tfi_hackathon — the price of pro-rata

TPSA / TF Wind Hackathon 2026. Cost of EirGrid's pro-rata constraint-group split versus effectiveness-ordered dispatch, and a year-to-date "band" rule between them, measured on North-West Constraint Group 3.

Start here: project/MASTER.md (the full specification), then research/validation-prorata-brief-2026-09-08.md and research/formula-decisions-2026-09-09.md.

Headline results, in the order a reader wants them:

• project/out/EVIDENCE_PACK.md — the argument, consultation-shaped.
• project/out/WP_RESULTS.md — the equal-relief replay on real 2026 data. Pro rata spilled 8.57 % more than an effectiveness-ordered cut delivering identical flow relief.
• project/out/RESULTS.md — the synthetic-year simulation (the earlier, weaker line of evidence; keep for the frontier shape).
• project/verify/VERIFICATION.md — the ten independent checks of MASTER §10, and every discrepancy found (D1–D9).

Layout

• participant-kit/ organisers' kit (networks, flowmath, gridkit). Make a venv: python -m venv .venv && pip install -r requirements.txt.
• project/src/ our code: measurement.py (2026 SEM-O data), engine_prep.py + run_engine.py (network prep, baseline dispatch, PTDF/LODF, overloads), rules to follow.
• project/out/ results (CSV/JSON/PNG/notes). Large parquet tables are not in git; regenerate with python project/src/run_engine.py <case>.
• data/ inputs. Files over ~20 MB are git-ignored; data/README.md says how to pull them from SEM-O, data/synth_out_extra/README.md how the extra synthetic years were built.
• project/verify/ independent verification (MASTER §10). None of these scripts imports rules.py, metrics.py, measurement.py, compare.py or robustness.py.
• research/ sources, verification notes, go/no-go tests.

Cases: WP2024s42 (main), WP2033s42, WP2033s43. Group: NW CG3. See MASTER §0.

Two things that will bite a new machine:

• Windows: run everything with PYTHONUTF8=1, and git config core.longpaths true before cloning.
• SEM-O retention: the 2026-06-08 window can no longer be pulled in full (the API now starts 2026-06-09 23:00), so out/measurement_*.csv cannot be regenerated from the API. Keep the local raw BM files. See project/verify/v06_measurement.md.