Summability demo — one example, three methods

Runnable elections behind the Summability topic hub. The point: summability is a property of the count, not the ballot. The same two districts are summable under STAR and Ranked Robin, and not under RCV-IRV.

Files

File(s)	Method	Summable?	Shows
irv_district_A.yaml, irv_district_B.yaml, irv_combined.yaml	RCV-IRV	❌	B wins both districts but is eliminated when merged — no precinct subtotal predicts it
star_district_A.yaml, star_district_B.yaml, star_combined.yaml	STAR	✅	precinct score totals + pairwise matrix add to the combined winner
rr_district_A.yaml, rr_district_B.yaml, rr_combined.yaml	Ranked Robin / RCV-RR	✅	the same ranked ballots IRV couldn't combine — the pairwise matrix adds cell by cell and recovers B

Note on the rr_* files. Ranked Robin runs on the same ranked ballots as IRV (irv_*) — only the count differs — so the rr_* files carry identical ballots to irv_* and just set voting_method: RankedRobin. They exist so the exact pairwise (For–Against–Equal) matrix in the RR write-up is reproducible on the LH engine (not only via the pref_voting cross-check): District A + District B add cell-by-cell to the Combined matrix, from which B beats A (11–9) and C (11–9). The folder stays a cross-method (IRV / STAR / RR) demo — it is not prefixed RCV_RR_.

Run them

# STAR (summable) — score totals + matrix (from the repo root)
uv run python STARVote_LH_tabulation_engine/starvote_larry_hastings.py method_comparisons/summability_demo/cases/star_combined.yaml

# RCV-IRV (not summable) — B eliminated when districts merge
uv run python STARVote_LH_tabulation_engine/starvote_larry_hastings.py method_comparisons/summability_demo/cases/irv_combined.yaml

# Ranked Robin on the SAME ballots — the summable pairwise matrix (LH engine)
python starvote_larry_hastings.py ../method_comparisons/summability_demo/cases/rr_combined.yaml

# Independent cross-check of the same ranked ballots (Copeland winner)
cd tools_adam/pref_voting_tabulation_engine
python pref_voting_tabulation.py ../../../method_comparisons/summability_demo/cases/rr_combined.yaml

The three write-ups

• STAR is summable
• Ranked Robin is summable
• IRV is not summable

The STAR files are guarded by the positive test harness; the rr_* files carry expected_winners (B for all three) and tabulate to it on the LH engine, confirmed independently by the pref_voting cross-check.