miatz™
Log inDocs Score
Platform tour

Four modules, and where each one stops.

Miatz Academy compiles the course. Miatz Certify makes it a certification. Miatz Kanan guards the attempt. Miatz T'aan interrogates the answer. Here is what each does, plainly — including the limits, because you'll find them anyway.

Get your Docs Score Talk to us
Context

The modules are stages of one loop

Academy is stage four. Certify is stage five. Kanan and T'aan are how stage five stays honest.

  1. 01

    Ingest

    Docs, API specs, repos, tickets — one knowledge graph.

  2. 02

    Score

    23 criteria. Per-page defects. A prioritised fix queue.

  3. 03

    Regenerate

    Rewrite what fails. You approve every block.

  4. 04

    Compile

    Tracks, lessons, labs, quizzes — each one cited.

  5. 05

    Certify

    Blueprinted exams. Signed Itz'at credentials.

  6. 06

    Distribute

    Static academy, Mysty tutor, MCP, exports.

  7. 07

    Observe

    Answer gaps, decay, renewals — back to stage one.

The modules

What each one actually does

Every module gets the same treatment: what it does, then what it doesn't. The second list is the one worth reading.

Miatz Academy

The living course, compiled from the corpus you already maintain.

What it does

  • Turns the knowledge graph into a curriculum graph: role-based tracks — admin, developer, support, CSM, partner implementer — then modules, lessons, labs and formative quizzes.
  • Every lesson cites the source chunk it was derived from, which is what makes provenance visible and makes automatic invalidation possible when the source changes.
  • Runs the LMS around it: enrollment, cohorts, prerequisites, scheduling, streaks, community spaces, notifications, SSO, team and seat management, commerce.
  • Ships the Mysty tutor beside the learner — grounded only in your corpus, citing anchors, saying "I don't know" and escalating rather than inventing. Powered by Mysty Pie.
  • Exports as SCORM 1.2/2004, cmi5 and QTI 3.0 so the same compiled content can run inside an LMS you already own.

Where it stops

  • Compiled content does not publish without a subject-matter expert approving it. That gate is mandatory and it costs your team time — budget for review, not just for ingestion.
  • Output quality is bounded by corpus quality. If the Docs Score comes back low, the honest sequence is regenerate first, compile second.
  • An LTI 1.3 provider — embedding Miatz courses inside an enterprise LMS — is roadmap, not shipping.

Miatz Certify

The governance that separates a certification from a quiz.

What it does

  • Blueprint builder: weighted domains drawn from the curriculum graph, so a form's composition is defensible rather than arbitrary.
  • Item banks with AI-drafted items under mandatory SME review, a beta pipeline that seeds unscored items, and per-item statistics — difficulty, discrimination, exposure.
  • An Angoff-lite cut-score wizard that walks your SME panel through setting a pass mark, so you never ship an unexamined 70%.
  • Randomised parallel forms, server-authoritative timing, attempt immutability, an audit trail, a retake policy engine and accommodations flags.
  • Exam versions bound to product releases, plus expiry and renewal policy — maintenance modules, free re-assessment, renew-by-badge or version-bound validity.

Where it stops

  • A product certification is not a licence and not accreditation. It attests that someone cleared your scheme, and the scheme's credibility is yours to build.
  • Braindump exposure is managed, not eliminated: item rotation, exposure caps and similarity forensics reduce it — nothing removes it.
  • Cut scores come from your SMEs. We supply the workflow and the statistics; we do not supply the judgement about what competent means in your product.

Miatz Kanan

Proctoring built in-house — kanan, Yucatec for to guard.

What it does

  • Identity verification at session start, then capture that is consent-gated and scoped per jurisdiction: webcam, screen and audio.
  • AI monitoring for presence, gaze, a second person, a second screen and audio anomalies — each flag routed to a human review queue rather than auto-failing anybody.
  • A sealed evidence locker with retention windows, so exam artifacts are segregated, short-lived and auditable.
  • Kanan Live: a human proctor console with real-time video, chat, incident flagging and session attestations.
  • Kanan Journey: a minimal-permission browser extension that observes a candidate performing real user journeys in your product against checkpoints compiled from your own tutorials and acceptance criteria.

Where it stops

  • Capture levels three and above are web-only today. The Miatz store builds deliberately do not declare camera or microphone permissions, so proctored attempts happen in a browser until that changes.
  • AI flags are signals for a reviewer, not verdicts. We do not ship a system that fails a candidate without a human in the loop.
  • Capture ships only behind explicit consent flows and regional legal review — DPDP, GDPR and BIPA-class rules all bite here. Some jurisdictions will mean some levels are unavailable to you.
  • Kanan Journey watches only allowlisted exam domains, only during the session, with a visible recording indicator. It is scoped to see nothing else, ever — which also means it cannot assess work done outside those domains.

Miatz T'aan

An AI oral examination — t'aan, Yucatec for speech.

What it does

  • Conducts a voice viva: adaptive follow-up questions generated from the same exam blueprint the written form draws on.
  • Scores against a published rubric, with the full recording and transcript retained as evidence.
  • Sends borderline results to human moderation instead of resolving them silently.
  • Answers the braindump problem structurally: a memorised answer does not survive the second follow-up question.

Where it stops

  • Rubric scoring of speech is harder than marking a multiple-choice form. Treat T'aan as a strong signal that a human confirms at the margin, not an oracle.
  • Voice assessment has real accessibility implications — accents, speech differences, anxiety, bandwidth. Accommodations are a first-class setting and you should expect to use them.
  • It is slower and costlier per attempt than a written form. Use it where the credential warrants it, not everywhere.
Integrity ladder

Seven levels, chosen per credential

How closely an attempt is watched should match what the credential claims. Every level is tenant-selectable, and the honest framing is that each step up buys confidence and costs money, friction and privacy surface.

  1. Level 1

    Open

    Untimed practice, formative quizzes and free micro-badges. Nobody is watching and nothing is claimed.

    Use for the free learning layer. Makes no integrity claim at all — and shouldn't pretend to.

  2. Level 2

    Secured browser

    Timed exams from randomised parallel forms, in fullscreen, with focus-loss and paste telemetry and an honour agreement the candidate clicks through.

    Deters casual cheating. Does not stop a determined candidate with a second device.

  3. Level 3

    Kanan (AI)

    Identity verified, then webcam, screen and audio captured with consent. AI flags presence, gaze, a second person, a second screen and audio anomalies into a human review queue.

    Web-only today. Consent-gated and jurisdiction-reviewed; flags are reviewed, never auto-failed.

  4. Level 4

    Kanan Live

    A human proctor watches the session in real time — video, chat, incident flagging — and signs a session attestation at the end.

    Highest confidence per attempt, highest cost per attempt. Reserve it for the credentials that carry weight.

  5. Level 5

    T'aan (AI viva)

    An AI examiner conducts a spoken examination, generating follow-up questions from the blueprint and scoring against a rubric, with recording, transcript and human moderation on borderline results.

    The anti-memorisation layer. Needs accommodations planning before you switch it on.

  6. Level 6

    Practical & evidence-based

    Graded sandbox labs, project submissions reviewed by AI and an SME, supervised ride-along certification, and shadow assessment where the AI mentor scores real in-product work against criteria.

    Competence judged from work rather than from an exam. Slowest to grade; the most convincing to a hiring manager.

  7. Level 7

    Kanan Journey

    A browser extension — or an agent observing the session stream — watches the candidate complete real user journeys inside your actual product, scoring follow-through, path quality, time on task and recovery from injected error states.

    Capture is hard-bounded to allowlisted exam domains for the session only, with a visible recording indicator.

Levels three and above capture video, screen or audio. That capture is consent-gated, regionally scoped, retained on a clock and reviewed by a person before it affects a result. If your jurisdiction or your candidates make that a bad trade, the right answer is a lower level with a narrower claim — not a capture policy you have to defend later.

The output

Every level ends in the same credential format

Whatever integrity level you pick, a pass issues an Itz'at: an Open Badges 3.0 credential — which is a W3C Verifiable Credential 2.0 — signed under your own did:web, revocable through a public status list, and verifiable without calling us.

The credential records which integrity level the attempt ran at, so a verifier can tell an unproctored practice badge from a journey-assessed professional certification. That disclosure is not optional and not configurable — it is the thing that keeps the Itz'at mark meaning something across every tenant who issues one. How the Itz'at works.

What an Itz'at is not: accreditation, a degree, a diploma, or a licence to practise anything. It is a signed, revocable statement that a named person cleared a named scheme at a named integrity level, on a date. That is a smaller claim than most credentials make, and it is one we can actually stand behind.

Questions

Platform FAQ

Do we have to use the whole ladder?

No. The integrity level is selectable per credential, so a free foundations badge can sit at level one while your professional certification sits at four or seven. Picking a level is a cost and credibility decision, and we'd rather you picked deliberately than defaulted high.

Is proctoring available in the mobile app?

Not for capture levels three and above. The Miatz store builds deliberately don't declare camera or microphone permissions, so those attempts run in a browser. Learning, practice and level-one and level-two assessment work everywhere.

What if we already have an exam vendor?

Then keep them. External proctoring networks can be plugged in as a passthrough for tenants who need third-party attestation, and QTI 3.0 item banks import and export. Our position is that the in-house stack is better and cheaper for most schemes, not that you're obliged to use it.

How do you keep AI-generated exam items from being wrong?

Every AI-drafted item goes through SME review before it can be scored, then through a beta pipeline where it's seeded unscored until its statistics look sane. Items with bad discrimination get pulled. The AI drafts; the panel and the psychometrics decide.

What does the platform cost?

Pricing isn't published yet. We're scoping early engagements individually rather than publishing a number we'd have to walk back. Talk to us with your scope — modules, audiences, integrity levels, learner volume — and we'll come back with one.

Which of this is shipping today?

Ingestion, the Docs Score, and the compile-and-deliver path are the earliest phases; the certification engine, native Kanan and credential issuance follow; Kanan Live, T'aan, journey assessment and the war-room simulator come after that. We'll tell you exactly where a module sits when you ask about it rather than showing you a roadmap with everything marked available.

Start where the evidence is

Before any of this matters, we score the corpus it would all be compiled from. The audit is free and the report is useful on its own.

Get your Docs Score Back to Miatz for business