consistency-ft × futarchy-fi/simulations

Consistency ↔ Futarchy

Two research lines that turn out to be one: futarchy-fi's simulations repo ran LLM agents inside real decision-market mechanisms and measured where they fail; consistency-ft trains exactly that faculty. This page is self-contained — it assumes you haven't seen the futarchy experiments.

The experiment, in one screen

The futarchy-fi/simulations repo (private — ask Kelvin for access) contains a decision-market engine ("Proposal Poker"): each of 150 proposals has a hidden quality x ~ N(0,1); five agents each get one private noisy signal sj = x + noise (precision proportional to the agent's wealth), and stake real (simulated) money on APPROVE/REJECT across 3 sequential rounds of a binary staking market, with participation costs, a 1% fee, a winner subsidy, and a costly verification oracle. Every arm sees identical draws (common random numbers), so arms differ only in the decision process:

ArmDecision processWhy it exists
Acoded rational-Bayesian agents + marketmechanism works with ideal agents
BLLM agents (Claude Haiku 4.5) + marketthe arm under study
Cpoll: average the same LLMs' pre-market beliefsdoes the market add anything over just asking?
Done LLM manager sees all five signalscentralized baseline
F / Gone agent covertly bribed / bribed and instructed to lie coherentlymanipulation and audit-evasion

Separately, the repo reproduces Galanis (2026) — LLMs trading in a prediction market over structured information — as an exact game solved with CFR+: the equilibrium aggregates well on every information structure, so when LLMs degrade on the hard structures, that's an agent limitation, not a market one.

The shared finding. The LLM traders are statically well-calibrated (Brier 0.076; stated 0.6 → realized 0.627) and stake in proportion to how informed they are (Spearman +0.86) — the two virtues markets rely on. But they do no within-market belief updating at all (precise claim below) — so the market's price aggregates worse than an unweighted poll of the same models (r = 0.882 vs 0.925 against truth), because stake-weighting adds noise while the beliefs never improve. The bottleneck is agent belief dynamics, not market design. Calibration-without-coherence is the exact failure consistency-ft exists to train away.

"Zero within-market updating" — the precise claim

It sounds unbelievable, so here is exactly what was measured. In round 0, each agent's prompt contains only its private signal — no market state (these answers double as the poll, Arm C). In rounds 1–2 the prompt additionally shows the public history: the running APPROVE/REJECT stake totals from the other agents' earlier moves. A Bayesian with an independent noisy signal should treat those totals as evidence — four other signals are worth roughly 4× their own — and move toward the market. What Haiku does instead, over 2,250 logged calls (raw JSONL in the repo):

Why this matters for consistency-ft: "update your stated probability coherently when shown new evidence" is precisely what the conditional consistency checks (Cond, CondCond, ExpEvidence in Paleka et al.) measure, and what exploitability training (RLE, our RLEF variant) directly optimizes. An agent whose beliefs are Dutch-book-coherent including its conditionals cannot both call the book "evidence" and leave its belief unmoved.

shared bottleneck: calibrated but incoherent beliefs consistency-ft train coherent probabilistic beliefs • 10 Dutch-book check types • arbitrage violation metric • RLEF: −Exploit − λ·anchor • KL-to-base-beliefs arm • coherent-projection math • 17-model consistency bench futarchy simulations measure mechanisms with real agents • proposal-poker venue + LLM traders • CFR+ rational benchmarks • manipulator / liar / bribe arms • BATCH-LMSR, Kyle, TWAP studies • ΣPnL = b[ln2 − LogLoss] identity • evolution harness (designed) tuned checkpoint as trader — does the market finally beat the poll? coherence audit → catches the "aligned liar" without waiting for outcomes also: coherent quoting for belief-net makers · soft inversion vs jamming adaptive adversaries (evolution harness) = the anti-Goodhart fix for our checks venue P&L as a resolution-free eval of every checkpoint also: the saturation lesson — measure headroom before training
Blue arrows: what consistency-ft ships into the futarchy stack. Violet arrows: what the futarchy research sends back. The capsule is the empirical finding both sides independently reached.
consistency-ft → futarchy

What we ship them

  1. The missing market agent. Their traders never update; our conditional checks train updating. Drop a tuned checkpoint into proposal-poker and A/B market-vs-poll on their CRN shards.
  2. A liar detector that works. Their Arm G "aligned liar" beat the stake-vs-belief audit 0/118 and profited; only slow outcome-based Brier caught it. A Dutch-book audit over related propositions is instantaneous — fabricated beliefs must stay coherent or pay a visible penalty.
  3. Coherent quoting. bayes-market enforces cross-market coherence in the engine; a coherence-tuned model carries it in the head — an incoherent maker is a subsidy leak equal to our violation metric.
  4. Soft inversion. Their batch mechanism is jammable by one inconsistent order under strict Bayesian inversion; projection-onto-nearest-coherent is the fix, and we build that machinery anyway.
futarchy → consistency-ft

What they ship us

  1. Adversaries that adapt. Their best-response manipulators showed static analyses understate exploitability 3–5×; their evolution design is exactly the "adversarial check generation" Paleka names as the open anti-Goodhart fix.
  2. An eval with economic units. ΣPnL = b[ln2 − LogLoss] means "can an adversary profit off your checkpoint in a venue" is a proper-scoring-rule eval needing almost no resolutions.
  3. The saturation lesson. Their v0 environment was too easy to tell arms apart; same trap as training on checks the base model already passes (Consequence 0.03 vs Negation 0.39). Measure headroom first.

Ranked joint experiments

E1 · DAYS — Coherence audit vs Arm G liar, in their existing harness. Cheap, novel for both repos, no training required.
E2 · AFTER PILOT — RLEF-tuned checkpoint as market agent: does the market beat the poll? Our generalization result and their aggregation result, one experiment.
E3 · CHEAP — Galanis Hard-structure eval of tuned checkpoints; their CFR+ and LLM-empirics baselines already exist.
E4 · WHEN NEEDED — Evolution-based adversarial check generation, once the static LP adversary saturates.

Full text: FUTARCHY-BRIDGE.md · model dashboard · RLEF × Forecasting proposal · sources: futarchy-fi/simulations (RESULTS.md, BATCH.md, MANIPULATION.md, KYLE.md, EVOLUTION.md, galanis-market).