7 · Aggregating Truth

🚧 Draft (core-idea sketch). Part A · Core Ideas. One organ — perception — not the root.

The problem. Combine the dispersed knowledge of many specialists into a shared, evolving world-model — without a single oracle that can be bribed, and without a point-in-time resolution that wealthy actors can push around. (This is the problem behind funding science, grants, and policy.)

The core idea. The retroactive consensus market, which decouples eliciting a forecast from declaring what happened from paying out:

• Forecasters publish timestamped probability distributions.
• Capital-holding resolvers privately and retroactively declare what they believe happened — optional, revisable, discardable.
• A reference-relative proper scoring rule pays each forecaster for moving belief toward that resolver's eventual verdict. The money-optimal forecast becomes the capital-weighted prediction of future resolver consensus.

Output: a live shared world-model and a skill ranking of forecasters — with no exploitable point-in-time oracle. It is the system's perception; it is the most worked-out instance of the recurring "pay for prediction" principle, but it sits on top of identity, timestamps, and money — it is not what they're built on.

Leans on: identity (1), timestamps (2), money (6), optionally the substrate (3) for machine-resolvable questions. Enables: governance (8) — the fact-finding half.

⚠️ Where it's thin. Two central, unproven theory gaps: reflexivity (if resolvers defer to the forecast, does the loop converge to truth or to a self-fulfilling fiction?) and causal validity (futarchy's conditional-vs-causal flaw). Both are reduced to measurable conditions in the Reference, but neither is closed. Part B.