C3: Failure inputs in the export (#30, #31)

[yaniv-kwb]

Do not merge before the release. Failure work stays out of main until C1–C5 are released together (roadmap decision D5). This PR targets main only because the pipeline deploys pull requests into main to dev (the production jobs run only for main itself). It is built on C2 (#29), so the diff also shows C1 and C2. Review only these commits:

• 7085d78 (C3: Reference test for the export package #30);
• a04e2f9 (C3: Failure inputs in the treatment table of the export #31);
• 305ec0f and 27b5268: fixes to C3: Reference test for the export package #30's test after CI failed;
• 5ca99b1, which only adds the C3 and C4 ticket files.

What changes

The export package records which failures a result is based on.

• Reference export test (C3: Reference test for the export package #30): a fixed benchmark assessment is exported through the application and compared with a reference stored in qmra/risk_assessment/tests/reference_export/:

  • the same files;
  • identical CSV tables and HTML report;
  • both plots present. They aren't compared byte for byte, because rendering can differ between machines.

  UPDATE_REFERENCE_EXPORT=1 rewrites the reference on purpose; the README next to it explains how. This is the check the "Export stability" Quality goal asks for.

• Failure inputs in the export (C3: Failure inputs in the treatment table of the export #31): treatments.csv gets two columns, "Failure frequency (days per year)" and "Failure duration (minutes)", with each step's values repeated on its three pathogen-group rows. The reference export is updated on purpose; only treatments.csv changed.

• HTML report unchanged: it shows no inputs at all, so it doesn't show failure inputs either (decided 2026-09-25). The roadmap's C3 entry is corrected accordingly.

C3 is a Small change without a Spec; the Tickets follow the roadmap entry and the decisions of 2026-09-25.

For the release notes

• The export package's treatments.csv has two new columns at the end: "Failure frequency (days per year)" and "Failure duration (minutes)".

Spec and Tickets

• Roadmap: docs/failure_roadmap.md, C3 (Small, no Spec)
• Tickets: C3: Reference test for the export package #30, C3: Failure inputs in the treatment table of the export #31 (local copies docs/tickets/10, 11)
• Builds on C2: C2: Failure calculation (#24, #25, #26) #29 (and C1: C1: Failure inputs per treatment step (#21, #22, #23) #28)

Checks

• Mock-data check (Controlled server):
  • The full suite passes: 65 tests, including the reference export test and a new HTTP test.
  • An assessment with failure inputs exports exactly its values per step, and a step without them exports 0 and 30.
  • The reference test fails on any changed file, column or value and names the file (checked by changing one stored value).
  • Still to do: the manual check on dev (see below).
• Real-data check (Workstation): not done yet.
• Quality goals checked:
  • Export stability: the export package is now compared with a stored reference in every test run. This PR changes it on purpose (two new columns), with the reference updated in the same commit.
  • Traceability: the export now records each step's failure inputs, so a reader can see which failures a result is based on. Origin and reference columns stay out of scope.

Tolerance in the reference test: the first CI run failed. 30 best-case values differed from the reference by up to 1.2·10⁻⁵ relative, because tiny probabilities (down to 10⁻¹³) are computed as 1 − exp(−k·dose), where most digits cancel, and CI's processor and math library differ. That difference is about 10¹⁰ times smaller than the health-based reference level. The test now requires identical text (names, columns, labels, layout) and each number within 0.1 % relative; a 1 % change or a renamed column still fails. Making the calculation itself numerically more precise (expm1/log1p) would be a separate change to the calculation.

Manual check on dev

• Save an assessment with failure inputs on at least one step, download its export package, and check exposure-assessment/treatments.csv: two new columns with each step's values on all its rows, 0 and 30 for steps without failures.
• The HTML report in the package looks as before (no inputs).
• An assessment saved before C1 exports 0 and 30 for every step.

This deploy replaces C1 + C2 on dev with C1 + C2 + C3, so please tell the team.

Definition of Done

• Tests pass
• code-review findings are handled (fixed, or answered in this PR)
• Mock-data check done (automated checks done; the manual check on dev is open)
• Real-data check done, or marked not needed above
• No secrets or Real data committed
• Glossary, ADRs and Vision document updated if this Change affects them (none affected; the vision's MVP item 4 already says the export records the failure inputs)
• Relevant Quality goals checked
• I can explain this Change in my own words

🤖 Generated with Claude Code