How Tangent Forecasts the Future

Approach

Mimic Zero is founded on Empirical Game-Theoretic Analysis (EGTA) and agentic simulation, coupled with AI judgemental forecasting. The system constructs a high-fidelity model of the real information environment (actors, relationships, behavioural patterns, and platform dynamics) then simulates it many times with controlled deviations to infer action and event probabilities from the distribution of outcomes.

Accuracy scales with fidelity. The closer the simulation mirrors reality, the more reliable the forecast. This is where our data layer matters: rich historical profiles, network structure, and behavioural baselines ground every simulation in observed reality. Our models and the overall Mimic Zero system ensure that agents act in line with the real counterparts they are imitating, producing emergent dynamics that reflect how events would actually unfold.

The network and agent construction for each simulation is learnt from similar past events. As more events are observed, the system builds a better picture of who appears, how they react, and what roles they play. Behaviours of similar actors also inform agent modelling, so even actors with limited history benefit from patterns observed across comparable profiles. History does not repeat, but it rhymes, and each new event improves the fidelity of future simulations.

Data Foundation and Historical Grounding

Mimic Zero ingests data continuously from X, Reddit, news sites, YouTube, price feeds, Polymarket, and other sources. ML pipelines run sentiment analysis, narrative tracking, entity tagging, and entity-relation extraction across this stream to build enriched author profiles, influence scores, behavioural baselines, and network position maps. This gives the system a structured view of who holds what opinions, how topics are perceived by different cohorts, and how actors relate to each other.

A graph database layer stores the structured relationships extracted from every post, profile, and external source (e.g. "Elon Musk is CEO of Tesla", "Fund X holds exposure to Protocol Y"). When a new event is detected, this graph lets the system immediately identify related actors, affected parties, and relevant context to populate the simulation accurately.

Together, this means actor discovery, role inference, and behavioural conditioning are grounded in observed data. Every simulation inherits a historical foundation that deepens over time as more events and interactions are processed.

Environment Construction

Before simulation, Mimic Zero constructs an environment bundle: a compact representation of the event, the bounded early feed, resolved actors, contextually relevant entities, relationship hints, and the artifacts needed downstream. This bundle defines what happened, who matters, and what is visible at the chosen evidence cutoff.

Entity discovery is intentionally conservative. The system builds from anchor-context evidence, supplements with pre-computed pre-event relations, and strictly avoids leaking later consequences into the build — ensuring a disciplined, low-leakage construction of the event world from historically valid evidence.

Graph Expansion and Archetypal Crowd Actors

Named actors alone do not capture the full social system. Many events are shaped by communities, amplification clusters, and counter-publics that span hundreds of accounts. Mimic Zero builds a graph layer that infers entity relations and uses clustered historical network structure to synthesize archetypal actors from real community patterns.

These archetypes represent recurring behavioural blocs found in the data — trader cohorts, analyst clusters, media amplifiers, tribal defenders, skeptical observers. This lets the simulation represent crowd behaviour as an endogenous force that emerges from the network structure.

Layered Persona Construction

Every actor carries a layered persona. A stable base layer captures identity, posting style, influence, and network position from author metadata and pre-event history. An event-specific overlay adds how the actor relates to the current incident — their exposure, obligations, and strategically available responses.

The same layering applies to archetype actors: a cluster-level behavioural profile conditioned on the current event. Each persona gives the simulation structured priors about who an actor is, how they tend to behave, what they currently perceive, and what actions are credible in context.

Game-Theoretic and Reactionary Behaviour

Not all actors are playing the same game. Some participants react impulsively — retweeting, piling on, drifting with sentiment. Others reflect on nearby context before responding. Others still are fully strategic: comparing options, weighing reputational constraints, and sometimes determining that silence is the best move. Mimic Zero places each actor on a spectrum from reactive to strategic, grounded in their past behaviour, perceived deliberateness, and how critical the situation is to them.

Strategic actors produce structured assessments — comparing candidate actions, attaching probability estimates, and retaining a reasoning trace across rounds. Critically, they act with bounded rationality as per the fundamentals of Empirical Game-Theoretic Analysis (EGTA) rather than attempting to solve the games they are in. They deliberate under uncertainty with incomplete information, which is what produces realistic emergent dynamics rather than implausible optimal play.

Hidden-State Forecasting to Model Information Asymmetries

Before any simulation runs, Mimic Zero identifies hidden variables that would materially change how an event plays out — questions like whether an exploit was insider-driven, whether a regulator will intervene, or whether a protocol can contain the issue internally. A forecasting layer estimates probabilities for each outcome.

These probabilities define distinct worlds. In each world, certain actors receive asymmetric private information reflecting that branch's assumptions — one world where insiders know the truth, another where they don't. By running all branches, Mimic Zero reveals which actions and reactions hold steady across different versions of reality and which are contingent on information that may or may not be public. The result is a forecast that accounts for what people know, not just what they do.

Multi-World Execution and Aggregation

Mimic Zero runs many versions (worlds) of each simulation, each with controlled variation — perturbed activity rates, sentiment biases, interaction order — to test whether predicted outcomes hold under different conditions or only emerge in narrow scenarios.

Results are aggregated with probability-weighted averaging. A branch that started with a 40% prior contributes more to the final forecast than one at 25%. This produces a single view of which developments are robust, which are contested, and which are edge cases.

Signal Propagation

Mimic Zero also treats information diffusion itself as part of the object of study. Not every fact enters the environment in the same way. Some signals are globally visible because they are issued by authoritative actors or because they have already crossed a public threshold. Other signals begin as local observations seen only by a subset of the network. Those signals may fizzle, remain niche, or propagate outward through repeated engagement and amplification.

This distinction matters because many incidents are defined not only by what is true, but by when different segments of the network come to believe that something is true. By modeling both global visibility and networked exposure, Mimic Zero can represent rumor cycles, asymmetric awareness, delayed crescendos, and information cascades as endogenous dynamics.