Rendered Futures

Proteus is not just a prediction market — it’s a settlement layer for Rendered Futures in the Timepoint ecosystem. Every resolved market validates a prediction against reality, strengthening the Bayesian prior for future simulations.

What are Rendered Futures?

Rendered Futures

Definition: Simulated future timelines generated by AI systems, scored against reality when the predicted moment arrives.Purpose: Transform prediction from binary (yes/no) to high-fidelity text-level forecasting.

From README.md:

Proteus validates Rendered Futures against reality. Winners are the best renderers — their predictions are candidates for graduation to the Clockchain as validated causal paths. Every resolved market strengthens the Bayesian prior.

The Timepoint Suite

Proteus is one component in a broader temporal AI ecosystem:

System Overview
Data Flow
Component Descriptions

Service	Role	Type
Flash	Reality Writer	Open Source
Pro	Rendering Engine	Open Source
Clockchain	Temporal Causal Graph	Open Source
SNAG Bench	Quality Certifier	Open Source
Proteus	Settlement Layer	Open Source
TDF	Data Format	Open Source

Timepoint Thesis: A forthcoming paper formalizing the Rendered Past/Rendered Future framework, Causal Resolution mathematics, TDF specification, and Proof of Causal Convergence protocol.

Distance Metric Evolution

Proteus uses Levenshtein distance for exact text prediction, but the continuous-metric primitive generalizes to other prediction types:

V1: Levenshtein (Current)
V2: Semantic Distance (Planned)
V3: Graph Distance (Future)

Prediction Type: Exact text of a social media postMetric: Character-level edit distanceExample: Predict Satya Nadella’s next tweet about CopilotWinner: Closest character-by-character matchStatus: ✅ Implemented in PredictionMarketV2

The Continuous-Metric Primitive

All three versions share a core principle: closest match wins on a gradient, not a cliff.This structure generalizes beyond text prediction to any measurable outcome space with a proper distance metric.

How Proteus Validates Futures

1. Prediction Submission

Pro Generates Rendered Future

The Timepoint Pro rendering engine simulates what a public figure will say.Output: Predicted text in TDF format.

Prediction Submitted to Proteus

The prediction is staked on-chain via createSubmission().

createSubmission(
    marketId: 42,
    predictedText: "Copilot is now generating 45% of all new code..."
) payable

Market Awaits Resolution

Multiple participants (AI models, humans, insiders) submit competing predictions.

2. Reality Occurs

The Moment of Truth

The target actually posts on X. The oracle fetches the real text via X API.

3. On-Chain Resolution

Oracle Calls resolveMarket()

resolveMarket(
    marketId: 42,
    actualText: "Copilot is now generating 46% of all new code..."
)

Levenshtein Distance Computed

For each submission, calculate d_L(predictedText, actualText) on-chain.

Submission	Predicted	Distance
Claude	”…45% of all…“	1
GPT	”…43% of all…“	8
Human	”Microsoft AI is great…“	101

Winner Determined

winner = argmin(distance)Claude wins at d=1. The 7-edit gap over GPT is worth the entire pool.

4. Graduation to Clockchain

Validated Causal Path

The winning prediction is a validated Rendered Future. It becomes a data point in the Clockchain:

Input: Target’s past behavior, world state, timing
Output: Exact text (validated)
Quality: d_L = 1 (near-perfect)

This strengthens the Bayesian prior for future predictions about this target.

Training Data Accumulation

From the whitepaper Section 6.5:

Every resolved Proteus market produces a naturally labeled training example: the predicted text, the actual text, the Levenshtein distance, and the market context (target handle, time window, number of competitors). Across many markets, this accumulates into a structured dataset of (prediction, actual, distance, context) tuples — purpose-built for fine-tuning persona simulation models.

Why Proteus Data is Valuable

Real-World Labels

Labels are actual outcomes, not synthetic.Verified by oracle consensus and timestamped on-chain.

Continuous Quality Signal

Not just “correct/incorrect” but how close (character-by-character).

Adversarial Diversity

Participants actively search for strategies others miss.Distribution spans AI roleplay, insider knowledge, null bets, random noise.

Silence Prediction

Resolved __NULL__ markets label conditions when targets don’t post.Standard training corpora can’t provide this signal.

Fine-Tuning Applications

Persona Calibration
Numerical Precision
Silence Prediction

Train a model on resolved markets for a specific target.Example: All resolved @elonmusk marketsGradient: Predicted “confirmed for March” when actual was “is GO for March” → learn target’s preference for colloquial phrasing.

Resolved __NULL__ markets label when targets don’t post.Standard LM training cannot provide this signal (corpus only contains text that was written).

Training data scales with market volume: More markets → more targets → more resolved outcomes → more labeled tuples. Unlike static benchmarks that leak into pretraining, Proteus data is adversarially generated in real time — the test set is always the next unresolved market.

TDF Integration (Phase 2)

Future versions of Proteus will express predictions as TDF (Timepoint Data Format) records.

Current: Raw Strings

struct Submission {
    uint256 marketId;
    address submitter;
    string predictedText;  // Raw text
    uint256 amount;
    bool claimed;
}

Phase 2: TDF Records

{
  "@context": "https://timepoint.ai/tdf/v1",
  "type": "RenderedFuture",
  "target": {
    "handle": "@elonmusk",
    "platform": "X"
  },
  "prediction": {
    "text": "Starship flight 2 confirmed for March...",
    "confidence": 0.87,
    "model": "claude-4.5-sonnet",
    "timestamp": "2026-03-01T12:00:00Z"
  },
  "window": {
    "start": "2026-03-01T00:00:00Z",
    "end": "2026-04-01T00:00:00Z"
  },
  "submitter": "0x742d35Cc6634C0532925a3b844Bc9e7595f0bEb",
  "stake": "1.0 ETH"
}

Benefits:

Interoperability with Pro, Flash, Clockchain
Richer metadata (confidence, model provenance)
Off-chain storage (IPFS) with on-chain hash
Supports semantic distance metrics (V2)

Clockchain Integration

The Clockchain is a Temporal Causal Graph accumulating Rendered Past + Rendered Future.

Graph Structure

Proteus’s Role

Submit Prediction

Proteus market created for future event.

Reality Occurs

Event happens (e.g., CEO posts).

Resolve & Validate

Proteus determines winner based on Levenshtein distance.

Add to Clockchain

Validated prediction becomes a node in the causal graph.Metadata: Target, timestamp, actual text, winning distance

Strengthen Prior

Future simulations use Clockchain data for better predictions.

Proof of Causal Convergence

The Clockchain accumulates validated causal paths. As more Proteus markets resolve, the graph’s Causal Resolution (measured by SNAG Bench) increases.This is analogous to Proof of Work in Bitcoin — the accumulation of validated predictions proves the system’s forecasting capability.

Key Differences: Proteus vs Traditional Markets

Traditional Prediction Markets
Proteus (Rendered Futures)

Outcome: Binary (yes/no)Resolution: Simple thresholdTraining Data: Binary labels (correct/incorrect)Integration: Standalone productPurpose: Price discovery on specific questionsExample: “Will AI achieve AGI by 2030? Yes/No”

Complexity-Theoretic Framing

From the whitepaper Section 11:

Text prediction as capability proxy: Predicting exact text requires simultaneous integration of:

World model: What events are happening?

Person model: How does this individual express ideas?

Timing model: When will they post? What’s salient then?

Style model: Punctuation, formatting, rhetorical devices

Capability Metric

capability(M) ∝ E[1 / d_L(M(context), actual)]

A model that consistently achieves low edit distance across diverse targets and contexts demonstrates integrated competence across all four submodels.

Levenshtein distance to exact text is a naturally occurring capability benchmark — one that emerges from the market mechanism rather than being designed by a benchmark committee.

The Fast Takeoff Scenario

From whitepaper Section 11.3:

Distance Range	Market Phase	Strategic Implication
d_L ≈ 100+	Noise	Random guessing; market is a lottery
d_L ≈ 50	Signal emerges	AI outperforms random; theme-level accuracy
d_L ≈ 10	Precision game	AI captures structure, vocabulary, phrasing
d_L ≈ 5	High stakes	Small capability differences → large payouts
d_L ≈ 1	Frontier competition	7-edit gap = entire pool

Binary Markets Commoditize, Levenshtein Markets Deepen

As AI improves:Binary: P(correct) = 60% → 95% → 99% → spread vanishes, market diesLevenshtein: d = 100 → 10 → 5 → 1 → stakes per edit increase, market deepensThis is the thesis of Proteus and Rendered Futures: the approaching AI capability explosion doesn’t destroy the market, it makes it more valuable.

Key Takeaways

Settlement Layer

Proteus validates Rendered Futures against reality

Continuous Metric

Generalizes from Levenshtein to semantic to graph distance

Training Data

Adversarial, labeled, continuous quality signals

Clockchain Integration

Validated predictions graduate to causal graph

TDF Interoperability

Phase 2: Express predictions in standard format

Capability Proxy

Low distance = high integrated AI capability

Levenshtein Distance

Mathematical foundation of Proteus scoring

Market Lifecycle

How markets progress from creation to Clockchain graduation

Timepoint Ecosystem

Overview of Flash, Pro, Clockchain, and TDF

Whitepaper

Full academic paper on text prediction markets

Phase Status: Rendered Futures integration is currently conceptual. V1 (Levenshtein distance on text) is implemented and deployed on BASE Sepolia. V2 (semantic distance) and V3 (graph distance) are future research directions.The Timepoint Thesis paper formalizing this framework is forthcoming.

Get Started

Core Concepts

Using Proteus

Smart Contracts

Development

​What are Rendered Futures?