Why Nova

The margin for error has gone

Thermal coal margins have compressed hard — worldwide. Across the major coal-exporting regions, the average export margin fell to about US$26 a tonne, from US$41 a year earlier, as prices slid toward the cost of production for much of the world's supply. In a world like that, efficiency is not an initiative — it is survival. Every tonne, every dollar of cost, every hour of availability now decides the quarter.

And timing is everything. Late news is bad news: by the time a problem shows up in a monthly report, the tonnes and the dollars are already gone. You cannot recover a shift that ended three weeks ago. The advantage goes to whoever sees clearly and acts soonest — and that is now a question of intelligence, not effort.

A new era

The world has never generated data like it does now. A single operation produces more of it in a shift than anyone could read in a month — spreadsheets, scanned reports, onboard instruments, plans, emails, ledgers, sensors. Dozens of systems. Millions of records. Growing every day.

For as long as anyone can remember, using that data has meant the same thing: open the file, find the tab, click, scroll, search, and hope you remember where it lives next time. That way of working is coming to an end.

We are entering an era where machines read everything — millions and millions of records — and make sense of it, so the person doesn't have to. No more click and search. Ask.

An intelligence engine, not another dashboard

Nova is not one more place to go looking. It is an intelligence engine that reads across every system an operation runs, holds all of it at once, and turns it into one picture you can simply talk to.

Ask a question in your own words — how are we tracking, where is the opportunity, what changed this month — and Nova answers with the number, the trend, and the source it came from, in seconds. The knowledge that used to take a day, or lived in one person's head, is now at everyone's fingertips.

And the intelligence is not the chatbot. It is built to live beneath the conversation — in how the data is read, linked and made trustworthy, in a smart layer that knows where every answer comes from, and in named skills that do the work the way a good office would. That foundation is real and running today.

Why you can trust it

An engine is only worth trusting if the numbers are real.

• Grounded. Every answer is traced to the document and page it came from.
• Proven. Before a figure is served, it is reproduced against the operation's own records — not asserted, proven.
• Governed. One engine, one truth, the right view for each person, all logged.
• Honest. Where a number is still to come, Nova shows the gap as a gap. It never draws a chart on an invented figure.

What idiya builds

It is easy to sell the picture — the glowing brain, the big dream. The hard part is the how and the when: reading the real data, at real scale, and making it answerable and trustworthy. That is the part everyone skips, and the only part that matters.

idiya builds the engine underneath the dream — the intelligence foundation an operation can actually stand on. It is already running, on a live operation, on real data. The mine does the digging. Nova is the copilot.

Nova — Operational AI

Prologue — why now

The economics of mining have turned. Thermal coal prices fell around a quarter through 2025, to roughly US$100 a tonne, and are expected to slide further in 2026 — down toward the marginal cost of production for a large share of the world's suppliers. Margins compressed with them: across major exporters the average export margin dropped to about US$26 a tonne, from US$41 the year before, and on current forecasts roughly a third of global production sits at or below the line where it makes money at all.

In a world like that, efficiency stops being an initiative and becomes survival. When the margin is US$26 a tonne and competition is climbing, a tonne left on the pad or a soft month at the wash plant is the difference between a profitable quarter and a loss.

And the clock matters as much as the number. Late news is bad news. By the time a problem surfaces in a monthly report, the tonnes and the dollars are already gone. The value is not in knowing what happened; it is in knowing in time to do something about it.

A new era

Every operation on earth is now generating an ocean of data — more, in a single shift, than any team could hope to read. The data is not the problem. Making sense of it, at that scale, in time to act — that is the problem, and until now it has had no real answer.

Nova is that intelligence engine. It reads millions of records across every system an operation runs, unifies them into one living picture, and lets anyone ask a question in plain words and get the answer, with its sources, in seconds. No navigation. No remembering. No search. Ask.

What Nova does

It reads everything

Every kind of record — structured and unstructured, typed and scanned, current and historical — including the numbers locked inside scanned and instrumented reports. The whole record, not a sample.

It makes sense of it

Millions of records from unrelated systems become one coherent, queryable picture — connected, so a question can be answered across the systems, not one file at a time.

It answers, instantly

Ask in your own words and Nova answers with the figure, the trend, and the source — in seconds.

It shows, live

A full operations view — Executive, Coal Chain, Dragline, Rail & Port, forward Mine Plan — every figure drawn live from the same engine, with one-click export to PDF and Excel.

It governs itself

One engine, one truth, the right slice for each person — board view, site view, contractor view — all logged.

It is honest about its edges

Where a data feed is not yet connected, Nova shows the built structure plainly rather than a fabricated number.

How it works — millions of records to one intelligence

• Read at scale. Every inbound record captured faithfully — including figures trapped inside scans — the whole corpus.
• Unify into one picture. Records from every system connected into a single model of the operation.
• Make it answerable. The unified picture made queryable and provable — every answer arrives with its lineage.
• Keep it current and open. One governed engine feeds the dashboards, Nova's answers, the exports and any other system.

Intelligence, layer by layer

It is easy to assume the intelligence is the language model. It is not — that is the whole design. Nova distributes intelligence across layers, with the largest share built to live below the model.

• 1. Intelligent data. Every record read, linked, domained, typed, tagged with provenance, reconciled and made answerable — before any question is asked. This is the moat.
• 2. The smart feed (MCP). A governed interface that knows what data and tools exist and assembles the right context for a question.
• 3. The skills. Named, versioned, testable units of know-how — reconcile-the-month, brief-the-board, chase-a-variance — auditable, improvable and portable.
• 4. The language model, at the edge. The interpreter and translator — the friendly last mile, not the brain. Swap the model and you change the voice, not the intelligence.

Proof it is real

This is not a concept. On Xanadu Basin (Polaris Mining Group, XSX: PLR), today, Nova reads the whole operation's record, unifies it, answers from it with sources, and drives a live operations site — and every figure it serves is reproduced against the site's own workbooks before it is shown.

Operational AI — past, present, future

• The past — understood. Nova reads the entire record and makes sense of it. Live today.
• The present — watched. Every record read as it lands; the state of the operation judged on arrival.
• The future — project, prescribe, prevent. What is coming, what to do about it, and the incident that never happens because the pattern was caught early.

Reading millions of records is where operational AI begins. Project, prescribe, prevent is where it goes. The mine does the digging. Nova makes sense of everything else.

The anatomy of an 8-week deployment

Eight weeks, not eight quarters

Most enterprise AI dies in the gap between "interesting" and "in production." idiya closes that gap in under eight weeks — not by cutting corners, but by running a disciplined path that front-loads the hard part (the data) and proves value early. Here is what the timeline actually looks like.

Weeks 1–2 — Discovery & data

We map the decision you want to improve and the systems that feed it, then point IdiyaCORE at your real files. No new data entry: we read what you already have — spreadsheets, scans, instrument feeds, ledgers — and stand up the ingestion. By the end of week two, your record is being read at scale.

Weeks 3–4 — Unify & prove

The records from every system are connected into one model, each number carrying its provenance. Every figure is reproduced against your own workbooks before we trust it, checked against source rather than just asserted. You see the first unified picture and confirm it matches reality.

Weeks 5–6 — Build the surface

Now the intelligence gets a face: the dashboards, the Data Field API, and — where you want it — an AI analyst built in, so anyone can ask questions in plain words and get answers with sources. We build against your actual use case, not a template.

Weeks 7–8 — Validate & go live

We run the system alongside your current process, tune it on real usage, wire up access and governance, and cut over. You go into production with something that is already trusted — because it has been proving itself against your own numbers for weeks.

Why it sticks

The reason our clients stay (100% retention) is that we don't hand over a model and walk away — we embed until it runs, and we keep running it. Eight weeks to live; then it only gets stronger as models improve, because the intelligence lives below the model, in the data engine we own.