The report of the fifth campaign of the evaluations, kept

[PierreMardon]

Kept for the next campaign to be compared with: evals/reports/2026-10-07-9d5de55f57a9/.

It measures the chain on main with the nine pull requests of 2026-10-07 that did the ten decisions of the developer (#97 to #105), none of which had been played. The six cases are played twice on one version for the first time: 02-reservations, 04-suspension, 05-fine-cap and 06-borrow-limit whole, 01-overdue-list and 03-overdue-reminders stopped at the hand over. 200 sessions, 30.89 USD at list price, the check of the judge included.

report.md and summary.json are as report --against <summary> --keep wrote them, by the harness of #106, #108 and #109, against the fourth campaign.

What it shows

• The eight runs played whole reached conformant and passed all of their hidden acceptance tests.
• The hand back at conformity is one to three sentences in the eight runs, 27 to 36 words where it was 115 to 158, names the files of the critical zones and ends on the developer's step (A hand back at conformity held to its one line, which ends on the developer's step #103). messages.next-step goes from 3 to 4.4.
• The words of the messages: no final message reports a cd (Prompts that say why commands run from the root, and no longer ban cd nor git -C #102), none gives the path of a report, and one names an agent, in twelve runs (Messages in the words the chain taught the developer, with its insides left out #105). messages.clear goes from 2 to 4.4, by a rubric that changed with the prompts.
• The assumptions of the plan: every page that holds a rule the plan settled on its own shows it as an assumption that approving confirms (A rule the plan settles on its own, marked in the plan and shown on the blueprint as an assumption #100).
• The diagrams: 05-fine-cap draws none in its two runs, where it drew one in each (A blueprint that draws no diagram for a chain with a single fork #101), and blueprint.diagrams is 5 on the twelve pages.
• The check of the judge: the lean page scores 4 and 4 on blueprint.no-padding, its padded version 2 and 2, where one point parted them. On the pages of the runs the score is 2 on ten and 3 on two, for facts said twice and no longer for what the frame asks for.

What did not move, and what it found

• following.no-noise, 2 to 3: the nineteen hand overs of the campaign each tell a step of the chain, checked, pushed, refreshed.
• The judge faults the hand back at conformity for not telling what was built, which is what the developer asked of it. The rubric says so in a pull request of its own.
• The proof of conformity lists a file of a critical zone under a blueprint that said none, in 3 runs of 8: A blueprint says the plan touches no critical zone, and the hand back at conformity sends the developer to read a file of one #107, which zones_announced tells since The code a developer is sent to read at conformity, measured against what the blueprint announced #108.
• No blueprint holds a title inside a section, and nine of twelve have a section of more than two paragraphs without one: The body of a blueprint is cut by the layout of the code, or left uncut where several things are decided separately #63, measured since The titles inside the sections of a blueprint, counted #109.
• 01-overdue-list takes a correction in its two runs, on an exit code the developer does not decide: the brief mended by A brief for the overdue list that no longer reads as a rule on wrong usage #85 did not remove it.

To say with it

• evals/README.md says what this campaign is worth to compare with: it is the first that plays every case twice on one version, and it was judged before the two lines the rubric draws for the hand back at conformity.
• Its ledger says that the weekly gauge rose by 6 points, and its streams show 3: the runs were played before A ledger that counts a point of the weekly gauge once when runs are played together #106. The figure is left as the ledger counted it, and the README says so.
• The campaign is twice the cost the index of the issues gave for it, 12 to 15 USD: it plays the six cases twice where four cases were planned, to close the issue on the spread between runs.

The conclusions, issue by issue, are in the issues #77 lists.