment.
- Suggested owner
- 02 · PROVE NEXT
- Done when
- Evidence to retain
- Exact artifact revision, test setup, expected versus observed result, and who reviewed it. Label the evidence stage; do not substitute a polished demo for a runtime test.
- 03 · WATCH FOR
- Reopen the question
- When the reviewed workflow, authority, dependency, or source-of-truth contract changes. Record the change and repeat the relevant check.
Need a second review? Request one at takyejun.com/research/ai-readiness/quick-report. Personal or work email is fine. Verification alone starts no analysis; selection is manual.
WHY THIS GRADE
Four views. One public signal.
Scores cover only assessable, scored public checks. Coverage is their weighted assessable share, not the percentage of your product reviewed. The evidence index is a rubric value, not a probability of correctness or safety.
Every check, accounted for.
EVIDENCE
Evidence to resolve.
These checks need stronger evidence, not an automatic fail. Each retains its observation, next proof, source reference, and claim boundary.
The brief expands unresolved findings. Matched signals and advisory checks remain in the check index. The optional full-trace appendix preserves every finding.
UNKNOWN ≠ FAILED
We refused to guess.
Public evidence can only go so far. These items were not penalized as failures; a real H.A.R.D. review can move them through stronger evidence without pretending one stage proves another.
A design description is not implementation evidence. An implementation is not runtime proof. The stage stays explicit so polish cannot silently upgrade maturity.
IF THE PUBLIC SCAN FOUND SOMETHING YOU CAN'T SHAKE
Bring us the version you're nervous to ship.
We'll try to break one consequential decision, trace the failure path, and tell you what would prove it. After email verification, Yejun manually decides whether it fits the free bounded review. No deck required. Verification alone starts nothing.
Personal or work email is fine. Your email is used to verify and follow up on this request; verification starts no model analysis. Do not submit credentials or private URLs.
- 01Verify your emailPersonal or company email is fine. The signed link only confirms you're reachable.
- 02Yejun reviews the requestNo automated inference starts from verification. The deeper review is manually selected.
- 03If selected, pressure-test the buildFocus on a consequential decision, failure mode, and what would prove it.
Method boundary
What this report did — and deliberately did not do.
The public evidence index uses fixed rubric weights: public observation 0.25, documented language 0.45, walkthrough 0.65, implemented evidence 0.80, runtime-tested evidence 1.00. This scanner can retain only the first two; it does not establish the later stages. View weights are 25 / 30 / 30 / 15. Keyword matches are candidates for review, not semantic validation.
15 protocol criteria, deliberately not 15 automated checks. The public grader only uses criteria that can be bounded to retained public evidence. Criteria that require observing the current workflow or measuring human oversight work stay outside this grade and belong in a real H.A.R.D. review.
Surfaces reviewed 0
Interface only — this report reuses Plot design system primitives. Plot is not a Product Signal scoring source.