Fig. 01 Environment ap-ledger-01 · three rollouts
DataOrigin builds simulation-based environments for frontier labs in hard science: Nuclear physics, CAD, metallurgy, civil, mechanical engineering and software development. Others sell RL tasks; we sell the lab. We train the next skill, multi-step research, where agents design experiments, run simulations and debug results. The simulator grades every step, so there is almost no room for reward hacking. Code models had GitHub to learn from. Physics up to nuclear and civil have nothing like it, and that's where models are weakest, so models cannot pick it up from the internet in pre-training. 1,000+ India's PhDs audit the tasks at a fraction of the cost, and every model failure we find makes the next environment harder.
01 · Environment. A replica of the production system, seeded with de-identified state. Every action the agent takes changes it.
6
production systems replicated as environments
820
tasks written from real tickets and incidents
4,100
graded checkpoints across the task set
160
practising specialists writing and grading tasks
42%
of ingested records rejected before delivery
0
scraped or unlicensed records in the catalog
Fig. 02 Why this exists
Sample tasks from the environments we are building. Pick a domain, open a task and see how frontier models responded.
Every lab reads the same web and buys the same chips. What still separates models is the distribution they were trained on, and the distributions that matter now, how a claim is settled, a build fixed, a chart reconciled, were never published. They sit inside institutions. Reaching them takes a licence, not a crawler.
Institutional records cover the tasks models fail at today: diagnosis, code review, claims handling, reconciliation.
Public text keeps the conclusion and drops the reasoning. Trajectories inside a real system keep both.
A benchmark score is a claim. A graded run inside a replica of the production system is a measurement.
Fig. 03 How it works
Every product is a stage of the same loop, and every stage leaves an artifact a lab can inspect. Shown for one episode of ap-ledger-01.
1Environment
A snapshot of a production system its owner licensed to us, identities removed, surrounding services stubbed to production behaviour.
2Rollout
Tasks are written from real tickets, with the edge cases, approval steps and failure modes left in. Every action changes state.
3Grade
A gate, then weighted criteria, written and applied by someone who does the job. Deterministic where possible, judgement where not.
4Dataset
Graded trajectories become datasets with licence and reviewer on every row, or the environment runs live as a reward source.
{ "episode": "ep-0412",
"env": "ap-ledger-01",
"score": 0.65,
"licence": "DO-L-2026-018",
"reviewed_by": "DO-EXP-0688",
"pii_residual": 0 }Fig. 04 Environments
Pick an environment and the console loads a task from it: the trace, the rubric, the row it became, and the full task card with instruction and method. It advances on its own until you touch it.
—
—
Instruction · as the agent receives it
Rubric · gate and weighted criteria
Methodology
Dataset
Results by model
Fig. 05 Data licensing
The records behind the environments are also licensed on their own. Each comes directly from the institution that produced it, is de-identified inside its walls, and ships with a licence ID and a named reviewer on every row. One sample of each is below.
{
"thread_id": "T-2026-0481", "channel": "whatsapp", "opened": "2026-03-04T09:12:00Z",
"student": "«STU-7f3a»", "counsellor": "«CNSL-21»", "queue": "admissions",
"subject": "Eligibility · MSc Data Science · September intake",
"messages": [
{ "from": "student", "t": 0, "text": "I have a three-year BCom at 64%. Can I apply for the September data-science intake, or do I need a bridging year?" },
{ "from": "counsellor", "t": 540, "text": "Three-year degrees are accepted from 60%. You need an English test by 15 June and a short statement on your quant background. Sending a shortlist of three programmes inside your budget." },
{ "from": "student", "t": 2210, "text": "Got the shortlist. Starting the first application today." }
],
"outcome": "application_started", "next_step": "english_test_booked", "handle_time_s": 2210, "csat": 5,
"licence": "DO-L-2026-018", "reviewer": "DO-EXP-0231"
}
--- a/bookings/reminders.py+++ b/bookings/reminders.py@@ -61,6 +61,10 @@ def due_reminders(appts, now):- window = (now, now + timedelta(hours=24))- return [a for a in appts if window[0] <= a.start_utc < window[1]]+ tz = ZoneInfo("Australia/Melbourne")+ local_now = now.astimezone(tz)+ window = (local_now, local_now + timedelta(hours=24))+ due = [a for a in appts if window[0] <= a.start.astimezone(tz) < window[1]]+ # one reminder per appointment, even when the DST shift makes two windows overlap+ return [a for a in due if a.reminder_sent_at is None]
Compare in clinic-local time, not UTC — the Fitzroy clinic got two reminders on the changeover night. Guard on reminder_sent_at as well.
Done in the second commit, with a test pinned to the April changeover.

It wasn't typical. But how do we make it… maybe not all the companies need to become a unicorn.
Unicorns are a vanity metric that kind of like, you know, they make waves, they put us on the map, but it's definitely not a KPI that we should follow of how successful an ecosystem is.
And still, if we want to create a sort of an incubator for systemic success stories and successful companies, what is needed and how far are we from it? Or are we there yet?
Well, I think from… I think we are there yet. From a certain angle.
And that angle is that thing that, you know, the American dream of you can start from, you know, you go to America, you start from anywhere and you can become a fully accomplished, you know, business person with everything that can be achieved.
And I think that American dream has been democratized and is now part of our infrastructure as well. And I think this is the most important part.
| Invoice | Vendor | PO | Goods received | Posted | Exception |
|---|---|---|---|---|---|
| INV-8633 | PO-12889 | «DATE−2d» | P12 | — | |
| INV-9404 | PO-36893 | «DATE+1d» | P12 | posted before receipt | |
| INV-7863 | PO-34519 | «DATE−5d» | P12 | — | |
| INV-6516 | PO-47997 | «DATE−1d» | P12 | — |
Conclusion. Four of twelve shown. One misstatement in the sample: INV-9404 (52,080.23) recorded in P12 with goods received in P13; reclassification entry posted at C-4.1. Projected 118,400 against performance materiality 135,000. Cut-off assertion supported after adjustment.
Fig. 06 Graders
Practising physicians, staff engineers, chartered accountants and counsel write the tasks and grade the runs. Credentials are confirmed with the issuer, the exam is timed, the trial is paid, and every grader is re-calibrated weekly on seeded gold sets.
Fig. 07 Provenance
Every record traces to a licence signed by its owner and a reviewer who practises in the field. Built so a hospital or a bank can sign without lowering its standards, and a lab can audit what it received.
Identifiers are removed inside the owner's environment. Each redaction is logged and residual risk is reviewed by a person.
Each record traces to a licence ID, a consent basis and a permitted use, written into the manifest.
Scope, exclusivity, term and revocation are set by the owner and enforced contractually after delivery.
Encrypted pipelines, per-access logs and audit rights for both the owner and the licensee.
src · where records are licensed from · rev · where graders sit
Fig. 08 For data owners
Hospitals, enterprises, engineering organisations and contact centres license what they already keep. We de-identify inside your walls, build and deliver, and report every licensee.
Clinical notes and imaging. Code, reviews and CI. Tickets, CRM, email and procurement. Calls, meetings and first-person video. The by-product of the work, not the work itself.
A one-time licence or a revenue share. You set scope, term, exclusivity and revocation, and the contract enforces them after delivery. No resale outside the agreed scope.
Direct and quasi-identifiers removed in your environment before anything moves. Every redaction logged; residual risk reviewed by a person; spoken identifiers cut from audio.
Per-delivery statements showing which licensee received what, under which terms, and what was paid. Revocation is honoured downstream. Nothing moves until the licence is signed.
For AI labs and enterprise teams
Tell us the capability gap. If an environment or a dataset covers it, you will have samples under NDA within days.
For data owners
License it on your terms. We handle de-identification, compliance and delivery; you keep control of scope, term and exclusivity.
partnerships@dataorigin.ai · we reply within two business days