feat(prose-tests): judge the walk as told, not the summary returned

[leeovery]

Summary

• We were judging one message and calling it a walk. An agent returns a single final message; a walk runs across dozens of turns. Every step the walker narrated at the time and then compressed out of its closing summary read to the asserter as a step never taken.
• This was the under-reporting we spent a long time attributing to the walker, then to the model. It was neither. A Sonnet walk of continue-feature was marked down for skipping the selection menu — its transcript shows it emitted the display, emitted the menu verbatim, stopped, consumed the scripted answer, and validated the selection. An Opus walk of the same case made byte-identical tool calls.
• The runtime already writes every turn, and the stop payload names the file. The hook now lifts those turns into the world beside the action log. Nothing is asked of the walker — the same property that makes the action log trustworthy.
• So the walker is asked for less. Its final message no longer has to be a complete log (the instruction that caused the compression). Commands, outputs and file writes leave its list entirely — the harness records those with more fidelity. What remains is only what no tool call can show: the arm entered, the guard that selected it, what went on screen.
• The asserter drops "a thin narrative is not a missing step" — it papered over this blindness and is now false.
• A missing walk log is refused loudly, like a missing action log, rather than silently falling back to the closing summary.

Test plan

• node --test across recorder, corpus and snapshot suites — 40/40 (two new cases: turn ordering and preservation of mid-walk emissions; empty-transcript no-op)
• Conventions lint clean
• Verified against a live walk: the asserter prompt now contains Step 0: Initialisation, 0.1 Casing Conventions, 0.2 Boot with its Otherwise arm, and 0.3 Knowledge Gate — none of which appeared in the message the walker returned

🤖 Generated with Claude Code

Stack

1. docs(design): prose-tests programme design log #544
2. feat(prose-tests): the framework — cases, worlds, runner, skill #545
3. test(prose): feature happy-path corpus — five worlds, seven cases #546
4. test(prose): bugfix corpus — the investigation-centric surfaces #548
5. test: retry recursive teardown removals — kill a class of phantom failures #549
6. fix(entry-skills): close the handoff fences — six files render their arms wrong #550
7. docs: a contributing page for working on the system #551
8. fix(entry-skills): every handoff arm says to invoke the skill #552
9. fix(implementation): environment setup belongs to the setup reference alone #553
10. fix(prose-tests): the asserter is told which substitutions were armed #554
11. feat(prose-tests): the mid-flow substitution, and a world only prose can describe #555
12. test(prose): claims assert consequences, not what was displayed #556
13. feat(prose-tests): record everything the agents do, results included #557
14. fix(discussion-entry): the handoff reports the source it actually had #558
15. fix(prose-tests): the stop hook records, and names the model that walked #559
16. fix(prose-tests): command output was never actually recorded #560
17. feat(prose-tests): judge the walk as told, not the summary returned #561 👈 current
18. feat(prose-tests): decide in code what an agent should not be deciding #562
19. test(prose): a case starts where a session starts #563
20. feat(prose-tests): walk on Sonnet, judge on Opus, escalate a failure #564
21. test(prose): give the eight read-only cases something that can fail #565
22. test(prose): only walks that can be observed, and checks that survive the trip #566
23. fix(prose-tests): the verdict names only the model the record names #567
24. test(prose): discovery, walked to the point where work first exists #568
25. fix(prose-tests): the asserter judges which of prose or walker was at fault #569
26. docs(conventions): a step whose reference routes every exit still signposts #570
27. test(prose): discovery's epic arm, to the same durability boundary #571