Document the program evidence visual system - #165
Conversation
|
Reviewed. This is the strongest piece of the cluster — it is the only one that defines the system rather than implementing a corner of it, and the model is genuinely good: two timelines (program artifact vs one live run), four views, and one sharp invariant that the whole cluster turns on:
I used that sentence as the review criterion for the other five PRs, and it held up as a test. It is worth keeping in exactly those words. Not merging it yet, for ordering only. This documents and screenshots four surfaces. Two now exist on main (
Merging now publishes a guide on docs.openadapt.ai describing a workbench a reader cannot use in two of the four surfaces, with screenshots captured from local mock and development-only builds — which the PROVENANCE.txt entries state plainly, to their credit. Nothing is wrong with the document. It should merge last, once #415 and #397 land. Then the guide and the product agree on the day it publishes. |
… ledger #168 landed the palette guard while this branch was open, and this branch adds four images it has never measured. Three are cool. The fourth is not. program-workbench-desktop.png #f2f1ed 33.1% blue < red FAILS program-state-live-relation.png #f6f7fb 23.1% blue >= red program-workbench-cloud.png #0c1220 19.7% blue >= red program-workbench-cli.png #ffffff 26.5% blue >= red `#f2f1ed` is one digit off the retired `#f2f1ec`, and the second colour, `#fbfaf6` at 10.2%, is an exact retired value. Nobody can see that. The cause is ordinary: openadapt-desktop added the workbench in its #137 and moved to the canonical tokens in its #139, and the capture came from a build between them. Re-shot from `ebc10b0`, the #139 merge, at the same 1440x1000 through the same `?surface=program-workbench-preview` route. Ground is now `#f5f7fa` at 35.1%, which is the canonical `--surface`. Nothing was tinted and the guard was not touched. Ledger entries added for all four, so the two "recorded set == disk set" tests cover them. Each names the upstream repository, and the desktop entry names the commit and says why that commit is the one that matters. Two other things this pass found. `dashboard-overview.png` is vendored from openadapt-web public/product-preview/dashboard-workflows.png, and that file was re-shot in openadapt-web #405 tonight. The copy here is refreshed so the byte-identity the ledger claims is true again, and the PROVENANCE entry now names the Cloud commit it actually came from. The PROVENANCE header said every product-chrome capture is a byte-identical copy of an openadapt-web file. These four are not: openadapt-web publishes those surfaces as live components, not as stills. The header says so now. Verified: 289 tests pass, `measure_asset_palette.py` reports 14 of 14 matching, strict mkdocs build clean. Merging main brought no conflict in PROVENANCE.txt. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
1 participant
What changed
This documents the two-timeline model for a compiled program and a live run. It covers the program map, exact event occurrence, recorded-to-current disagreement, evidence, repair, and projection boundaries.
The guide includes screenshots from the website, Cloud, Flow, and Desktop. It also records the rendering tradeoffs and the threshold for adding a graph library.
Validation