Preference vs. Support, as a live election — the rank can't see what the score can¶

Two elections with byte-identical rankings. The only thing that changes is how hard the two wings score the centrist Blair — a grudging 1 in one, a genuine 4 in the other. Because the orders never move, RCV-IRV and Ranked Robin return the exact same winners in both — they read only order, and the order didn't change. STAR is the only method that moves, because STAR reads the one thing that did change: support. This is the Preference vs. Support lesson made concrete.

▶ Live on BetterVoting: - Tolerated (Blair = 1): vote · results ↗ (election ywx39y, BV2225) - Supported (Blair = 4): vote · results ↗ (election 82gg36, BV2226)

→ Companion pages: Preference vs. Support · Scores vs. ranks · Center squeeze (one electorate, method varies) · What makes a good winner?

The ballots — identical rankings, one number changed¶

Three candidates on a spectrum: Alex (a pole), Blair (the center), Cole (the other pole). 36 voters, three blocs. Read the order off either score column and the two elections are the same ballot:

Voters	Ranking (identical in both)	TOLERATED scores	SUPPORTED scores
15	Alex > Blair > Cole	`5, 1, 0`	`5, 4, 0`
6	Blair > Alex > Cole	`1, 5, 0`	`1, 5, 0`
15	Cole > Blair > Alex	`0, 1, 5`	`0, 4, 5`

Only the wings' score for Blair changed: 1 → 4. Nobody's preference order moved a millimetre.

The result — one table, the whole lesson¶

Method	Reads…	TOLERATED (Blair = 1)	SUPPORTED (Blair = 4)	Moved?
RCV-IRV	order only	Alex	Alex	❌ can't — center-squeezes Blair both times
Ranked Robin	order only	Blair	Blair	❌ can't — Blair is the Condorcet winner both times
STAR	order + support	Alex	Blair	✅ the only method that responds

IRV and Ranked Robin cannot tell the two electorates apart — the rankings are identical, so their tallies are byte-for-byte identical. Only STAR registers that in the second election the wings genuinely support Blair (4s) rather than merely tolerate him (1s), and shifts the winner accordingly.

Read it even-handedly — no method is the villain¶

RCV-IRV eliminates Blair for having the fewest first-choices in both elections — the textbook center squeeze. It never sees Blair's second-choice strength at all.
Ranked Robin looks good here: it finds the consensus Blair from the order alone (he beats both poles 21–15 head-to-head). But it crowns a barely-tolerated Blair exactly as readily as a beloved one — it has no way to know the difference. That's the honest limit of an order-only method, not a knock on Ranked Robin's design.
STAR is the one that says: when Blair earns only 1s, he hasn't the support to win — elect the candidate with a real base (Alex); when Blair earns real 4s, he's earned the seat. Support, tracked.

The point isn't "STAR wins the argument." It's that support is real information, and only a score ballot carries it — the same claim the Preference vs. Support page makes with 1,0,1,0 vs 5,4,5,4, now as 36 real ballots you can re-run.

The engine says so — TOLERATED (Blair = 1) → STAR elects Alex¶

Abridged for the lesson — not verbatim engine output

Scoring Round
   Alex          -- 81 -- First place
   Cole          -- 75 -- Second place
   Blair         -- 60            ← thin support: Blair misses the runoff
 Alex and Cole advance.

Automatic Runoff Round
   Alex          -- 21 -- First place
   Cole          -- 15
 Alex wins.

[Divergence from STAR]
  STAR               = Alex
  RCV-RR (Condorcet) = Blair   (differs from STAR)

(RCV-IRV also elects Alex here — it squeezes Blair out. Full detail: bv2225_ywx39y_center_tolerated mirror · page.)

The engine says so — SUPPORTED (Blair = 4) → STAR elects Blair¶

Abridged for the lesson — not verbatim engine output

Scoring Round
   Blair         -- 150 -- First place   ← real support: Blair now leads
   Alex          --  81 -- Second place
   Cole          --  75
 Blair and Alex advance.

Automatic Runoff Round
   Blair         -- 21 -- First place
   Alex          -- 15
 Blair wins.

[Divergence from STAR]
  STAR                   = Blair
  RCV-IRV                = Alex   (differs from STAR)
  Note: Ranked Robin (RCV-RR) agrees with STAR, so RCV-IRV is the lone
        outlier — the classic center-squeeze signature.

(Same rankings as the tolerated election, so RCV-IRV → Alex and Ranked Robin → Blair are both unchanged; only STAR moved. Full detail: bv2226_82gg36_center_supported mirror · page.)

Reproduce it¶

python3 STARVote_LH_tabulation_engine/starvote_larry_hastings.py \
  method_comparisons/preference_vs_support/cases/bv2225_ywx39y_center_tolerated.yaml
python3 STARVote_LH_tabulation_engine/starvote_larry_hastings.py \
  method_comparisons/preference_vs_support/cases/bv2226_82gg36_center_supported.yaml

Each STAR file's [Divergence from STAR] block runs RCV-IRV and Ranked Robin on the same ballots — that's where the "same ranks, IRV & RR don't move, STAR does" table comes from. Both elections are also live on BetterVoting (results links above), and BetterVoting's own tabulations confirm every cell of the table — STAR → Alex/Blair, IRV → Alex/Alex, Ranked Robin → Blair/Blair — verified from the frozen full exports (…tolerated_bv_export.json · …supported_bv_export.json, each Election + Ballots + Results). BV and the LH engine agree exactly.