Skip to content

fix: align intersection slowdown and publish reproducible model results - #66

Merged
fjarmula merged 4 commits into
mainfrom
codex/shared-slowdown-results
Oct 31, 2026
Merged

fjarmula merged 4 commits into
mainfrom
codex/shared-slowdown-results

Conversation

@fjarmula

Copy link
Copy Markdown
Member

Both simulation modes now spend the configured rolling slowdown once per unique controlled intersection. Standard keeps its additional cautious crossing at red, yellow or unsupported movements; green decisions and failed priority requests preserve the independent common delay. This removes the artificial disadvantage previously applied only to GreenWave.

The reproducible runner fixes three GreenWave-ranked station candidates at density 1.5 and compares both modes at five densities for 200 deterministically sampled incidents. It exports signed actual driving-time savings, failures/on-site cases, all intersection strata, diagnostics, independent fastest stations and revision/dataset provenance. Walking and post-arrival recovery are excluded.

README and bilingual /docs publish the reviewed 3,000-pair / 6,000-simulation run: 89% faster, 11% tied, no slower cases; mean 71.94 s saved across all station ranks and 37.02 s for rank 1 across densities. The report records methodology, full strata, hashes and limitations. These are model outputs, not measured real-world benefits; repeated station/density combinations are not independent incidents.

The old submission history is verified in a local bundle. Main retains an explicit cleanup recipe and archived presentation changes for recreating submission from finalized main, preserving current application source and the current README. No old submission history is merged.

Validation:

  • Backend Ruff lint/format, mypy and 1,579 tests passed.
  • Frontend lint/typecheck/format, 1,311 tests and production build passed.
  • Regression coverage includes green/red/yellow, multi-line and adjacent intersections, origin/destination boundaries, fallback, exact Standard dry-run timing and live/final savings.
  • Actual-engine smoke passed 12 pairs / 24 simulations; the maintainer ran all 3,000 pairs successfully from clean revision e79afd5.
  • Independent motion, harness and publication reviews completed; findings fixed with regressions.
  • Playwright CLI verified both driving modes and results-page content/link/responsiveness in Polish and English at 1440, 393 and 320 px.

No real traffic-controller integration or production deployment is introduced by this change.

@fjarmula
fjarmula merged commit eb2ede6 into main Oct 31, 2026
7 checks passed
@fjarmula
fjarmula deleted the codex/shared-slowdown-results branch October 31, 2026 21:33
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