Outcome Tracking

Outcome Tracking

Predictions are graded against reality, at fixed horizons, with no human intervention.

The horizons

Horizon	Typically
30 days	short-term calls
90 days	quarter-scale judgements
180 days	strategic positioning

A durable Outcome Clock (scheduler.py) fires at each horizon and grades whatever has come due. It survives restarts — a prediction made today is still graded in six months, whether or not anyone remembers it was made.

Every prediction carries an invalidation condition

Governance Engine rejects a forecast that does not state what would prove it wrong. This is not bureaucracy: an unfalsifiable prediction can never be graded, so it can never be learned from. Requiring the condition up front is what makes the whole loop possible.

What a grade does

Result	Consequence
Confirmed	the belief is reinforced; the member's Reputation rises
Refuted	the belief's confidence is lowered, attributed to reality
Minority was right	the dissenting member's reputation rises — and the pattern is recorded
Unverifiable	recorded as unknown; not guessed, and not counted as either

The last row matters. A judge that guesses when it cannot observe manufactures a record that looks like learning and is not.

The loop closes here

Deliberation → Verdict → Prediction → (30/90/180d) → Grade
     ↑                                                  │
     └────────── House Mind + Reputation ←──────────────┘

Most agent systems stop at Verdict. Everything to the right of it is what makes the House an institution rather than a very articulate session.

Related

The same discipline is published as a standalone standard — Genesis Reality Grading — whose governing law is that a provider's assertion is not evidence: a system's own claim of success carries no weight, and only observable outcomes count.

→ Next: Tools and Skills