/ field notesASSESSMENT / EVIDENCE BEFORE SCORES
Know what
was verified.
Read the scope, test evidence and remaining boundaries together. Publication alone does not establish implementation readiness.
Provisional
The public design reference contains fourteen runnable page recipes and thirty-two component families. It has not met the canonical benchmark's all-scores 90 gate.
| Area | Result and boundary |
|---|---|
| Public coverage | 121 indexed public routes, two console entry probes and one deliberate 404; fourteen page families. |
| SVG review | Independent review passed 45 assets and 151 recorded contexts. The canonical vector audit also passed. |
| Designer Dave | Three fresh screenshot reviews each returned 7/10. The 9/10 visual target was not reached. |
| Geometry | All 32 families have source/local records at 390 and 1440 pixels; 21 have supported part correspondences at both widths. Correspondence is not full fidelity, and matched CSS rules do not establish each property’s winning declaration. |
| Behavior | The targeted browser suite covers search, form validation, dialogs, tabs, clipboard recovery, themes, navigation, local uncertainty states, reduced motion and difficult content. |
| Junior handoff | The tenth trial passes all 19 functional criteria, with 21 Independent and 3 Assisted stages; one of four complete tasks passes. The independent verifier supplied missing execution evidence. This was a context-limited Sol high trial. |
| Publication | The publisher rejected the upload because its shared deployment exceeds the configured aggregate-size limit. No live deployment or live verification is claimed. |
Open gates
- The exhaustive control audit has not established a pass. The generator now includes controls visible only at narrower widths. Incomplete runs, native-event limitations, search setup limits and every raw failure remain retained.
- The source-to-component CSS and anatomy mapping remains partial. Exact source evidence, local contracts and unresolved comparisons are retained separately.
- The tenth handoff trial passes functional acceptance, but three Verify stages are Assisted because the implementer omitted named contract checks. Current extension guidance clarifies those checks after the trial; no new handoff pass is claimed.
- This runtime cannot export the provider-originated roster and judge attestations required by the canonical benchmark. Local agent names and file hashes cannot substitute for those attestations.
- Publication and live checks are blocked by GSites shared capacity. The starter archive is below its separate 25 MiB per-file limit.
Review and verification evidence is retained in the research package. Publication is a provisional public-evidence deliverable, not a certified benchmark pass.
Evidence boundary
Public TypeSafe marketing, docs and console entry are in scope. Authenticated workspace, account permissions, billing, model runs and private operational behavior remain excluded. Source search, public CSS and asset recovery, and local reconstructions preserve a useful reference without claiming access that did not occur.
No live form was submitted, no account was created, and no paid model request was executed. All local state examples use synthetic data.
What a passing result requires
The canonical benchmark requires complete public inventories, current source-matched examples, working pointer and keyboard controls, reviewed assets, four independently implemented junior tasks, two fresh judges on one frozen identity, a live deployment crawl and a passing canonical validator. Every applicable native score must reach 90%.
Actual test results and remaining work are retained in the research package. Review findings are repaired and re-captured before a new freeze.
Open the four handoff tasksMaintenance
Version 0.1.0. Researched 18 Sep 2026. The reference maintainer owns this local package. Re-capture changed TypeSafe routes before treating older screenshots or source hashes as current. Verify asset rights and browser support before using the system outside this private research context.