Omnel · Aviation · a programme for training departments, in the Gulf and beyond

Training evidence your pilots can hold, your instructors can trust, your authority can verify.

Evidence-based training grades nine competencies and asks every operator to keep its instructors concordant. Two things the industry has never had: a grade the pilot keeps, with the instructor's reasons and the pilot's own words on it; and a concordance programme whose yardstick was fixed, in public, before the cycle began. Both run on Omnel today, for other sectors.

No machine grades anyone. People grade; the infrastructure keeps the record and makes it unfalsifiable.
Nothing chosen after the fact. The yardstick, the sessions and the disclosure rule are sealed before the first grade.
The key is not ours. Identified records open only with a key the pilots' representatives hold.

The three mechanisms A first study

machines that grade a pilot. Instructors grade; the record is kept, signed and logged.

key for identified records, held by the pilots' representatives, never by us and never by management.

competencies of evidence-based training on each grade record, with the instructor's reasons beside them.

volunteer instructors for a first study; a result in six months, published whatever it says.

Three open problems

Written by the regulators themselves, not by us.

Evidence-based training is the method the industry chose. Its own guidance names what it has not solved: how to keep instructors concordant without turning the metrics against them, how to keep the pilots' trust in what is done with their training data, and how to show that a change in training worked.

Instructor concordance is “the main area in which further maturity and guidance is needed.”

EASA research on data-driven EBT, final report, 2024

The rules “do not specify statistical thresholds of minimum variance of concordance.”

EASA, EBT manual, 2026 edition

Instructor metrics must never serve “to spy on instructors or to pressure individuals to change their grading.”

IATA, EBT implementation guide, second edition, 2024

Access to training data must be “agreed by all parties involved”: management and the flight crew's representatives.

EASA, oversight guidance for baseline EBT, 2025

Researchers add two findings of their own: examiners who grade the same session differ widely while each reasons well; and the debriefing, where the pilot explains their performance, leaves no record of what the pilot said.

Three mechanisms

A receipt, a yardstick, a key.

Each one already runs on Omnel for other sectors: Reviewed for the decision a person can verify, the Register for the evaluation that cannot be rewritten, the public log that anyone can witness. The programme gives them the vocabulary of the flight deck.

01 · Reviewed, for training

The heard grade

After a session, the instructor grades the nine competencies and writes the reasons. The pilot adds their own statement, in their own words, through a short guided form that no model touches. The grade, the reasons and the statement are bound together and signed with the instructor's own device. The pilot receives a receipt that verifies offline, with a published key, in ten years as today. Nobody can rewrite it: not the operator, not the instructor, not us.

Pilot: Keeps the receipt. Reads the reasons. Has their statement on the record, next to the grade.
Instructor: Signs what they wrote. Golden cases and a minimum reading time, applied by the system, keep the quality of every record.
Operator: Holds the file. The public log carries a fingerprint of the record, never its content.
Verify: Anyone with the receipt checks the signature against the published key, without us.

Illustration. Nine grades, the reasons, the pilot's words, one signature, and the receipt the pilot keeps.

02 · The Register, for the concordance programme

Concordance you can prove

Before the cycle, the training department deposits its yardstick: the metric, the acceptable spread, the share of golden cases, the decoys, and the rule for who learns what. The deposit is fingerprinted and time-stamped in a public calendar. Instructors then grade the same sealed sessions blind, each with their reasons. The figure comes out with an exact bound, not an average. Each instructor sees their own reliability, alone. The operator sees the fleet, never a name. And the log shows the yardstick never moved.

Before: Yardstick and sessions sealed; disclosure rule agreed by management and the flight crew's representatives.
During: Blind grading through tokens; golden cases hidden among real sessions; reasoned disagreement recorded as a category, not erased.
After: One figure with its bound, a signed report, and a line in the public log; a two-year history your authority can verify without us.
Who sees: The instructor: themselves. The operator: the fleet. The authority: the chain. Nobody: a name next to a number.

Illustration. The yardstick is dated before the first grade; the marks settle; the log keeps the order.

03 · Reviewed, blind

The key your pilots hold

Identified records, the names and the sessions behind the grades, are encrypted in the browser before they leave it, with a key the pilots' representatives hold. Omnel stores fingerprints, signatures and log lines; it cannot read the records. The operator reads the fleet's figures, not the names. Every opening of a record writes a line in the log, so the absence of a line is itself the proof. The data protection procedure that evidence-based training asks the parties to agree on stops being a promise and becomes something a pilot can check.

Where: Encrypted at rest; the programme is built for residency in the United Arab Emirates.
Key: With the flight crew's representatives, as the gatekeeper of flight data monitoring has long been; here the gate is cryptographic.
Log: Public, chained, witnessed by machines that are not ours. A record opened is a record logged.
Status: In construction: the blind, in-country variant is the first thing the programme adds to what runs today.

Illustration. The key sits with the pilots' representatives; the counter of openings stays at zero, and that is the proof.

The calibration loop

One session a year measures the session. A clip at a time measures the instructor.

Concordance programmes usually rest on one exercise a year: every instructor grades the same recorded session, and the spread is read. One session cannot tell an instructor who grades severely from one who read that session differently, and it shows only the behaviours that session happened to contain. The loop keeps the yardstick and changes the instrument: short sealed clips, a few minutes at a time, all year round.

Illustration. With one session the band stays wide open: the session and the instructor cannot be told apart. Each clip narrows it, until the loop can say something about the instructor rather than the session.

1
Seal
Three examiners grade each clip apart, and point to the moment that shows each behaviour. Where they agree, their grade becomes the reference; where they do not, that behaviour is set aside rather than imposed. The reference is fingerprinted and logged before any instructor sees the clip.
2
Grade
Each instructor receives a clip chosen for what it can still teach about their own grading, and grades it blind: the competencies, the behaviours observed, the moment of each.
3
See
Then the reference opens: the examiners' grades, their reasons, the moments they pointed to, and where the fleet's grades fell. The instructor's own figures are theirs alone.
4
Watch
The loop picks the next clip where an instructor's figures are least certain, and follows each instructor's agreement with a test that stays valid however often it is read. A drift is seen when it starts, not at next year's exercise.

The library stays. Every sealed clip keeps its examiners' grades and reasons: the department's judgment, kept when examiners retire. If they wish, it is also the ground their measured echo answers instructors from — Tacit, measured against them, and saying when it does not know.

Observable behaviours

Grading on behaviours gives agreement three layers, not one.

When the grade is built from observable behaviours, the record holds more than a number: which behaviours were seen, and when. The loop measures agreement on all three.

The grade

Within one point of the reference, with its bound, and the direction of any gap: more severe or more lenient.

The behaviours

Seen when they happened, missed when they happened, or noted where nothing happened.

The moment

Whether the behaviour noted is the one the examiners pointed to. A behaviour noted without a moment counts as unsupported.

Three patterns, three remedies

A column lights upEveryone diverges on one behaviour: the definition needs work, not the instructors.
A row lights upOne instructor diverges on almost everything: a calibration conversation, never a sanction.
A block lights upOne clip divides everyone, examiners too: the clip leaves the library.

Illustration, not data. Instructors in rows, behaviours in columns, clips in blocks: where the disagreement sits says what to fix.

What the machine does

Transcribes in-country. Indexes candidate moments for each behaviour, to navigate a clip, never to grade it. Chooses the next clip. Sorts the reasons instructors give, to find the same grade given for different reasons. Keeps the examiners' reasoning at hand.

What it never does

Grades a pilot. Sets the reference. Ranks an instructor. Shows an instructor's figures to anyone but the instructor, unless the rule deposited before the cycle says so.

Between clips, the everyday grades speak too. Each instructor's severity is estimated over all their sessions, then confirmed against peers who graded the same pilots — the method EASA's research describes for every module. What supports that analysis is not shown to the instructor, and it never decides anything alone.

Beyond a dashboard

What a concordance dashboard does not do.

Outlier charts on everyday grades exist, and so do yearly video exercises. The loop is built to answer the questions they leave open: for the instructors, for the pilots they grade, and for the authority that oversees both.

The reference is sealed, and so is its doubt

The examiners grade apart; their agreement becomes the reference, sealed and logged before anyone grades. Where the examiners themselves split, the behaviour is set aside: no instructor is ever scored against a reference its own examiners could not hold.

The decisions the draw made

On everyday grades, a severity model that accounts for who was graded finds the below-standard grades a severe instructor gave where an average instructor would more likely than not have met the standard, and the reverse. A second look for a pilot, not a label on an instructor.

A change you can prove

A new grading method or a standardisation day is evaluated by a question deposited before the change: the metric, the periods, the cohort. The base refuses it once the period has begun. The answer comes once, by the rule deposited, « no change shown » included.

Where the disagreement sits

Instructor, behaviour, clip: the disagreement is decomposed, and each pattern comes with its remedy (rewrite a definition, hold a conversation, retire a clip) rather than a list of names.

Read often, never wrong for it

Every instructor's figures carry a bound that stays valid however often they are read, and a drift detector whose false alarms are bounded by design. « Not yet clear » is said, with about how many clips remain.

Verifiable without us, kept by the department

Deposits, seals, reports and answers are lines of a public log, and reports are signed. The library of sealed clips, grades and reasons stays the department's judgment when its examiners retire.

Plainly

What runs today, what the programme adds.

We do not describe as built what is not. The green rows are in production and proven against it; the amber rows are what a first training department would help us finish.

Element	Status	What it is
Reasoned decision, the person's statement, signed receipt	runs today	Reviewed: the decision signed with the reviewer's own device, the statement assembled without a model, the receipt verifiable offline with a published key.
Protocol deposited before results, blind judging, exact bound	runs today	The Register: items in escrow, a draw with public randomness, judges behind tokens, decoys, a bound rather than an average, a signed report.
Public chained log, witnesses, time-stamping	runs today	Every protocol, report and receipt leaves a fingerprint in a log anyone can replay, cosigned by witnesses, with a kit for a witness that is not ours.
Open specification, terms for organisations, data processing agreement	runs today	Published and readable before any contract; the specification lets anyone verify a report without us.
Concordance metric with its bound	runs today	Agreement within one grade point as the deposited metric, with its exact bound; reasoned disagreement kept apart; the far-off rate (two points or more from the reference) counted on its own.
Calibration loop: sealed references, blind clips, three layers of agreement	runs today	References sealed and logged before a clip is seen; blind grading through seats the operator issues and each instructor opens and holds; agreement on the grade, the behaviours and the moment, each with its bound; the next clip chosen where the figures are least certain.
Drift watched, everyday grades read	runs today	Each instructor's agreement followed with a test valid at every look; severity estimated on everyday grades and confirmed against peers who graded the same pilots.
The decisions the draw of the instructor made	runs today	A many-facet severity model on everyday grades: below-standard grades from instructors standing severe where an average instructor would more likely than not have met the standard, and the reverse. For a second look, never a verdict; never shown to an instructor.
A change you can prove	runs today	The evaluation of a change, deposited while its after-period has not begun (the base refuses it otherwise), answered once by the deposited rule, the answer logged.
The department's console and the instructors' grading room, in English	runs today	Programmes, clips, the examiners' sheets and the seal, seats, the fleet's reading, evaluations and everyday grades in the console; each instructor grades on their own screen, from a seat whose key the department never holds.
Competency grade record on the nine competencies	programme	A record type for evidence-based training: grades and observable behaviours, the instructor's reasons, the pilot's statement, one signature.
Encrypted records under the pilots' representatives' key, in-country	programme	Encryption in the browser, the key with the flight crew's representatives, the store in the United Arab Emirates; we hold fingerprints only.
The frozen minute: freeze, two seats asked at once, validated transcript, blind coding	study	A three-to-five-minute block in the training phase; the pilot's own words at second zero, kept by the pilot; fleet maps for the training needs analysis. Proposed, not built.

A first study

Pre-registered, with volunteers, published whatever it says.

The way to try this is not a contract. It is one training department, thirty volunteer instructors, ten recorded sessions, and a protocol written before anyone grades. The result comes out in six months, and it is published because the rule to publish it was deposited first.

1
Scope
One fleet. Thirty volunteer instructors. Ten recorded simulator sessions, sealed before the draw. Written clearance from the operator and from the flight crew's representatives before the first grade.
2
Yardstick
The metrics named and deposited: agreement within one grade point with its bound, the confidently-wrong rate, the share of reasoned disagreement. The disclosure rule agreed by all parties, before the figures.
3
Grading
Blind, through tokens, each grade with its reasons. Pilots who consent add their statement to their own session's record and receive the receipt.
4
Result
One figure with its bound, a signed report, a line in the public log, and a registered report written for review. No machine graded anyone. Nothing left the country.

The open specification Terms, data processing agreement, service level How we write protocols before results

A second study · the frozen minute

What a pilot thinks at second zero, and whether the crew thinks it together.

Evidence-based training grades what an instructor can see. What a pilot thinks when a failure appears, nobody can see: it exists only in the pilot's own words, and only at that moment. The frozen minute captures it. In a recurrent simulator session, a block of three to five minutes. A failure drawn at random from a sealed pool. The simulator freezes; nobody acts. A neutral recorded voice asks the captain and the first officer, separately and at the same time, what they see, what they think, what comes next. Forty-five seconds each. Then the session resumes.

Illustration. One frozen minute in a sixteen-second loop: the scenario runs, the failure falls, the simulator freezes, the voice asks, both seats answer at once, each validates their transcript, the pilot keeps the receipt. The operator sees one line in the log.

It records what no grade can

The competency grade says how the event was managed. The minute says what each seat understood at the moment it began, in their own words: the priority they named, the threat they saw, the plan they had, the doubt they declared.

It belongs to the pilot

The transcript is read back and validated by the pilot, then kept by the pilot for ten years under the key the pilots' representatives hold. The operator reads the fleet, never a name. The log proves that no minute was opened.

It feeds the training needs analysis

Evidence-based training asks every operator to build its programme on its own data. The minute adds a layer no other source has: how the fleet reads each failure, by seat, fleet and experience band. The next cycle is measured with the same yardstick, so the department can show that a change worked.

How it runs

Six steps, fixed before the first minute

DrawThe failure comes from a pool sealed before the cycle, drawn by the Register's public draw at the session. Nobody chose this failure for this crew.
FreezeThree to five minutes inside the training phase, never the check. The simulator freezes, nobody acts, the instructor presses one button and says nothing.
AskA neutral recorded voice, the same words for every crew, published in advance. Captain and first officer answer separately and at the same time, forty-five seconds each.
KeepTranscribed in-country, read back to the pilot, validated as their own words; the audio is destroyed. The pilot receives the receipt; the record is encrypted under the pilots' representatives' key.
CodeA codebook deposited before the first minute: the priority named, the threats, the memory items, the declared uncertainty, the resources called, the plan. Human coders, blind to name and rank, with their concordance and its bound. A model may sort; it never scores a person.
MapFleet maps by failure, seat, fleet and experience band, never an individual. A registered report asks one question: does the crew's divergence at second zero predict how the event is managed?

Not an assessment, not a check, no profile of anyone. Fluency is excluded from the codebook by deposit, because a fleet speaks many first languages. No personal device in the simulator. Status: a proposed study, nothing built; the first thing a volunteer department would design with us.

Illustration. Each dot is one seat's first words; the two seats of one crew do not always land in the same column. That divergence is the measurement.

Later, and only as research

The examiner's echo, and the fleet asked the same question

Experience leaves with the people who hold it. Omnel's Tacit keeps an expert's judgment as cards the expert publishes one by one, with an echo that answers colleagues from those cards alone and says when it does not know, measured in the Register. For a training department this becomes two studies, not products. We will never claim such an echo is faithful. We claim to measure it, against the person, and to publish the floor it reached.

The reference examiner's echo

A senior examiner's judgment, kept as cards they publish themselves; an echo that helps calibrate instructors and outlives a retirement, measured against the person.

The fleet asked the same question

The same operational scenarios put to every volunteer instructor's echo, each answering alone, to map where a fleet reads the same procedure differently.

Tacit, in detail The Standard it is measured against

Illustration. One scenario, every echo answering alone: most read it one way, a few another, some abstain. A research instrument, pre-registered, never an assessment of a pilot.

What it is not

Three lines we will not cross.

Not an assessment by a machine

No model grades a pilot, ranks an instructor or infers anything about a person. People grade. The infrastructure keeps the record and the order of events.

Not a screening tool

No psychological, medical or physiological data enters the programme. The records are training records: grades, reasons, statements. Nothing in it touches licensing or fitness to fly.

Not a tool of management over crews

The key is with the pilots' representatives, the disclosure rule is agreed before the figures, and every opening of a record is logged where anyone can see it.

Talk to us

A person reads it, and writes back.

Written from the United Arab Emirates, for training departments in the Gulf and beyond. If your department runs evidence-based or competency-based training and would consider a first study, say so at the door; a person answers.

Request access What Omnel sells today

Or write to support@selione.ai.

What helps us answer well

Your training framework: evidence-based, competency-based, or on the way.
How concordance is kept today, and what your instructors think of it.
Who represents your flight crew on training data, and whether a research or ethics route exists.
Your window: a fleet transition, a new simulator centre, a recurrent cycle.

We run no demonstrations on real training records. The illustrations on this page are illustrations; the first real figures you see will be yours, under a protocol you deposited.