Read this before anything else
We measure one thing. We are not going to pretend it is three things. The number on the certificate is a score on a sealed exam. It is not a claim that the agent is accurate in general, that it does not make things up, that it is safe, or that it is better than anyone else’s.
The certificate cites the stated cutoff. We do not claim that is the true cutoff. Vendors state these dates loosely and revise them. If the true cutoff is later than the stated one, a question we treated as post-cutoff may already have been in the training data. There is no way to detect that from inside. We name the stated date and where we got it.
The seal proves the exam was not edited after it was sealed. It does not prove that a stranger witnessed the seal before grading. Until the hash is published, dated, somewhere public before grading, that part rests on our word, and we would rather say so here.
The sample below is the prototype running on local fixtures. It is not a live-web read. Every prompt is tagged [FIXTURE]. We have not pointed this engine at the internet.
The sealed certificateprototype on fixtures, exam-ord-1001-2025-09-15
Printed by the prototype on the frozen clock date 2025-09-15. Recorded answers are withheld. The taking agent was a scripted stand-in named agent-alpha, not a live model.
| Field | Value |
|---|---|
| Agent | agent-alpha |
| Run | run-ord-1001-2025-09-15 |
| Exam | exam-ord-1001-2025-09-15 |
| Stated training cutoff | 2025-06-01 |
| Cutoff citation | MODEL_CUTOFF in the local fixture file (fixtures.py; vendor_page_fetched=False) |
| Sealed at | 2025-09-15 |
| Graded at | 2025-09-15 |
| Score (exact match) | 2/3 |
| Admitted questions | 3 |
| SHA-256 seal | 100c372e2ff36edb61bb8dbe628ac038a151263927bd8752a704d2789657b449 |
| Certified | yes |
This certificate cites the stated cutoff. We do not claim that is the true cutoff. The seal proves the exam was not edited after sealing. The seal does not prove a stranger witnessed the seal before grading.
One public scoreboard rowfrom the same prototype run
The public board lists certified runs only. This is the row that matches the certificate above. The prototype later printed a second week-2 row; this page shows the row for this certificate.
| RUN | AGENT | DATE | SCORE | PCT | SEAL |
|---|---|---|---|---|---|
| run-ord-1001 | agent-alpha | 2025-09-15 | 2/3 | 66.7% | 100c372e2ff3 |
Verify any row: recompute SHA-256 over the sealed questions and answers and compare to the seal. The public board does not print recorded answers or refused question ids.
The free probe — prompts only3 admitted of 6 candidates
This is the honest half of the product. You can run these prompts against any agent. You cannot grade them from this file: we hold the answers, and the seal at the top proves we fixed them before the agent saw a prompt. The intended weekly pack is ten questions; this prototype admitted three from six local candidates. We are not going to invent seven more.
| What it says | |
|---|---|
| Stated training cutoff | 2025-06-01 |
| Cutoff source | fixtures.py MODEL_CUTOFF (invented; no vendor page fetched) |
| Sealed at | 2025-09-15 |
| Exam id | exam-ord-1001-2025-09-15 |
| SHA-256 seal of withheld answers | 100c372e2ff36edb61bb8dbe628ac038a151263927bd8752a704d2789657b449 |
| # | id | Prompt | Category |
|---|---|---|---|
| 1 | q-harbor-bridge | [FIXTURE] Which contractor was awarded the Harbor Bridge retrofit on 2025-08-14? | procurement |
| 2 | q-orbital-launch | [FIXTURE] What was the launch outcome of mission ORB-19 on 2025-07-02? | aerospace |
| 3 | q-rate-decision | [FIXTURE] What policy rate did the Northern Reserve set at its 2025-09-04 meeting? | macro |
Six candidates were offered. Three were admitted. Three were refused, for these reasons, as the prototype printed them: resolution_not_after_cutoff, sources_disagree, too_few_independent_sources. A question is refused if it resolved on or before the stated cutoff, if two sources disagree by so much as a full stop, or if only one organisation stands behind it. If nothing survives, you get no exam and no charge.
The cutoff this sample usesfrom the fixture file, not a vendor page
We did not fetch a vendor page. A cutoff we cannot cite is a cutoff we do not test a named commercial model against.
| Field | Value |
|---|---|
| Model | fixture-demo-model (invented; not a named commercial model) |
| Stated cutoff | 2025-06-01 |
| Source | the local fixture file, field MODEL_CUTOFF |
| URL | none |
| Vendor page fetched | no |
What $195 buys
One certified run against a stated cutoff we can cite: a sealed exam, graded by exact match, a dated certificate with the seal, and one public scoreboard row. A person emails you the certificate after you pay.
A weekly seat is $395 per month. That amount is quoted in the email thread, not on the rail above. Each week is a fresh exam, a new seal, and a new date, because a newer model makes last month’s questions pre-cutoff.
- Not a quality score.N out of M on this exam, nothing else.
- Not a live-web read, not yet.The sample is fixtures. Counting how many real questions survive exact agreement between two independent sources is a live-web test we have not run.
- Not witnessed before grading.The seal catches edits after sealing. A stranger did not watch us seal it first.
Ask before you pay
No pay button on this one yet. Email operations@ustechautomations.com. Name the model and the cutoff you want certified against. We reply with whether we can run it and the checkout link, before you spend anything.
What you would be paying for: $195 once, for one certified run against a stated vendor cutoff we can cite. A $395 per month weekly seat is quoted in the email thread, not on this page’s rail. There is no pay button on this page yet. After you pay, a person emails you the sealed certificate. One public scoreboard row is published for that run. The certificate is only as good as the stated cutoff we cite.
Email us the model and cutoffNo pay button on this page yet. The certificate is only as good as the stated cutoff we cite.