[adamjohnwright]

This replaces #770 after its review (which replaced #769 and #768).

Changed from #770:

• Render page: not ready until the analysis is coloured #770's head did not build. It was committed from an index that held main's copies of five files, staged for a red run. That removed AnalysisService.settled, the views' markers and __renderState.analysis. They are restored here, and the committed tree equals the tested one.
• Reading order: the page reads the result's error before anything reads the result, which would throw the error with its URL and token.
• Gone or unread: the page tells a result that is gone (404/410, which a release removes) from one that could not be read just now.
• Token stripping: an error's URL loses a token in its path as well as its query.
• The 410 test asks for "gone", and checks that the token is not in the error.

With an analysis token, the render page reported ready as soon as a view was drawn. That was before its analysis colouring arrived, and, for a result holding one resource besides TOTAL, before the second load that colours it for that resource, as the pathway browser shows it. The render service then cached the figure under the token, uncoloured or coloured for TOTAL. A token whose result could not be read was dropped, and the figure was drawn with no colouring at all and reported ready.

What changes

• AnalysisService.settled: the result on show is the one the address settles on, including the default resource's second load.
• Each view says what it shows:
  • the diagram and the illustration say which analysis they last coloured;
  • the overview says whether its tree is the one fetched for the current token and resource.
• The render page:
  • waits on these before it is ready;
  • publishes what it carries in __renderState.analysis;
  • reports an error when the analysis cannot be read, saying whether it is gone;
  • strips any token from an error's URL;
  • leaves the reaction view, which is never coloured, out of all of this.
• Render cache key v12, so figures cached without their colouring are not served.

Tests (e2e, with a made-up token answered in the test, replaying the suite's Release 97 recordings)

• For a diagram, an illustration and the genome-wide view: the result for the default resource is held back 3 s, and the load each view's colouring comes from is held back 6 s. No load may arrive after the page says it is ready.
• A gone token (410) is an error, not a figure.

Proofs

• On the previous page code, all four tests fail.
• With only the diagram's own check removed, its test fails. This is new since Render page: not ready until the analysis is coloured #769: there, every load was held back alike, so a page ready on the result alone passed.
• The illustration colours itself in the same update that delivers its result, so no timing separates its own check from the result's. The check stays as a guard, and the code and the tests say so.

Belongs to pdf-export's spec 002 (the analysis report), whose plan makes this its step 0. It also fixes figures rendered with a token today. Moving the whole e2e suite to Release 98 is a separate change.

🤖 Generated with Claude Code