The standard

Proof of Learning

Protocol version v1-mock-to-mastery · Event schema v4-full-chain-evidence · Transfer contract v1-authored-delta

Last reviewed 2026-08-19 · Studara

What this document is

The rules, and every way evidence is refused

Studara will not say a student has learned something on the strength of a correct answer. This document is the complete rule set we require before we will say it, and the complete list of ways evidence is refused.

It is published so that a student, a parent, a tutor or a school can check whether we followed our own rules, rather than take our word for it.

What this document is not

This protocol is not a claim that it works

The design hypothesis is that joining an immediate unassisted challenge, delayed parallel-form retrieval, authored transfer and exam-like performance is more useful than treating any single correct answer as mastery. That hypothesis has not been tested against a simpler retrieval-practice baseline, and until it has, the sequence is not an efficacy claim.

We publish that here, first, because a standard that hides its own unproven status is not a standard.

We make no claim about grade improvement. We have not run a controlled outcome study. When we do, it will be published on the evidence page whichever way it comes out.

The chain

Evidence is accepted in one order and no other

Each stage must be reached from the one before it.

The 9 stages, in order
`unobserved`	Nothing has been measured.
`diagnosed`	A specific cause of the lost mark has been identified.
`repair_in_progress`	Teaching has occurred. This is not evidence.
`immediate_challenge_passed`	Answered unassisted, on a different question.
`delayed_retrieval_passed`	Still answered after a real delay, unassisted, on a parallel form.
`transfer_passed`	Answered when the surface changed.
`exam_performance_passed`	Answered under exam conditions.
`verified`	A receipt has been issued.
`review_due`	Previously verified, now due for re-checking — or reopened because later unassisted, measured evidence contradicted it.

Stage 1

A diagnosis names one of six causes

The last is a permitted answer. A system that cannot say it does not know will always say something.

The 6 diagnosis kinds
`knowledge_gap`	The required knowledge was not there.
`prerequisite_gap`	Something earlier in the chain was missing.
`misconception`	A specific wrong belief produced the wrong answer.
`exam_process`	The method was known; the exam procedure lost the mark.
`retrieval_failure`	It had been learned and could not be recalled.
`uncertain`	We could not identify the cause. This is a permitted answer.

The rules

R1 to R11

R1 — Teaching is never evidence

The repair stage is recorded as unmeasured and contributes nothing to proof. A student being taught something proves nothing about whether they learned it. No amount of instruction advances the chain.

Refused as: the repair_started branch records the stage without evidence

R2 — Assisted work is never evidence

Every accepted attempt must carry assistancePresent: false. An answer produced with a hint, a worked example on screen, or a prompt is not counted.

Refused as: assisted_attempt

R3 — A question may never be reused

Enforced twice on every stage: a repeated question is rejected, and a repeated attempt is rejected. You cannot reach a verified receipt by answering the same question five times.

Refused as: reused_item and reused_attempt

R4 — Each stage must be a different question than the last, and must name it

Every stage records the item it came from, and the attempt is rejected if that reference does not match the previous stage's item, or if it is the same item. The chain is therefore linked: a stage cannot be skipped, reordered, or satisfied by an unrelated question.

Refused as: challenge_not_independent, not_parallel_form and not_exam_like

R5 — The delay is a real calendar delay

The elapsed time must be at least 1 day, measured from the previous stage's timestamp rather than from the start of a session. A retest later the same minute is not a delayed retest. A receipt records the actual number of days that elapsed for that student — we publish the real figure, never a typical or representative one.

Refused as: delay_too_short

R6 — The delayed check must be a parallel form

The relationship to the immediate challenge must be exactly parallel_form: a different question testing the same method. Anything else is rejected.

Refused as: not_parallel_form

R7 — Transfer must declare what changed

A transfer attempt must carry a contract naming its class and the dimensions that changed, and that statement is checked. A cosmetic rewording cannot be presented as transfer.

Refused as: invalid_transfer_contract

R8 — Exam conditions means four things, all of them

An exam-stage attempt must be timed, must carry the mark tariff, must use an exam-like response format, and must be aligned to the specification source. Any one missing and it is rejected.

Refused as: not_exam_like

R9 — Unmeasured evidence is refused, not estimated

Evidence is accepted only when a rule decided it or a checked marker decided it. Anything unmeasured is refused. Where we did not measure, we say so rather than infer.

Refused as: unmeasured_evidence

R10 — Fail closed

If evidence is missing, malformed, out of order or duplicated, the stage is not awarded. The default is not proven. There is no partial credit and no benefit of the doubt.

Refused as: every branch returns a refusal rather than a downgraded pass

R11 — Every refusal is kept

The protocol replays an append-only event stream and retains every exclusion for audit, with its reason. Rejected evidence is not deleted. A record that only keeps its successes is a marketing artefact, not an audit trail.

Refused as: replayProofOfLearning returns the full exclusion list

The transfer classes a contract may declare

The 5 transfer classes
`representation`	The same method in a different representation.
`surface`	The same method behind different surface details.
`discrimination`	Choosing this method when a similar one would be wrong.
`integration`	The method combined with another to reach an answer.
`far_transfer`	The method applied in a distinctly different context.

What counts as measured

Evidence validity
`deterministic`	A rule decided it. Accepted.
`validated_marker`	A checked marker decided it. Accepted.
`unmeasured`	Nothing measured it. Refused — see unmeasured_evidence.

R11

All 13 ways evidence is refused

This list is complete. It is rendered from the same array the protocol refuses with, so a reason cannot be added to the product without appearing here. Rejected evidence is retained for audit rather than deleted.

The 13 refusal reasons
`invalid_event`	The event was malformed, or a required reference was missing or over-long.
`duplicate_event`	That exact event had already been accepted. Replaying it changes nothing.
`wrong_stage`	The stage before it had not been established. Stages cannot be skipped or reordered.
`unmeasured_evidence`	The evidence carried no measurement. We do not infer where we did not measure.
`assisted_attempt`	A hint, a worked example or a prompt was present. Assisted work is never evidence.
`attempt_not_passed`	The attempt did not pass. A failed attempt does not advance the chain.
`reused_attempt`	That attempt had already been counted at an earlier stage.
`reused_item`	That question had already been counted at an earlier stage.
`challenge_not_independent`	The immediate challenge did not reference the diagnostic question, or was the same question.
`delay_too_short`	Fewer than 1 day elapsed since the previous stage. A retest in the same session is not a delayed retest.
`not_parallel_form`	The delayed check was not a parallel form of the immediate question, or was the same question.
`invalid_transfer_contract`	The transfer attempt did not declare a valid class and the dimensions that changed.
`not_exam_like`	The exam attempt was untimed, untariffed, not in an exam response format, not source-aligned, or did not follow the transfer question.

Conformance

When a receipt may cite v1-mock-to-mastery

A receipt may state that it was issued under this protocol only if all of the following hold. A receipt failing any of them is marked unverifiable or incomplete, and says so on its face.

Every stage from diagnosed to exam_performance_passed is present, in order.
Every accepted attempt was unassisted.
No question reference and no attempt key appears twice.
Each stage names the previous stage's question as its source, and differs from it.
At least 1 day elapsed before the delayed stage, and it was a parallel form.
The transfer stage carries a valid contract naming its class and changed dimensions.
The exam stage was timed, tariffed, exam-like in format and source-aligned.
Every stage's evidence was decided by a rule or by a checked marker.
The receipt's assertions are complete.

Receipts

A receipt is one of four things

Two of them are unflattering. We print those.

The 4 receipt statuses
`verified`	The full chain was established.
`review_due`	Previously verified; it should be re-checked. This is not a failure and it is not hidden.
`superseded`	A later receipt replaces this one.
`unverifiable`	The chain is broken or incomplete. The receipt still exists and still says what it could not establish.

Changelog

Versions of this standard

2026-08-19 — v1-mock-to-mastery published

First public release of a protocol that has been enforced in the product since before publication. Publishing it changed the documentation, not the rules.