Clubhouse methodology.
How the questions are made, who may answer them, and exactly what the numbers claim — published so the record can be evaluated, not merely believed.
Methodology v1 · applies to every Clubhouse question and receipt
Reality creates the question.
Every canonical Clubhouse question exists because something verifiable happened — a transaction, a decision point, a race with a real clock. A question engine normalizes league transactions into candidate questions; a human editor reviews every candidate before anything opens. No question publishes itself.
Before a question can open it must pass the Question Constitution, enforced as code:
- Materiality
- a real trigger.
- Evidence
- frozen, hashed snapshots of the underlying record.
- Neutrality
- loaded wording is rejected mechanically.
- Completeness
- the reasonable choices are all present.
- Reality
- impossible options cannot be offered.
- Timing
- the closing condition is declared up front.
- Provenance
- every entity the question claims to be about is pinned by canonical id.
After the first vote, wording and options are immutable. A material error produces a withdrawn or superseded record — the original is preserved, never edited away.
The commercial firewall.
No external party can create, word, time, sponsor, or alter a Clubhouse question. There is structurally no commercial field on a question — nothing to buy, no placement to sell. Question selection is editorial and engine-driven; it is never sold, never influenced by a partner, and never coordinated with a club.
Fan sentiment and Posterior’s own model assessment are separate products and are never blended. The Clubhouse carries no odds, no lines, and no picks. When a Clubhouse read and a Posterior model read appear on the same page, they are labeled as what they are.
Who may vote.
Voting requires an authenticated Posterior account — there are no anonymous votes. Each account holds one active vote per question. While a question is open you may change your vote: the latest vote is the one that counts, and every revision is preserved append-only in the permanent record. Voting closes when the declared closing condition fires; results then become public to everyone.
Results are revealed only after you vote — a deliberate anti-herd posture, so the numbers you see can never be the reason you voted. Internal test accounts vote normally for audit purposes and are structurally excluded from every aggregate.
Segments, floors, and the exact words.
Respondents declare a primary club. Each vote carries a snapshot of that declaration at cast time, so a club’s segment reads always mean the same thing: qualified Posterior respondents identifying that club as their primary club. We never say “fans” of a club, and we never call any cut a representative sample of one.
No percentage — overall or per segment — renders below n ≥ 25. Below the floor the surface shows nothing rather than a number a small sample cannot support. Methodology v1 applies no statistical weighting; when weighting or behavioral qualification is introduced, the methodology version changes and applies forward only.
Snapshots, versions, and receipts.
A nightly job writes aggregate snapshots per question, per segment, per day. Every snapshot carries the methodology version it was computed under and is rebuildable from the preserved vote-event record — history is append-only, and earlier days are never rewritten. That is what makes the longitudinal questions answerable: what a club’s respondents believed before a decision, and how the read moved after.
When reality resolves a question, the resolution seals a permanent public receipt: wording, options, evidence, results, and outcome, hash-sealed. Receipts keep the methodology version they were computed under forever.
Brigading posture, at a high level.
Voting is rate-limited, one-active-vote-per-account, and authenticated. Participation signals are captured at cast time so coordinated bursts are detectable in the record. We publish the posture and not the parameters: the point of this page is that the discipline exists and is enforced as code — not to hand a manipulator the manual.
What we deliberately do not claim.
Clubhouse results describe the people who answered: verified Posterior accounts who chose to vote. They are not representative population polling, and raw percentages are never marketed as such. They are not a census of a fanbase, not a probability, and not a prediction.
Movement between two questions is descriptive, not causal — a change in approve share after a move does not establish that the move caused it. Where a sample is too small to say anything honest, the Clubhouse says nothing. Empty is a legitimate answer.
The prediction model’s own methodology — signals, calibration, grading — is a separate discipline.
Signals, calibration, grading — published at /methodology.