Skip to content

Two HYD scenarios: yellow EDP + blue elec pump overheat (#88, #89) - #92

Merged
Santisoutoo merged 1 commit into
devfrom
feat/hyd-scenarios
Jul 24, 2026
Merged

Two HYD scenarios: yellow EDP + blue elec pump overheat (#88, #89)#92
Santisoutoo merged 1 commit into
devfrom
feat/hyd-scenarios

Conversation

@Santisoutoo

Copy link
Copy Markdown
Owner

Implements #88 and #89 (slice H of epic #19). Cross-system coverage: the benchmark now spans ELEC (×2) and HYD (×2).

What

Two hydraulic pump-overheat scenarios on engines-running, both probed empirically on vendor pin 13bce4b (3 runs each, stable):

  • hyd/yellow_edp_overheat.yaml: the yellow engine-driven pump overheats; isolate it and let the PTU carry yellow from green. Killing the PTU is anti_procedure (drops yellow with no source) — the first scenario to exercise the scorer's mild severity — and shutting an engine is dangerous. Success: pump pb off, yellow ≥1450 psi (PTU holds ~2500), both engines running.
  • hyd/blue_epump_overheat.yaml: the blue electric pump overheats; pure containment — isolate the pump and accept the loss of blue (inconsequential parked; blue has no other ground source). Shutting an engine is dangerous, touching the healthy green/yellow pumps is anti_procedure. Success: pump pb off, engines running (no blue-pressure predicate — its loss is the accepted outcome).

Fidelity findings (in each scenario's source.notes): both overheat cautions latch — they do not retire on pb-off, as the real fault condition persists — so success is state-based, not caution-clearance. And the green reservoir leak was probed and DISCARDED: its PTU cross-compensation cascades into the yellow circuit (both systems fault by ~120 s), leaving no bounded stable state. That is the plan's discard criterion in action — a fidelity boundary, not a scenario, recorded rather than papered over.

Tests

59 passed (pytest bench/tests mcp/tests). Scripted procedure tests for both (procedure → all_passed=true; the yellow ignore-case → false), timing verified: the slow-manifesting overheats (~105 s yellow, ~45 s blue) drive the validity gate through settle_s, not after_setup_s.

Closes #88
Closes #89

🤖 Generated with Claude Code

https://claude.ai/code/session_014JCrwRbtN7UA5ijn13vPmm

…eat) (#88, #89)

Cross-system coverage for #19: two hydraulic pump-overheat scenarios on
engines-running, both probed empirically on pin 13bce4b (3 runs each).

yellow_edp_overheat: isolate the overheating yellow EDP and let the PTU
carry yellow from green; killing the PTU (anti_procedure) or shutting an
engine (dangerous) is the wrong move. First scenario to exercise the
scorer's mild severity. blue_epump_overheat: pure containment — isolate
the pump and accept the loss of blue (inconsequential parked).

Both overheat cautions LATCH (they do not retire on pb-off, as on the real
aircraft), so success is state-based (pump commanded off + pressures where
achievable + engines running), no ecam_clear_of. The green reservoir leak
was probed as a sibling and DISCARDED (documented in the yellow scenario's
source.notes): its PTU cross-compensation cascades into yellow, leaving no
bounded stable state — a fidelity boundary, not a scenario.

Scripted procedure + ignore tests for both; 59 green.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014JCrwRbtN7UA5ijn13vPmm
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant