Status: pre-launch. This page describes target design; this document does not establish a deployed service.
Scope: forecast organ feeding the non-economic forecast capability merged
into Signal inside Skyzai Personal. Stakes route separately to Pay; orders use
the Markets compatibility handle and Pay's separately gated settlement path.
realityfutures.com is an adjacent venture, not a buyer package. This organ
evaluates COULD; it never executes. See The Product Family.
RealityFutures — Prediction Markets (COULD)
realityfutures.com | Organ 2 of 4 | Pipeline stage: COULD (predict) | Code-only P-score: 0.59
RealityFutures explores what COULD happen. Its forecasts and scenarios keep uncertainty visible; Agentz separately prepares unsigned proposals from them. The earlier market-only description is preserved below.
Current Evidence Note (2026-04-19)
- 18 Solidity contracts and a web app are present in code
- Forge test lane passes
- organism-local loops still use a statistical F2 fallback instead of live deployed market-backed emission
The organ capability
RealityFutures explores scenarios and forecasts with explicit sources, assumptions, uncertainty and limitations. The prediction-market mechanism below is retained research, not a required implementation or evidence of calibrated forecasts. An LMSR price is a model-derived market estimate; the softmax formula alone does not prove the probability of a real-world event. See the ForecastPacket contract.
Historical LMSR and forecasting proposal — not verified performance
RealityFutures is the organism's probability engine. It performs one function: price what COULD happen, using skin-in-the-game prediction markets. It does not observe (that is Circle). It does not judge (that is APU). It does not act (that is Skyzai). It prices.
ReFu is a prediction market platform where participants bet on future outcomes. The market price IS the probability. This is not metaphor — it is a mathematical property of the LMSR (Logarithmic Market Scoring Rule) that ReFu uses as its automated AMM.
LMSR: Price = Probability
The Logarithmic Market Scoring Rule (Hanson 2003) is the AMM engine:
Price(outcome_i) = exp(q_i / b) / sum(exp(q_j / b))
Where:
- q_i = quantity of shares for outcome i
- b = liquidity parameter (controls market depth)
Properties:
- Price = probability: The market price of a share that pays 1 if outcome occurs IS the market's aggregate probability estimate for that outcome.
- Bounded loss: The AMM's maximum loss is b x ln(n) where n is the number of outcomes. Known, finite, predictable.
- Skin in the game: Every probability estimate costs real tokens. Cheap talk is impossible. Lying is expensive.
Why Prediction Markets Beat Experts
- Experts have incentives to hedge. A consultant who says "70% chance" faces no penalty if wrong. A market participant who buys at 70% and the event doesn't happen loses real money.
- Markets aggregate private information. A trader who knows something the market doesn't can profit by moving the price. The price incorporates their private information without requiring them to disclose it.
- Markets self-correct. A wrong price is a profit opportunity. Arbitrageurs push the price toward truth.
AI Enhancement: TimesFM + Chronos
ReFu augments human prediction with two AI forecasting models:
- TimesFM (Google): Foundation model for time-series forecasting. Provides baseline probability estimates for markets with historical data.
- Chronos (Amazon): Pretrained probabilistic time-series model. Provides uncertainty bounds.
AI predictions seed the market. Human participants adjust the price based on private information the AI cannot access. The combination outperforms either alone.
Adjacent-venture economics — not package topology
The following possibilities belong to the adjacent realityfutures.com
venture. They do not create a Forecasts product, a third Skyzai package, or
authority to execute an order or settlement.
| Stream | Mechanism | Target |
|---|---|---|
| Conditional market-maker loss bound | Retained LMSR model: b × ln(n), not an adopted fee or spread | Market operator; funding and assumptions require separate definition |
| Market creation fees | Small fee to create a new prediction market | Analysts, researchers, organizations |
| API access | Probability feeds for downstream consumers (Circle, APU, external) | Developers, quant funds |
The retained LMSR model describes a conditional market-maker loss bound, not an adopted spread, funding guarantee or proof of non-extraction. Fees, available funds, collateral, resolution rules and operator incentives need separate owner decisions and evidence.
AXIOM Output
In the target design, when a ReFu market resolves under its rules, the resolution would be recorded as an AXIOM: an N:1 convergence that keeps every source visible. It is a money-weighted recorded outcome, not a settled truth; live market-backed AXIOM output is not yet public.
AXIOMs feed:
- Circle's intelligence graph (updating source reliability based on predictive accuracy)
- Agentz's unsigned proposal context; apu.bot may independently reason over admitted, permissioned inputs
- FlowWallet conditions (streams conditioned on AXIOM states)
Current Status
- code-only P-score: 0.59
- contract/test lane: verified locally
- live market-backed Kind 31338 emission: not yet public/runtime truth
- deployed web app / public market loop: not yet the claim this page should make
- statistical fallback in organism-local proof loops: yes
A forecast carries uncertainty. A settled bet records an outcome under its rules, not universal truth.