Blind triangulation. One brew, everyone scores it independently, nobody sees anyone else's numbers until all of you have submitted.
Why blind matters. If you see a champion's score before logging yours, this measures anchoring, not perception — and the reliability figure it produces is worthless. Submit first, discuss after.
One batch, identical cups, same temperature. Vary the coffee style across sessions — calibrating only on light washed teaches the system exactly one thing.
Tap to toggle. You need at least two champions present — with one, there's no measurable spread and every disagreement gets charged to Joseph.
Everyone submits before anything is revealed.
Score what's in the cup, not what you want from it. Move every slider even if you're unsure — a blank biases the agreement rate more than a rough guess does.
Your own words, verbatim. If you taste chamomile, write chamomile — don't translate it into what you think the "right" answer is. The mismatch is the data.
Locked entries can't be edited — that's what keeps the session blind.
Append this object to the sessions array in Platform/data/palate-calibration.json. The store is append-only — corrections are new sessions, never edits.