NOTES · FROM THE LANDING PAGE
Every number on this page comes from simulations that ship with the platform and can be rerun in a minute. A claim we cannot reproduce is not on this page.
A simulated organisation with twenty problems, five of which really are the deep causes. Fifteen people work through it, two hundred times per row, deciding each pair the way this platform does: 75% of the votes cast. Because it is a simulation we know the right answer, so we can count how many of those five each method ends up with — the only honest way to score a process, since no real room has an answer key.
| The room ranks by what hurts today | Stop at the vote | Carry on and build the map | Map, with seats held for evidence |
|---|---|---|---|
| not at all | 3.1 of 5 | 4.3 of 5 | 4.8 of 5 |
| a little | 2.4 of 5 | 4.0 of 5 | 4.9 of 5 |
| mostly | 0.8 of 5 | 3.6 of 5 | 4.9 of 5 |
| almost entirely | none | 0.2 of 5 | 3.5 of 5 |
Read the bottom row. When a room ranks purely by what it feels, the vote names none of the five real causes, and building a map on top rescues only 0.2 — a cause dropped at the ballot cannot be recovered later. The last column is the fix: hold five shortlist places for overlooked ideas backed by a source, and the room finds 3.5 of 5 at its worst. And read the top row: even a room with no blind spot at all finds more by mapping than by voting, 4.3 against 3.1, because a pairwise question at 75% is a better instrument than a ranking.
Group size has a floor, not a ceiling. Panels of five find 3.0 of the five causes, fifteen find 4.9, and larger panels find the same. Below about a dozen a panel loses causes; above the size that can read one wall it gains nothing. That is why a large room is split into groups, and why a federation needs five rooms to shrug off one bad one.
There is a cliff, and it is human. When people get roughly one in five of the influence questions wrong, the method still finds 3.3 of the 5 causes. At one in three wrong, the result is no better than guessing: a fifteen-person majority can correct scattered mistakes, not a room that is mostly mistaken. This is why one question is put to a whole panel at a time, rather than fifty questions to one person.
Some answers come free. If the group has agreed that A influences B and B influences C, nobody needs to be asked about A and C — it already follows. That shortcut removes 5–24% of the questions, and across 800 paired runs it cost no accuracy at all.
Duplicates are an attack, even by accident. Let one popular complaint sit on the wall as three differently worded cards and it eats three times the attention. In simulation that halved the number of real causes reaching the map. Merging duplicates is not tidiness; it protects the ballot.
Same simulated organisation, twenty problems, five real causes. Fifteen people decide each pair at 75% of votes cast. Some seats belong to an attacker who reads the wall and votes honestly everywhere except where the truth hurts one cause — the smart version, the one an AI agent would play. Every number below is a hundred simulated rooms at that setting.
| seats against one cause | causes named | target in top five | flagged |
|---|---|---|---|
| none | 4.2 of 5 | 87% | 0% |
| two of fifteen | 3.8 of 5 | 63% | 85% |
| three of fifteen | 3.6 of 5 | 47% | 100% |
| four of fifteen | 3.5 of 5 | 19% | 100% |
| six of fifteen — captured | 3.5 of 5 | 26% | 32% |
| four of thirty, two groups | 4.8 of 5 | 98% | 48% |
| eight of thirty, two groups | 4.7 of 5 | 84% | 83% |
A minority cannot add an edge. It can push one cause down the map: two seats in fifteen keep it out of the five deepest a third of the time, four seats four times in five, and the rest of the map looks fine. But the same seats end up in the minority on every pair touching that cause, and the record flags that — 85–100% of attacked rooms, none of the honest ones, as a count, never a name. Six seats capture a group outright; the cure is the last two rows — a room of thirty is two groups dealt by lot, and a deadlock goes to the other one.
No script works. The pairs don't exist until the vote closes, so an attacker has to read and answer like everyone else, at three seconds a pair minimum. That stops a rush, not an AI. What an AI cannot supply is a person who vouches for it.
Rooms must be vouched for. Rooms are free to open, so fabricated rooms could outvote real ones on the shared map. A room’s verdicts count only when it belongs to a verified organisation, has a steward the instance trusts, or was vouched for by a room that already counts — and one name is one counted voice per question, however many rooms it opens. Every room is still shown on the merged map, marked counted or unverified, so a reader always sees who spoke and whose verdicts weighed. In simulation — five real rooms against five fabricated ones — the rule takes a real cause from surviving in 23 runs of 100 to 96.
Not shown by simulation. Hiding sticker counts during the vote changes nothing in the numbers — the evidence seats already catch what the crowd misses. It stays because the method votes in silence.
How careful people have to be. If each person gets the hard pairs right four times in five and the easy ones nine times in ten, one group recovers 4.2 of the five real causes and two groups recover 4.9. If people are wrong three times in ten, 70% of pairs never settle and the map is guesswork. So a pair a group cannot agree on is not put back to the same people: each side writes one sentence on what would be lost if the other side won, published unnamed beside the pair, and the pair goes to a different group, who read those and answer it fresh.