Label every item proven, better, or new with Mark Pincus's test and nobody else's: proven is what already works for your audience on your platform, copied click by click; better is what 10 out of 10 users of the proven product would choose; if only you think so, it is new; and the launch isolates exactly one new. Run on Heidi, the company brain Falkster.AI builds, the board reads: Proven, the features, which a funded competitor can replicate because models are rented. Better, empty, for Pincus's reason. One removal, pooling learning across tenants, logged as a bet rather than a better, because removing something proven is a bet like any other. One new that failed, the Signals layer, which held everything, displayed it cleanly, and compounded nothing. One new that is the company, the tenant's earned record of what agents did, which acts a human reversed, and which rules survived contact with outcomes. Where I read Pincus differently is his rule to pursue less new early in a platform shift: for a product on rented models, less new rounds down to a rented model and a wiki import, a snapshot that ports, so the discipline is exactly one new, not less.
Label every item with Pincus's test, write empty under Better when that is the honest answer, and isolate exactly one new. On Heidi the one new is the earned record, and the Signals layer is the new that failed.
The full board is in Mark Pincus Left the Better Column Empty. Here Is Heidi's Board. The four moat questions it sits on are in The Data Flywheel Has a Reverse Gear, and the written-versus-earned line is in Context Is King. Written Context Is Rented.
Last reviewed 2026-10-27 · 1 min read