Most of what people do, they have done before. We compute it against published outcomes. Our first market is film and TV: the streaming views behind over 128,000 titles, in 20 markets and 34 platforms.
U = f(En, El, Pb, P | Π)
We compute the streaming views behind film and TV, anchored to what platforms themselves publish: one number both sides of the table can check.
TalentFor actors, directors, writers, producers and composers
See the computed streaming views behind every title you worked on.
EnterpriseFor studios, streamers and enterprise clients
Computed views for every title in your slate and library, period by period, checked against published outcomes. See how audiences met every comparable title before you commit.
Evolution and experience keep the behaviours that protect us, help us succeed or bring us pleasure. So we repeat them: in the same places, at the same hours, in the same rhythms.
At scale, on the same clock, it repeats.
Vast isn’t infinite. Chess is bounded too.
We built the record, period by period.
They had no clock. Anchored in time, behaviour can be read.
Behaviour is vast, but bounded. It is carried forward by its own past, and what worked is what we kept.
86 billion neurons and trillions of synapses make the combinations enormous. Enormous is not infinite: every sense has a limit, so the number of scenes we can tell apart is finite.
10120
possible games of chess: vast, yet bounded. Behaviour is like chess.
of everyday actions are habits, done almost daily and usually in the same place.
Wood, Quinn & Kashy, 2002of a person’s next move can be read from their past moves, across 50,000 mobile phone users.
Song et al., Science, 2010countries show a similar daily rhythm of mood, across 509 million messages.
Golder & Macy, Science, 2011of coin tosses land on the side they started on, across 350,757 flips by 48 people. Even a coin toss carries its past.
Bartoš et al., University of Amsterdam, 2023
The loops that survive are not random. They follow a rule, and a rule can be computed.
In 1936 Kurt Lewin showed that behaviour is a function of the person and their environment. We made each part of that function readable in a scene, then added the element that makes behaviour computable: time.
=f(,,,|)
Choose a term to see what it reads.
What a participant takes from a scene: the reading the equation computes.
The setting, scenery, location, time and everything else that frames the scene.
The identifiable things in it: furniture, clothing, tools, currency, communications and more.
What came before: a creator’s history and previous works, the storyline, the people who made it.
The people in the scene, real or acted, with their history, personality and motivations.
Who is reading the scene. Established first, because the same scene means different things to different people.
U(t) = Σ wᵢ(t) · Xᵢ(t) + ε(t)
At each moment, understanding is a weighted sum of four scores, each between 0 and 1. The weights add up to one, and ε is what no input explains yet. Each period is read on its own, against what was true at that time.
The weights start open and tighten with every observation, period by period. The more that is read, the less is left to chance.
An illustration of the method, not a measurement: each weight’s range narrows as periods are read, and the unexplained part shrinks.
The compute to read a bounded space at scale is here. We built the system that resolves behaviour against what actually happened, period by period.
Published, verifiable outcomes from 202 sources: the ground truth.
Abstract data for the same period, and where and when each title could be seen, country by country.
Bayes’ rule updates the odds of each outcome as each period’s data arrives. The weights tighten with every test.
P(A = x | B) =P(B | A = x) · P(A = x)P(B)
A = the published outcome, x = one value of it (such as a view count), B = that period’s abstract data.
The residual is the part no input explains yet. Each new input shrinks it.
Not guessed.
Computed.
Derived from known data, held, and updated whenever its inputs change, like a formula in a spreadsheet. Pattern recognition from evidence, anchored in time.
The foundation is the record: over 128,000 titles, 45 months and 202 sources of ground truth. Five layers sit on top of it.
An API suite with custom and industry-standard connectors.
Encrypted, and held offline in physical vault storage.
Research agents that gather the latest practice in each field, in many languages.
Specialist agent teams for each domain, with access to Wyatt, tested before they are used.
Agentic reasoning, inference-time compute and multi-agent collaboration, using temporal periods to reason about the behaviour patterns that succeed.
Deep Blue never counted every game; it searched what mattered. So do we. Without time, behaviour is only pattern recognition. Anchoring it to when it happens makes it computable.
What looks like noise is often inputs not yet read. We test that two ways: with people, and at scale.
Computed, not platform-reported: each period is checked against published ground truth from 202 sources. This is the engine behind Oh Talent.
In 300+ blind table-setting sessions, 92% of readings fell within two bands, against 44% by chance (95% confidence interval 88.4–94.6%; κ = 0.86, agreement beyond chance).
Take the Table TestThe space of scenes is finite (Proposition 4.1), and adding inputs never raises the residual (Equation 30). The floor R* is set by noise, not by the model, and what remains is measured per person, per period.
As six inputs are added, the residual falls from 0.00436 to 0.00010. 168 computational checks, none failed.
Every test updates the equation, including the ones it fails. That is the VITAL principle: validity is not a certificate issued once. It is the state of the weights after every test so far, and it grows with each one.
Film and TV came first: enough outcomes are published there to check the model, period by period. The same stack now serves music, and the research reaches further.
Toy Soldiers is a creator-owned, direct-to-fan music platform. Oh technology gives each artist a computed reading of their fans.
Every line of business reads from the same Oh technology stack.
Plan from computed behaviour, checked every period.
Take psychology from description to measurement.
Research into how the scenes around people in care affect them.
We read audiences. We never steer them against their interest.
Privately held, and based in Los Angeles.
Investors and press: write to us for materials.
