Terrafa.
Talk to us Book a walkthrough

Terrafa AI. The operating system for commodity operations.

Continuum is a unified, contextualised foundation for the whole of a commodity operation: one industrial knowledge graph holding the physical, commercial and fiscal sides of the business as a single operational model, so the operation owns the definition of its own business rather than assembling it across five systems. Every function works from that model, and every decision, buy, hold, move, blend, sell, fund, cover, commit, is taken against the enterprise rather than one silo.

Beneath the foundation sits the layer the market does not have. The physical half is generated at the source, continuous and in place, and every value carries its computed uncertainty through to custody transfer and settlement. Around the model, the operation's counterparties, its capital and the systems acting on its behalf work from the same record.

Book a walkthrough Why it is built this way
/The fusion

Four industries each sell this operation a quarter of the answer.

Every platform in this market is built to decide and act on figures nobody generated. That is the ceiling the industry works under, and it is where Terrafa starts.

Transformation

Sold by the data platforms, on figures entered somewhere else.

Efficiency

Delivered to the plant, while the money moves in the commercial layer above it.

Advantage

Calculated on the market outside, never on the product inside.

Protection

Guaranteed at the moment the inspector is present, and nowhere between.

Terrafa is all four at once. One operation, held as one model, standing on a physical foundation it produces itself. That combination is why a commitment, a claim, a draw and an audit can rest on the same account of the business.
/Solutions

What the operation does differently on Continuum.

DEFEND THE PREMIUM

Product sold as what it demonstrably is

The renewable or specification premium backed by the operation's own record, shared with the buyer on its own terms.

CLOSE THE CLAIM

Settled at the interface, before it becomes a dispute

Intakes and outturns read against their paperwork while the parties are still on site.

BLEND TO SPEC

Tuned to what the contract owes

The margin held back as safety allowance stays in the batch instead of leaving with the product.

FUND ON WHAT IS THERE

Exposure sized rather than estimated

Working capital and the borrowing base rest on stock the operation can show, and duty is known at the transaction.

Continuum Core · Map · Physical embedding Feed live
Continuum Core map view: a terminal site plan with live tank and meter readings pinned onto it, each value carrying its uncertainty, and one flow drawn provisional because it is under-determined.
A checked number, on the operation's own site plan. Every pinned figure carries the confidence attached to it, and the berth-to-tank flow is drawn provisional because the platform cannot yet determine it.
Every figure, checked

14,203 bbl ± 118. 9,882 bbl ± 96. 312 bbl/h ± 22. No number appears without the confidence attached to it, and the meter's error ellipse names its least-sure direction.

What it cannot check

The berth-to-tank flow is under-determined, so it is drawn provisional rather than asserted. The platform shows what it does not know as plainly as what it does.

Your plan, not ours

Readings are pinned onto the operation's own site image in image coordinates.

/A day in the operation

The same Tuesday, run on Continuum.

06:00 · INTAKE

The delivery is read against its paperwork on arrival

The mismatch that used to surface three days later becomes a conversation at the gate.

09:20 · THE TANK

The retest list is set by exposure

One precautionary retest stands down, another moves up, and the laboratory budget follows the risk.

13:05 · THE BLEND

Blended to the specification the contract owes

Tuned on the parcel as it stands rather than on last week's sample.

16:40 · THE COMMITMENT

The desk commits before the window closes

Priced on what the operation holds, what it is contracted to, and the cash behind it.

18:15 · THE CLOSE

Finance closes against real stock

One account of the operation, all day, in every seat.

/Industries

Any business whose money sits in physical product.

The same objects everywhere: an asset holding product, a movement between assets, a quantity, a quality, a commitment, an obligation.

RefiningWhat the plant is making against what it is committed to sell
Chemicals and petrochemicalsGrade, yield and commitment held together
Terminals and storageCustody of other people's product, and the record that defends it
Blending and distributionRecipes tuned to the contract, premium defended
Trading with physicalThe book reconciled to the product it stands on
Midstream and pipelinesBalances and transfers settled where they occur
Bunkering and marine fuelsQuantity and quality agreed at the point of supply
Biofuel productionProduct that changes in storage, and buyers who expect evidence
Biomethane and digestionFeed, output and obligation on one account
Energy from waste and biomassWhat enters the plant decides availability and margin
Groups and portfoliosEvery site and jurisdiction as one position
The molecules arriving nextHydrogen, ammonia and sustainable fuels, from first molecule
/Customers

A multi-site bioenergy generator, UK and Ireland.

The operation

Five generating sites, one under development for carbon capture. Fuel arrives from several suppliers with moisture, calorific value and contamination declared on certificates. Generation, availability and the obligation sit apart from fuel purchasing.

What was put in

Reading points at intake and in storage, and one account holding fuel, supplier contracts, generation and the obligation together. Fixed scope, tested against the operation's own history before anything was relied upon.

What it showed

The gap between what was declared and what was delivered into the boiler, per supplier and per load, in cost per megawatt hour. Moisture variance moving output. Contamination traced to source.

What changed

Loads accepted, rejected or repriced on evidence at the gate. Suppliers ranked on what they deliver. Generation planned against the fuel physically present, and exposure under the obligation known at the transaction.

Working from the same account
The offtakerA position backed by evidence rather than a chain of declarations.
The suppliersPaid for what they deliver, with disputes settled at the gate.
The lenderExposure under the obligation sized instead of estimated.
The verifierOn the capture site, every tonne removed rests on figures that can be underwritten.
/Generated at the source

Every platform in this market stands on figures that already exist. Ours produces its own.

The physical side is read where the product sits, alongside the testing that standards require. Coverage is laid into the operation and maintained as a service, so it is an operating decision rather than a capital project.

READ IN PLACE

Composition and condition read without taking the product anywhere or altering it, across everything an operation handles.

COVERAGE, NOT INSTRUMENTS

Reading points supplied and maintained by us, priced per point, funded from spend that already exists.

WHY IT COMPOUNDS

Intelligence built on entered figures stops improving at the quality of its input. Ours improves with every point put in, across every operation.

Continuum Core · Network · Provenance As of 14:32 · Live
Continuum Core network view: measured leaves feeding transfer functions that compose into dashboard figures, with uncertainty propagated along every edge and one figure drawn provisional because it is under-determined.
How the platform answers the question. Every figure is a graph, not an entry: measured leaves on the left, the transfer that combines them in the middle, the figure it produces on the right.
Where the number came from

fig.total_inventory reads 24,085 bbl ± 152, composed from two measured tank levels through a declared sum. The figure is not typed in anywhere; it is derived, and the derivation is on screen.

Uncertainty that survives the maths

Where the transfer is linear the σ is exact. Where it is not, the platform refuses to linearise and switches that branch to Monte-Carlo rather than quietly approximating.

The honest gap

fig.expiry_risk is drawn provisional: its frailty term is not identifiable from these leaves. The platform states what would pin it down instead of returning a number it cannot stand behind.

Nobody should have to keep track of how good their own numbers are. That job normally falls to whoever remembers which figure came from a fresh reading and which came from a three-week-old certificate. Here it is not a job at all: confidence is carried by the structure, so it survives every step between a reading and a decision without anyone maintaining it.

It is baked in because the graph is measure-theoretic. A leaf is a measure, a transfer is a density against it, and a figure is a conditional expectation E[X|𝒢] — the projection of the quantity onto what is actually known. Every value is adapted to the filtration ℱt, so nothing on screen was computed using information that had not arrived yet.

Which is why linearity is checked rather than assumed. Through a linear transfer, uncertainty propagates exactly: σ_out = √(J Σ Jᵀ) is the answer, not an estimate. Through a curved one the same formula is a first-order approximation that usually understates the spread, so the platform refuses to linearise and switches that branch to Monte-Carlo. Slower, and correct.

/How it is laid in

It starts on the systems you already run.

Read-first, no replacement and no parallel running. A first operation is live inside a quarter, against acceptance agreed in writing before anything is relied upon.

01

Connect

Read-first connections to the systems the site already runs, and to the files where the working numbers live today.

02

Generate

The physical side produced at the source, with coverage deepening the account wherever the operation wants it.

03

Fuse

One account of the operation, reconciled continuously, every figure carrying where it came from and how sure it is.

04

Serve

Every seat gets its own view of the same account: operations, commercial, finance and management.

05

Decide

The platform recommends, acts under authority the operation sets, and declines when the figures do not support the call.

/Product

One platform. Four products.

Core● Live

Continuum Core

"What do we have, and what should we do about it?"

The operation on one account, a view for every seat, and decisions weighed against the whole business.

Reach◦ 2027

Continuum Reach

"Can we show it to the people we sell to and borrow from?"

The account opened outward on the operation's own terms, to customers, counterparties and capital.

Vectis◦ 2027

Continuum Vectis

"What is forming, and when should we act early?"

The forward view, calibrated against the operation's own record of what actually happened.

Sense◦ Deepening

Continuum Sense

"How deeply can the operation be read?"

Reading points laid in and maintained as a running service, priced per point, deepening the account.

/Pricing

Per operation, funded from spend that already exists.

Everything is operating expense. Nothing sits on the balance sheet and nothing waits on central capital approval.

Continuum Start

Fixed scope · one quarter

Brings the operation onto the platform against agreed milestones, tested on its own history before it is relied upon.

The platform

Per operation, per year

One operation, one fee, growing as more of the business runs on it. No seats and no per-user arithmetic.

Coverage

Per reading point, per year

Reading points supplied and maintained as a service on multi-year terms, funded from the testing budget already in place.

The operation sees several times what it pays, against giveaway, demurrage, claims and the working capital tied to stock it cannot show. Rates and a worked example on your own numbers: info@terrafa.uk

/Questions worth asking

Including the hard ones.

How is this different from market intelligence platforms?

They describe the market outside the fence, from cargo movements and reported prices. Continuum describes the operation inside it: what you hold, what it is worth, what you are committed to, and what to do next.

How is this different from an industrial data platform or a historian?

Those organise what the site's instruments already report about the equipment. Continuum holds the product, the contracts, the duty and the cash together, and produces the part of the picture that no existing system holds.

How is this different from a scheduling or optimisation system?

Optimisation improves a plan against the numbers it is given. Continuum decides what those numbers are, and prices the decision against the physical position, the contract and the cash at once.

Does it replace our laboratory, terminal or trading system?

No. It connects to them read-first and leaves them running. Nothing is ripped out and nothing runs in parallel.

Does it replace mandatory testing?

No. Accredited testing is set by standards and stays exactly where it is. Continuum adds continuity between those tests and directs discretionary retesting by exposure rather than by the calendar.

Is this a sensor with software attached?

It is the other way round. The operation runs on the platform, and reading points deepen what the platform can see. The platform runs without waiting for them.

What happens when it is not sure?

It says so. A figure it cannot stand behind is marked as such, and the platform names what would resolve it rather than returning a number that looks certain.

Who owns the data, and can our own systems use it?

The operation owns its data, and it is never shared, pooled or sold. Figures are served to your analysts, your spreadsheets and your own systems with their reliability attached.

Do we need hardware to start?

No. A first operation goes live on the systems the site already runs. Coverage is added afterwards, wherever the operation wants the account deepened.

How does this behave with AI agents?

Every value we serve carries where it came from and how sure it is, so a system acting on our account knows when to proceed and when to stop. That is the difference between an agent that acts and an agent that should not.

Who is behind it?

A UK company answering to operators rather than to an equipment maker's roadmap, with sensing science originating from Imperial College London and advice from King's College London, and a board drawn from refining, advanced fuels and terminal operations.

How long until it is live, and what does it cost?

A first operation is live inside a quarter. Pricing is per operation and per reading point, all of it operating expense. Write to us for rates and a worked example on your own numbers.

/Resources

From inside the operation.

EXECUTIVE BRIEF

What an operation is losing between its systems

Where the money goes when the physical, the commercial and the fiscal are held apart.

MARKET

What fifty-five platforms all promise

Unify, contextualise, decide, act. One shared foundation, and the layer none of them is buying.

REGULATION

Twice the litres, and the same certificate

The compliance arithmetic moved to physical volume, and the quality risk moved with it.

WORKED EXAMPLE

Your operation, on your own numbers

We build the case on your history before you commit to anything. Write to us and we will scope it.

The infrastructure gets built once.

We talk to operators, partners and investors building toward the same thing: commodity operations that run on figures worth trusting. If your business commits capital against numbers nobody has checked, we should talk.

Book a walkthrough Ask for pricing