Skip to content

Bv2105r2 bv github issue

Title: A partial ballot whose marks are all equal is dropped from the tally as an abstention (BV2105 r4dqvd, re-confirmed on w3vvff)

Summary

A ballot that scores one candidate and leaves the rest blank is counted as an abstention and excluded from the tally entirely — its score never enters the totals. On a 4-ballot election this reports nTallyVotes 2 / nAbstentions 2 where the truth is 3 tallied and 1 abstention.

This is a tally-level defect, which is what separates it from the neighbouring abstain tickets: #1053 and #1090 are UI-label and export-ambiguity bugs whose underlying counts were correct. Here the count itself is wrong.

First seen on r4dqvd in Oct 2025. Because that election is closed (and a closed election may serve a stored result), I re-cast the identical ballots on a fresh election, w3vvff, on 2026-08-04. Same wrong numbers. So this is current behavior, not a historical artifact.

Environment

  • bettervoting.com, STAR with num_winners: 2 (Bloc STAR), 3 candidates, 4 ballots.
  • Original: r4dqvd (BV2105, created 2025-10-31, now closed).
  • Re-check: w3vvff (BV2105-r2, created 2026-08-04) — https://bettervoting.com/w3vvff/results

Steps to reproduce

  1. Create a STAR election, 2 winners, candidates Vanilla, Chocolate, Strawberry.
  2. Cast four ballots — one of each kind:
Voter Vanilla Chocolate Strawberry
1 5 5 5 all-5s
2 — — — fully blank: a true abstention
3 1 — — scores one candidate — the probe
4 2 5 4 an ordinary full ballot

(— = left blank, i.e. score: null.)

  1. Tabulate and read summaryData.

Expected vs. actual

Quantity Expected Actual (both r4dqvd and w3vvff)
nTallyVotes 3 2
nAbstentions 1 (voter 2 only) 2
Vanilla's total 8 (5 + 1 + 2) averaged over 2 ballots — the 1 is absent

Voter 3's ballot is discarded. You can see it is dropped from the tally rather than merely mislabeled: Vanilla's reported figure is an average over two ballots (5 and 2), not three — the 1 never enters the sum.

The winners are Chocolate, Strawberry either way, and the seat-2 score tiebreak is correct. The discarded ballot only helped Vanilla, who loses seat 2 regardless. The result is right; the count is not — but a cast ballot dropped from the tally can decide a closer election, which is why I'm filing it.

Why this may be #884 working as written — and why I think it's still wrong

Stating the counter-argument first, because it's a fair one. #884 established that a ballot whose marks are all equal counts as an abstention. A ballot bearing a single mark is trivially all-equal, so it falls through the same test. On that reading this is policy, not a defect.

I don't think that survives contact with how the tally treats blanks. BetterVoting counts a blank as 0. So voter 3's ballot isn't "one mark and two unknowns" — it is Vanilla 1, Chocolate 0, Strawberry 0, which is not all-equal and which strictly prefers Vanilla to both others. Judged against the scores actually counted, the all-equal test shouldn't fire at all.

Put plainly: a voter who rated exactly one candidate expressed a preference. Recording that as "no preference" both loses the score and overstates the abstention count.

Scope: this is about partial ballots whose non-blank marks are all equal. Partial ballots carrying two or more distinct marks count correctly today — e.g. 26khr3, whose Ada 5, Bruno 1, blank ballot is tallied normally.

Suggested fix / acceptance criteria

Apply the all-equal abstention test to the scores as tallied (blanks coerced to 0), not to the non-blank marks alone.

Acceptance test, using the ballots above:

  • nTallyVotes = 3, nAbstentions = 1.
  • Vanilla total 8, averaged over 3 ballots.
  • Voter 2's fully-blank ballot remains an abstention.
  • Winners unchanged: Chocolate, Strawberry.

Independent verification

The same four ballots through Larry Hastings' starvote engine (which abstains only a genuinely blank ballot) give 4 ballots, 1 abstention, Vanilla total 8, winners Chocolate + Strawberry — i.e. the same winners BetterVoting reports, from a correct count.

  • #884 — the all-equal abstention policy this most likely follows from. If the behavior here is intended, that's the thread to settle it on.
  • #1053 — equal-max (5,5) ballot labeled "Abstained" in the confirmation dialog. Same family, but a UI message; here the ballot is missing from the count.
  • #1090 — explicit all-0 ballot labeled "Abstained"; JSON correct. Again UI/export, not the tally.
  • #1407 — reconciling LH and BV reports on the pets election; overlapping territory for fully-flat ballots.
  • #1056 — not a duplicate, listed only to prevent the confusion I fell into: it is the demo-election 401 on ballot download (Editable Ballots regression, closed via #1058) and quotes the same BV2105 test document. Different defect, same election.

Happy to supply more ballot shapes if it helps pin the predicate — e.g. 1,1,— (two equal marks plus a blank) versus 1,—,—, which would separate "single mark" from "all non-blank marks equal."