The prediction ledger

You set a trust weight on every manager you follow. The prediction ledger is where that number stops being a feeling and starts being a record.

It is a ledger of what people actually said — verbatim — and how it turned out. It covers the managers you track, and you.

Why bother

Everyone remembers their good calls; almost nobody keeps score of the rest. The people you follow certainly aren't keeping score for you, and the vivid recent call is exactly the one your memory over-weights. If you're going to delegate any part of your allocation to someone else's judgment, you want their record, not your impression of it.

Logging a claim

Open Managers → Predictions and log a prediction. Four fields carry the weight:

• The claim, verbatim. In the speaker's words, as said. Not your summary of it.
• The falsifiable condition. What would have to be true for this to count as right — written by you, separately, and set before the outcome is known.
• The claim type (below).
• A horizon, if there is one. Plenty of claims don't have a date; those stay open until you rule on them.

Keeping the claim and the condition apart is the design decision that makes the whole thing trustworthy. Nobody — not the manager, not you, not an AI — gets to soften a claim after the fact by rewriting what was "really" meant. The words stay put; the standard is explicit and set in advance.

Only published claims belong here. If an AI assistant recalls a manager saying something, that's a lead to go verify, not a record to file. The ledger is a record of what was said, not of what was remembered.

Four claim types, scored separately

They are never pooled into one number. Someone can be excellent at judging a business and hopeless at calling the market — and you're usually only borrowing one of those skills.

Resolving

Claims with a horizon come due and land in a resolution queue rather than quietly expiring. Claims without one stay open and visible, still worth ruling on whenever the answer becomes obvious.

Nothing resolves itself. The system never decides someone was right; it just makes sure the question comes back to you. When you rule, you record the outcome and, if you want, a note on what actually happened.

If a manager later revises a claim, the new one supersedes the old, and both stay in the record.

Scorecards that refuse to flatter

Per manager and per claim type, the ledger shows the record. With one deliberate rule: below eight graded claims it shows counts, not percentages. "3 of 4" is honest; "75%" from four data points is a number pretending to be evidence, and it would be the most quoted number in the product.

Small samples aren't hidden — they're shown as what they are, and they grow into percentages once there's enough to say.

Zeitgeist

The Zeitgeist view groups open predictions by theme across everyone you follow, so you can see where the people you trust agree and where they diverge — three managers all betting on the same cycle is a concentration risk you won't spot one profile at a time.

Themes come from the tags you assign. It reports what you told it rather than inferring clusters, which means it can't quietly invent a consensus.

Doing this by talking

Logging claims is transcription work, and your connected AI assistant can do it: hand it a transcript or a letter, have it pull out the falsifiable claims verbatim, review them, and let it file the batch. It can also work the resolution queue with you — but you rule on every outcome.