TRUSTXD EXAMOS navigate  ·  F fullscreen  ·  01 / 10
Strategic brief · 2026 Confidential — for evaluation

The end of paper leaks.

TrustXD ExamOS is zero-leakage assessment infrastructure — hardware-attested identity paired with just-in-time exam generation, so a national examination can be sat from home with cryptographic integrity.

0papers in existence before a candidate authenticates
2synchronised camera angles, temporally cross-matched
3permitted verdicts — and no integrity score, ever

Prepared for national testing agencies, ministries of education and university examination boards.

01 — The problem

Exams do not fail at the margins. They fail at the design.

FAILURE 01

The paper exists before the exam

A printed paper is a valuable secret held by presses, couriers, strongrooms and centre staff for weeks. Leakage is what that design produces when it works as built.

FAILURE 02

Identity is a moment, not a state

A photograph and a signature at a desk at 9 a.m. Proxy substitution survives because nobody is checking who is sitting there at 10:40.

FAILURE 03

Integrity is reported as a score

Remote proctoring returns "87% suspicious" and a career turns on it. A candidate cannot cross-examine a number; a court cannot audit one.

Each failure ends the same way: cancellation, re-examination, litigation, a lost year for candidates who did nothing wrong — and a lasting reduction in public trust that no subsequent administration fully recovers.

02 — Why now

Four conditions arrived at once.

None of these existed together five years ago. Every previous attempt at at-home high-stakes assessment failed on at least one of them.

  • 01Sealed-chip attestation is universalApp Attest and Play Integrity ship in every current phone — a cryptographic proof that a frame came from real hardware.
  • 02A second camera is already in the roomThe dual-angle rig needs no new hardware for the overwhelming majority of candidates.
  • 03Generative models can rewrite inside boundsReliable enough to re-parameterise items within declared ranges — and now cheap enough to do it per candidate, at national scale.
  • 04The legal frame is settledThe DPDP Act 2023 tells us exactly what consent, retention and redress must look like. Compliance is now a spec, not a guess.
03 — The insight

You cannot leak a paper that does not exist yet.

Every incumbent guards the secret harder — more strongrooms, more cameras, more biometric gates at the centre door. We removed the secret.

  • ADon't guard the paper — don't create itPapers are assembled per candidate at T−0, after identity clears, and never exist in advance.
  • BDon't check identity once — hold itTwo cameras at an arbitrary angle, correlated continuously, with a sealed chip proving the hardware is real.
  • CDon't score integrity — record itDeterministic evidence, three verdicts, an independent human on every flag, and the candidate reads the same file.
04 — The productFive modules · three assurance levels

Three locks in series. The paper opens last.

MODULE A

Identity lock

On-device ID parsing, active liveness, a 468-point facial mesh, and a second camera at 60°–120° whose motion must correlate with the first. Every frame signed by a sealed chip.

MODULE B

Display lock

A 3-second token painted on the exam screen, read by the phone, returned over an independent channel and validated inside a 2-second window on the server's own clock.

MODULE C

Paper lock

The item bank stays encrypted until locks A and B both hold. Then the generator writes this candidate's paper — and the IRT calibrator proves it is the same test.

MODULES D + E

Evidence & rights

Deterministic trails, three verdicts, independent reviewers, itemised bilingual consent, continuous self-view, and a 30-day purge that proves itself.

Level 1 browser-only · Level 2 dual-camera · Level 3 dual-camera plus attested status agent for national examinations.

05 — The hard question

"Different papers can't be fair."

They can — if you can measure it. Item Response Theory turns fairness from a promise into an enforced tolerance, checked before any candidate sees an item.

  • ±0.05logit tolerance on difficultyEnforced per item. A variant outside the band is rejected, never delivered.
  • 0.094marks — typical form effect on a 60-item paperAgainst a standard error of measurement of 3.16 marks.
  • 3%of the noise a candidate already carriesThe paper they were given is the smallest source of uncertainty in their score.
Base item (dotted) Generated variants, n = 8 ±0.05 parity envelope

Item characteristic curves. Full derivation in the psychometric equivalence specification.

06 — Defensibility

The system your legal cell can actually defend.

NO BLACK BOX

There is no integrity score

No field stores one, no API returns one, no configuration enables one. Every flag names the rule and the threshold that produced it.

THREE OUTCOMES

Clear · Review required · Not assessable

Nothing else. "Not assessable" says the evidence was insufficient — it says nothing about the candidate, and always carries a free re-sit.

REPRODUCIBLE

The same trail, seven years later

Append-only, signed, deterministic. A tribunal replays the session and gets byte-identical evidence — with the rule-set version it was judged under.

Candidates see their own evidence, their own reviewer access log, and their own retention clock. A system nobody can inspect is not a trust system, whatever it detects.

Findings of misconduct remain the institution's own decision. TrustXD supplies the evidence file and holds no opinion on guilt — which is precisely why the evidence file is worth something.

07 — Market & model

Priced per session, sold per programme.

  • 01National testing agenciesMulti-shift competitive examinations — the anchor customer and the hardest proof point.
  • 02State boards and common entrance testsHigh volume, high political sensitivity, strong equity mandate.
  • 03Universities and deemed universitiesInstitution-run entrance tests where centre logistics dominate cost.
  • 04Banking and public-sector recruitmentFrequent cycles, large cohorts, existing remote-assessment budgets.

Revenue: per-candidate-session licence priced by assurance level, plus an annual platform subscription, plus a funded civic-node programme for candidates without a second device.

Where the cost actually goes

Indexed comparison of delivered cost per candidate. Centre-based delivery is set to 100.

Illustrative model, not a quoted price. It removes venue hire, printing, secure transport, invigilator deployment and candidate travel; it adds compute, media egress, reviewer capacity and the civic-node programme. Actual figures are modelled per programme during a pilot.

08 — Path to national scale

Earn the national exam. Do not start with it.

P0 · SHADOW

One university test

~5,000 candidates, run alongside the existing centre-based exam. Results not issued from TrustXD.

EXIT: ZERO DELIVERY INCIDENTS

P1 · PARALLEL

Results issued

Two university tests plus a banking recruitment shift, ~120,000 candidates, live results.

EXIT: FLAG RATE ≤ 4%

P2 · REGIONAL

State entrance test

~600,000 candidates with civic nodes in aspirational districts, proving the equity model.

EXIT: NO DISTRICT SKEW

P3 · NATIONAL

Full-scale administration

A national competitive examination at 2.5 million concurrent sessions.

EXIT: CLEAN STATUTORY AUDIT

Each phase publishes its own evidence: appeal-upholding rates, Not-assessable rates by district, and an independent DIF review. If the equity numbers do not hold, the phase does not advance.

09 — The ask

Run one paper on it. Then decide.

  • WE ASKOne examination, one cohortRun in shadow beside your existing delivery, with your own psychometricians reading our parameters.
  • WE ASKA named audit contactLedger access from day one, not after the pilot report is written.
  • WE ASKTwelve weeksFour to integrate, two to seed the item families, six to deliver and analyse.
  • YOU GETThe complete evidence ledgerYours, exportable, in a documented open format. No lock-in and no data hostage.
  • YOU GETPublished fairness analysisFull DIF sweep and form-equivalence report, whatever it says.
  • YOU GETA costed national modelDelivered cost per candidate for your programme, built from your pilot's real numbers.
Request the pilot brief Read the PRD Architecture blueprint

TrustXD ExamOS · · Concept deck. Figures marked illustrative are modelling assumptions, not results.