teleo-codex/inbox/queue/2026-04-24-overcomingbias-hanson-decision-selection-bias-futarchy-fix.md

---
type: source
title: "Robin Hanson Partially Engages Rasmont — Proposes Four Fixes for Decision Selection Bias in Futarchy"
author: "Robin Hanson (@robinhanson)"
url: https://www.overcomingbias.com/p/decision-selection-bias
date: 2026-04-24
domain: internet-finance
secondary_domains: []
format: article
status: unprocessed
priority: high
tags: [futarchy, decision-markets, selection-bias, mechanism-design, hanson, rasmont]
---

## Content

Robin Hanson published "Decision Selection Bias" and "Futarchy's Minor Flaw" posts acknowledging the decision selection bias problem in futarchy and proposing design modifications to address it.

**The acknowledged problem:** The price→info→decision sequence creates selection bias in conditional market prices. "Futarchy can indeed give biased decision-conditional prices if one allows decision selection bias sequences of price then info then decision."

**Proposed fixes:**
1. **Randomized acceptance:** Randomly reject 5% of proposals that the system would otherwise accept. This ensures observations of the counterfactual state (non-adoption), allowing traders to price conditionally on non-adoption accurately.
2. **Insider trading access:** "Persons with access to decision-maker information should be permitted to trade in these markets."
3. **Timing announcements:** "The timing of the key decisions should be clearly announced just before such decisions are made so that speculators trading then need not fear the decision will be based on future information."
4. **Sequential per-timestep decision markets:** Create decisions at each time-step with three options (choice A, choice B, or wait), avoiding selection bias throughout.

**What Hanson does NOT engage:** Rasmont's structural payout-structure objection — that even with rational causally-reasoning traders and perfect information, conditional market prices track welfare-conditional-on-adoption rather than welfare-caused-by-adoption. Hanson's fixes address information-timing problems; Rasmont's critique is that the problem is intrinsic to the conditional payout mechanism itself.

## Agent Notes

**Why this matters:** This is the first substantive engagement with the Rasmont critique after 3+ months of silence. Hanson is the intellectual author of futarchy; his acknowledgment that there IS a "minor flaw" is significant. But the rebuttal is partial — it addresses the timing/information version of the bias, not the structural payout version.

**What surprised me:** Hanson's proposed "randomize 5% of acceptances" fix is operationally strange for high-stakes governance decisions. A futarchy system that randomly rejects 5% of its own approved decisions would face serious legitimacy problems — participants would not accept that their approved proposals get randomly overruled. This fix may work in theory but creates governance legitimacy costs that Hanson doesn't address.

**What I expected but didn't find:** A direct engagement with Rasmont's specific claim that "welfare conditional on adoption ≠ welfare caused by adoption" is intrinsic to the payout structure. Hanson treats this as an information problem (traders don't have enough data to distinguish correlation from causation); Rasmont treats it as a structural problem (the market mechanism itself selects for correlation-exploiters). This gap is the live intellectual tension.

**KB connections:**
- [[futarchy is manipulation-resistant because attack attempts create profitable opportunities for arbitrageurs]] — Hanson's response implicitly strengthens this: the selection bias problem is about information, not manipulation resistance. Arbitrage still works.
- [[Futarchy solves trustless joint ownership not just better decision-making]] — Belief #3 is the affected belief. Hanson's partial fix doesn't resolve the deeper Rasmont objection. Belief #3's confidence should be noted as "weaker on causal decision quality, stronger on execution/downside protection."
- [[MetaDAOs futarchy implementation shows limited trading volume in uncontested decisions]] — MetaDAO's binary pass/fail mechanism may implicitly limit selection bias by reducing the option space (no "wait" option in binary decisions). Hanson's sequential model uses 3 options. This is a design difference worth investigating.

**Extraction hints:**
- Claim: "Hanson's decision selection bias fixes address information-timing problems but not the structural payout gap between conditional and causal welfare estimates — the Rasmont critique partially survives the rebuttal"
- Claim: "Futarchy's 5% random rejection fix creates governance legitimacy costs that make it inapplicable to high-stakes single decisions — the fix works for low-stakes iterated decisions only"
- Note for extractor: This should be linked to the existing Rasmont archive and treated as a partial rebuttal. The divergence between Rasmont and Hanson is worth flagging as a divergence-candidate: two competing views on whether decision selection bias is structurally intrinsic or operationally correctable.

**Context:** Hanson is the original designer of futarchy (2000 paper). His engagement after 3+ months marks the first substantive response from the intellectual lineage of futarchy to Rasmont's critique. The LessWrong post still has zero comments as of last check. Hanson's posts reach a different audience (Overcoming Bias readership, rationalist-adjacent, less crypto-native).

## Curator Notes

PRIMARY CONNECTION: [[futarchy is manipulation-resistant because attack attempts create profitable opportunities for arbitrageurs]] — the Hanson engagement reframes the futarchy robustness question from manipulation resistance to epistemic accuracy of conditional prices

WHY ARCHIVED: First substantive response from futarchy's intellectual author to Rasmont's 3-month-old "parasitic" critique; changes the status of the KB's most serious unresolved challenge

EXTRACTION HINT: Focus on (a) distinguishing what Hanson actually fixes (information-timing bias) vs. what Rasmont says is unfixed (structural payout bias); (b) the legitimacy cost of random rejection; (c) whether MetaDAO's binary mechanism implicitly mitigates the bias Hanson's sequential model addresses