Whitepaper

Updated 10/06/26, 16:26

The binary is the problem

Retail market tools sell one number and a direction. Up or down, win or lose. That frame is not a simplification. It is the failure mode. A trader anchors to a single price, sizes risk against it, and learns the true width of the distribution only when it moves through him.

Desks do not think in points. They think in distributions: where the mass sits, how fat the tails are, what the present does to the spread. That machinery has never existed for the public in an honest, continuously maintained form. The closest public tools extrapolate past price and call the resulting cone a forecast.

Heatmup builds the distribution and gives it away. For any covered asset it estimates a full probability surface over future price and time, renders it as a heatmap a non-specialist reads in one glance, and resolves it later in public. P50 is the median. Ninety percent of modeled density falls between P5 and P95. Not a target. Not advice. The shape of the outcome space a holder already occupies, whether or not he can see it.

Per-cell estimation, not a widened line

A time-series extrapolator reads one input: price history. It projects a median and inflates a band as variance compounds. The cone widens because volatility widens, not because the model understands today. It cannot. It has never seen the news, the macro print, or the last twenty-four hours.

Heatmup inverts that. It does not draw a curve and stretch it. It estimates every cell on the grid, each price level at each future timestamp, as a discrete prediction. A single forecast can resolve from up to a million independent predictions, each conditioned on current information and produced by a different reasoning or simulation method, then aggregated into the surface a user sees.

The geometry confirms the underlying structure. A widened line is one statistical object, smooth by construction. The Heatmup surface carries structure, density, and regime shape: tight through the core where the lenses converge, wide in the tails where they genuinely disagree. That disagreement is not noise to be suppressed. It is the signal. When the options lens reads bearish and the news lens reads bullish, the spread between them is the most honest object on the page, and the heatmap keeps it rather than averaging it into a false consensus.

HMX: aggregation, not a chatbot with a finance prompt

HMX is the engine. It is a quantitative aggregation system, not a conversational model answering market questions in prose. Rule-based computation executed at a rate and scale no analyst matches, combined by mathematics rather than by a desk head's conviction.

It reads each asset through several independent lenses in parallel. Each lens computes the same grid from a different standpoint and commits its own view. The engine then synthesizes those views, applies time-decay so recent estimates outweigh stale ones, and collapses the result into one coherent distribution. No human selects which lens to believe. The combination is governed by method, not opinion.

Breadth beats brilliance here, and the reason is statistical rather than rhetorical. Independent estimators making uncorrelated errors see those errors cancel under aggregation while the shared signal survives. One excellent model is wrong in a single direction. A large set of ordinary ones, wrong in different directions, averages toward truth. Running many perspectives is the method, not a budget compromise. The cost engineering that makes this viable at high frequency is the part Heatmup treats as proprietary, and it is why HMX behaves like infrastructure instead of a demo.

Refreshed in hours, projected across years

The full pipeline re-runs unattended on a tight cycle. The surface a user opens reflects conditions hours old, not weeks. As the present moves, the distribution moves with it. This is a live forecasting layer, not a static report that ages on a page.

Horizon is wide. The heatmap expresses the model's view as far as a decade out, so a reader sees near-term shape and long-horizon fan on the same surface, P5 through P95 widening the way real uncertainty widens.

There is, today, effectively no public, free, regularly updated source of probabilistic price forecasting that prices in the present. Extrapolators ignore it. The funds that produce genuine distributions never publish them. Heatmup sits in that empty quadrant.

Resolution is the product

A forecast that is never graded is content. Heatmup is the inverse. Every forecast receives an immutable ID and a timestamp, is archived, and is later resolved against the realized price. The raw forecast history is available for independent validation. Hits and misses publish with equal prominence. The record, not the picture, is the asset.

The discipline is enforced by the surface, not appended to it. Show the band, never the lone line. Grade on calibration across many forecasts, not on being right once. Publish the accuracy record and the methodology openly and link them from every forecast. A polished forecast reads as authority, and authority on something a person acts on is precisely what can harm him. The only honest response is to put the track record in public and let it argue for itself.

That record is also the defensibility, and it is the part a competitor cannot lift. A presentation layer is copyable overnight. The history behind it is not. A graded public record accrues only in real time, one resolved cycle at a time, and it cannot be backfilled, because computation cannot be timestamped into the past. The moat is not the interface. It is the ledger of resolved forecasts, and a clone inherits a track record of zero.

Honest by structure

Prediction markets are the nearest public cousin, and their limitation is structural, not fixable. In a prediction market the price is the forecast, so trading it moves the thing being forecast, and the venue must clear positions to exist. Heatmup holds no position and clears nothing. It describes the distribution without perturbing it. A venue that has to trade cannot make that claim.

The best forecasters alive sit inside quantitative funds, and they will never publish. An edge dies the moment it is public, and they sit on the other side of the trade. The entities most able to compete in the open are the ones structurally barred from doing so. The field is empty by design, and Heatmup occupies it transparently rather than extractively.

Reflexivity-free description, public resolution, and open access are not features a competing product bolts on later. They fall out of taking no position and grading everything in public from the first cycle. A forecast that resolves in the open, win or lose, is evidence.

Free, and built to compound

Every forecast is free to view. No paywall, no signup to see the heatmap. This is strategy, not generosity. The compounding asset is an uninterrupted record of resolved forecasts graded in public, and the fastest way to build it is maximum distribution, which means giving the forecasts away. Future depth may sit behind modest subscriptions. The core probability layer stays open.

Markets are the right first domain: they resolve cleanly, fast, and unarguably, and the audience is the largest and most active online. The architecture is domain-agnostic. Any high-stakes domain that resolves cleanly maps the same way, from sovereign and supply-chain risk to macroeconomic indicators and grid stability. The objective is to make calibrated, publicly graded probability a utility rather than a privilege of the desks that can fund a research team.

Heatmup Oy operates as a provider of quantitative market analysis and does not provide directive financial advice. Our operations are governed by the transparency requirements set forth in EU Market Abuse Regulation (MAR) and MiFID II. We maintain a strict separation between data synthesis and investment recommendations. The integrity of our forecasting engine is subject to public verification via our continuously updated track record.