Glossary & guide for teachers

Summary (Inhaltsangabe)

What is a summary (Inhaltsangabe) and what matters most?

Picture a child telling a friend at break time what a book is about: briefly, factually, in their own words, without acting out the plot. That is exactly what a summary does. It stays in the present tense, keeps the events in order and leaves out direct speech and personal opinion. Teachers mostly look at completeness, the right tense and clear, neutral wording of one's own.

The exact weighting of the criteria is set in the end by your school or the curriculum.

See also: Retelling, Narrative

Narrative

What makes a good narrative in primary school?

A narrative takes the reader along on a real experience or an invented story and builds up tension: first an opening, then a climax where you hold your breath, and a rounded ending. Vivid words, direct speech and a clear thread make it come alive. Unlike a report or a summary, it is allowed to embellish and may be as gripping as it likes.

Which criteria count and how much is set by the teacher in their rubric.

See also: Report, Retelling, Picture story, Story from prompt words

Report

How does a report differ from a narrative?

A report is the sober cousin of the narrative: it informs factually and concisely about an event and answers the key questions: who, what, when, where, how, why. It uses the past tense, keeps the chronological order and drops tension, embellishment and personal opinion. Think of a newspaper article or an accident report: facts, not feelings.

See also: Narrative, Profile card (Steckbrief)

Dictation

What is assessed in a dictation?

In a dictation the teacher reads out a text that the children write down as accurately as possible: it tests confidence in spelling and punctuation. Usually the number and type of mistakes are counted, with a spelling error often weighted differently from a missing comma. How strictly it is counted is set in the end by your school or the agreement in place.

Description

What is a description and which kinds are there?

In a description the child paints a precise picture with words, so the reader can imagine the thing without seeing it: factual, ordered and in the present tense. Depending on the topic there are many variants: describing an object, a person, an animal, a process, a route or a picture. A process description explains step by step, like a recipe; a route description guides you safely from A to B.

See also: Profile card (Steckbrief), Report

Retelling

How do a summary and a retelling differ?

Both retell a text, but in a different tone. A retelling recounts a heard or read story vividly in one's own words: it keeps the suspense and the order, mostly in the past tense, and may be colourful. A summary, by contrast, condenses briefly and factually in the present tense. In short: a retelling wants to entertain, a summary wants to inform.

See also: Summary (Inhaltsangabe), Narrative, Fairy tale

Picture story

What is a picture story?

For a picture story the child gets a series of pictures, often a wordless comic strip, and turns them into one coherent, gripping story. The pictures set the order and the main events, imagination fills in the rest: names, feelings, direct speech. Teachers often look at whether all the pictures are linked sensibly and whether a clear thread emerges.

See also: Narrative, Story from prompt words

Story from prompt words

What is a story built from prompt words?

Here the child is given a handful of prompt words, say dog, thunderstorm and key, and has to weave a story in which all of them fit naturally. The prompt words are the spark for imagination and give a frame at the same time. As with any narrative, structure, suspense and a clear thread that ties the words together naturally are what count.

See also: Narrative, Picture story

Profile card (Steckbrief)

What goes into a profile card (Steckbrief)?

A profile card is the short form of a description: a clear list of the most important features, often in bullet points rather than full sentences. For a person, an animal or a character you list things like name, appearance, size, special traits and likes in an orderly way. You know it from animal fact cards or the classic friendship book: quick to grasp and boiled down to the essentials.

See also: Description

Letter

What is the difference between a personal and an official letter?

With a letter it all depends on the recipient. A personal letter to grandma or a friend may be warm and relaxed, with a greeting like Dear Grandma and a casual tone. An official letter, say to a school or a club, stays polite and factual, with a fixed layout: sender, date, subject line, a formal salutation and a closing formula. Both need a clear structure.

See also: Opinion piece / argumentation

Opinion piece / argumentation

What is an opinion piece or argumentation?

In an opinion piece the child argues a reasoned position on a question, for example whether phones should be allowed at school. Instead of just telling a story, you gather arguments, back them with examples and order them sensibly. In secondary school this grows into a full argumentation that also weighs the opposing side. The aim: a convincing, easy-to-follow point of view.

This form mainly comes in at secondary level and builds on the arguing practised in primary school.

See also: Letter

Fable

What is typical of a fable?

In a fable animals act like people: the sly fox, the vain raven, the busy ant. Each animal stands for a character trait, and at the end there is a lesson or moral you can carry over to real life. The stories are short and pointed. In class children often rewrite fables or invent their own to match a given moral.

See also: Fairy tale

Fairy tale

How do you recognise a fairy tale?

Fairy tales begin with Once upon a time and often end with happily ever after. In between anything is possible: talking animals, witches, magic and the number three that keeps coming back. Good and evil are clearly divided, and in the end good usually wins. Children write their own fairy tales or retell familiar ones, using these well-known building blocks.

See also: Fable, Retelling

Poem

What makes a poem, and what is an Elfchen?

A poem plays with language: with rhyme, rhythm and images, arranged in lines and stanzas rather than running text. In primary school short forms are popular, above all the Elfchen: eleven words spread over five lines (1, 2, 3, 4, 1). Such fixed blueprints take away the fear of the blank page and make writing playful.

See also: Fairy tale

Rubric (assessment grid)

What is a rubric and why does it help with correction?

A rubric is a structured assessment grid: per criterion (e.g. content, structure, language) it sets what earns how many points. Rubrics make assessments consistent across the whole set of papers and comprehensible for children and parents alike. In Laura, the rubric is freely configurable.

Laura ties every proposal to the stored rubric criteria (see the review/rubric module).

See also: Assessment matrix, Expectation horizon

Assessment matrix

Is an assessment matrix the same as a rubric?

Largely yes: "assessment matrix" is the common term in Berlin school practice for a criteria-based assessment grid. In Laura, binding rubrics set school-wide can be locked so that all teachers assess by the same standard.

The term's usage varies by federal state; the lock function (school_mandatory) is Laura product logic.

See also: Rubric (assessment grid)

Expectation horizon

What is an expectation horizon?

An expectation horizon describes in advance which content points a task should meet: the "model answer" for the assessment. It makes objective sub-tasks fairly comparable; for judgement tasks (e.g. an essay) pedagogical leeway remains on top.

Grade scale

How are points converted into grades?

A grade scale maps point totals to grades (e.g. grade 1 from a certain percentage). In most federal states it isn't binding uniformly by percentage, but is set at school or teacher level. Laura converts transparently by the disclosed scale: as a proposal, not as a fixed grade.

Percentage thresholds are guide values; to be justified in the paper header or internally at school.

Correction marks

Which correction marks does Laura use?

Correction marks are the teacher's usual margin notations (e.g. R for spelling, Z for punctuation, A for expression). Laura supports a configurable list and places them in the inline overlay at the sourced spot: which marks apply stays with the teacher or school.

Concrete mark sets vary by school or federal state; configurable in Laura (see the font/overlay docs).

See also: Inline correction overlay

Evidence Layer

What does "Evidence Layer" mean at Laura?

The Evidence Layer is the proof layer: for every assessment proposal, Laura shows the supporting spot and the original excerpt from the exercise book. It stays visible what a proposal rests on: the assessment rests on evidence, not on a gut feeling.

Laura's core principle: OCR/HTR delivers evidence, not truth.

See also: Evidence Card

Evidence Card

What does an Evidence Card show in the review view?

The Evidence Card is the UI card that brings together five layers per assessment criterion: evidence, interpretation, proposal, decision and audit. It makes visible why a proposal arises and prevents a pure "gut-feeling assessment".

See also: Evidence Layer

Confidence traffic light

What do the colours green, yellow, red and black mean?

The confidence traffic light shows how sure the handwriting recognition is at a given spot: green means confident, yellow asks for a second look, red calls for a check, and black stands for illegible. Uncertain spots go into a guided review view and are never quietly guessed. So the teacher sees at a glance where to look more closely. Black is not a fourth step on the same scale, it is a different statement: there is nothing legible here. That is not especially high uncertainty, it is a refusal to guess. This is why every colour is always accompanied by a word.

Like a traffic light: green may go, yellow means look again, red means stop and check.

See also: Confidence class, Review queue

Confidence class

What lies technically behind the coloured traffic light?

Behind every traffic-light colour sits a rating of how confidently a single word was recognised: confident, uncertain, critical or illegible. This rating decides whether a word is put forward for review and whether Laura may learn from it for the handwriting. It is, in a sense, the foundation from which the visible traffic-light colour arises.

Four steps from "clearly read" to "illegible": they steer what needs checking.

See also: Confidence traffic light, Confidence cascade (4 stages)

Review queue

Why doesn't Laura put every word up for checking?

Laura doesn't hand you every single word a second time. Whatever she read confidently passes through; only the shaky spots, the parts flagged yellow or red, land in a small queue. That way you work through the few doubtful cases on purpose, instead of re-reading text that was already clear. And whatever you decide there, Laura remembers for next time.

See also: The teacher decides, Manual correction (Edit)

Review requirement

Which spots must the teacher check before the grade?

Before the final grade, the teacher checks all red and black tokens as well as yellow tokens with low average confidence. This rules out an uncertain recognition flowing into the assessment unchecked.

See also: Review queue

The teacher decides

Does Laura hand out grades automatically?

No, never on its own. Laura makes suggestions and shows you openly where she is sure and where she isn't: but the grade is yours to give in the end. No result slips through unasked. Every move you make is recorded cleanly, so it stays traceable later who decided what. And it is precisely from these deliberate decisions that Laura keeps learning, not from guesses of her own.

This is the non-negotiable heart of it: no grade and no report card is ever produced at the push of a button.

See also: Review queue, Decision type

Human-in-the-Loop

What does "human-in-the-loop" mean at Laura?

Human-in-the-loop means: every pedagogically relevant decision is confirmed by a teacher before it applies. The AI takes on the groundwork and the proposals, but the human stays in the decision loop: Laura proposes, the teacher decides.

See also: The teacher decides

Proposal (assessment proposal)

What is a proposal in Laura?

A proposal is Laura's assessment suggestion before the teacher decides: including confidence and the marking that it comes from the AI. A proposal is never a fixed grade, but the starting point for the teacher's decision.

See also: Decision type

Override (teacher override)

Can the teacher override a Laura proposal?

Yes. With the override, the teacher actively overrides a proposal and replaces it with their own decision, with a justification. The process stays traceable in the audit trail, so it's clearly documented who changed what and when.

See also: Traceable trail, Teacher Override

Teacher Override

What is a teacher override?

"Teacher override" is the teacher's view of the override: the deliberate correction or rejection of an AI proposal by the responsible teacher. It's the mechanism that puts the core principle "the teacher decides" into everyday practice.

See also: Override (teacher override)

No-Surprises PDF

What's special about Laura's feedback PDF?

The no-surprises PDF is only created after the teacher has confirmed, corrected or discarded all proposals. It contains exactly the released decisions: no silent automatic changes. Partial results are shown transparently.

See also: XMP release metadata

Report-card module

How does Laura help with wording report cards?

The report-card module proposes wordings from sourced Evidence Cards, not from a free prompt. The teacher can edit or reject every proposal. Report-card sentences stay tied to documented performance instead of freely generated.

Concept/module description; wordings are always a proposal, the teacher decides.

See also: Evidence Card

Disadvantage compensation and grade protection (LRS)

How does Laura handle a reading and spelling difficulty (LRS)?

With a reading and spelling difficulty there are two levers. Disadvantage compensation only adjusts the conditions, for example more time or larger print, while the requirement stays the same. Grade protection goes further and adjusts the assessment itself, for example when spelling mistakes count for less. In Laura this is a stored profile that only takes effect when the teacher or the school actively switches it on, never on its own. What is allowed is set, in the end, by your school.

Whether and how far grade protection applies varies from state to state: your school's specifications are decisive.

See also: Assessment authority (rule hierarchy), The teacher decides

Assessment authority (rule hierarchy)

Who actually decides how grading is done?

Grading rules come about like a chain, from the top down. The federal level only gives a rough framework, as a recommendation. The state turns it into binding requirements that differ from state to state. The school fills them in, the subject group gets more specific, and right at the front stands the teacher, who decides for the individual child. Each level sharpens the one above it, and in a justified individual case one may depart from it.

Laura meticulously records who set or overrode which rule and when, so that every deviation stays traceable.

See also: Disadvantage compensation and grade protection (LRS), The teacher decides, Override (teacher override)

Data processing agreement (DPA)

When does a school need a data processing agreement?

When a service provider processes personal data on the school's behalf, the GDPR requires a data processing agreement. It governs the purpose, scope, security and deletion of the data. It's the basis for the school staying responsible under data protection law.

Art. 28 GDPR. DE: Auftragsverarbeitungsvertrag (AVV).

Data protection impact assessment (DPIA)

When is a data protection impact assessment needed?

A data protection impact assessment (DPIA) checks the risks of a processing operation for the people concerned in advance. For systematic processing of children's data it is generally mandatory. It documents risks and safeguards before a procedure starts.

Art. 35 GDPR. Whether mandatory in a given case is clarified by the school's data protection officer.

Deletion & retention log

How does Laura prove that data is deleted on time?

Deletion and retention events are logged, so it's provable when which data was deleted or retained. That supports the right to erasure and a verifiable retention concept.

Reference: Art. 17 GDPR (right to erasure); logged in a dedicated table.

GDPR

What does it mean concretely that Laura is built to the GDPR?

The GDPR is the EU General Data Protection Regulation. At Laura the safe path is the default: processing happens locally first, and processing in the cloud has to be switched on deliberately and is only possible on a legal basis, with a log. Stored data stays on servers in Europe. So data protection is Laura's preset, not a checkbox ticked afterwards.

For parents: at Laura data protection is the default, not a box you tick later.

See also: Local-first processing

Consent

When is parental consent needed?

Consent is one possible legal basis for processing that isn't already covered by school law, such as cloud processing of especially sensitive data. It must be freely given, informed and revocable. Which basis applies is clarified by the school or the data protection officer.

Art. 6 and Art. 9 GDPR; not legal advice: a check by the school is needed.

See also: Consent Gate

Pseudonymisation

What's the difference between pseudonymisation and anonymisation?

With pseudonymisation, clear names are replaced by identifiers; the mapping is stored separately and can only be resolved with additional knowledge. Anonymised data, by contrast, allows no personal reference any more. Laura keeps the clear-name mapping local and encrypted, separately.

Art. 4 no. 5 GDPR; implementation in Laura: the cloud only sees pseudonym IDs.

Technical & organisational measures (TOM)

What are TOM in data protection?

TOM are the concrete safeguards for personal data: e.g. encryption, access control, logging and deletion concepts. They prove that data is adequately secured and are part of a school's data-protection evidence.

Art. 32 GDPR.

Consent Gate

What is a consent gate?

A consent gate is a mandatory module before especially sensitive steps (e.g. audio processing or cloud escalation). It captures the legal basis, storage period and right of revocation before data of that kind is processed at all.

Reference: Art. 6 and Art. 9 GDPR. A product gate that enforces a legal basis.

See also: Consent, Processing mode

Local-first processing

Does the student data stay on the teacher's device?

Local-first means that by default nothing leaves your system. Processing anywhere else is not the standard, it is a decision you make: on controlled European servers or, only with a legal basis, via cloud services. An unknown or unset mode uploads nothing at all, it stops instead. Whatever is stored permanently sits on servers in Europe.

For parents: by default the work stays inside the school's own system. If anything does go outside, it is because the school deliberately chose that route, and the child's name stays local: only an identifier travels.

See also: Processing mode, GDPR

Row-Level Security (RLS)

How does Laura prevent schools from seeing other schools' data?

Row-Level Security is a database mechanism that binds every row to its tenant: a teacher only sees their school's data. Cross-tenant access is technically excluded. At Laura, RLS is mandatory on every table.

RLS on every table; school_id as the tenant boundary.

See also: Multi-tenancy, Tenant

Multi-tenancy

How are several schools kept apart in one system?

Multi-tenancy means: several schools or providers share one instance but are strictly separated by Row-Level Security; cross-tenant access is excluded by design. Laura runs on European servers, and each school stays in its own sealed space.

For parents: separate, locked filing cabinets in the same building.

See also: Row-Level Security (RLS), Tenant

Tenant

What is a tenant?

A tenant is an isolated school or provider unit within multi-tenancy. It's the boundary at which Row-Level Security takes effect: one tenant's data is technically invisible to other tenants.

See also: Multi-tenancy, Row-Level Security (RLS)

Processing mode

Where are the exams processed?

The processing mode sets where a task is handled: from the local default right on the device, through controlled processing on European servers, to cloud paths that are only allowed on a legal basis and once explicitly enabled. There is no silent shift into the cloud, every step outward is a deliberate decision.

Staged paths, always local first; which paths a school allows at all is set when it is configured.

See also: Local-first processing, Kill switch

Kill switch

Can a school switch off individual AI functions?

A kill switch is a switch that fully deactivates a whole module (e.g. cloud OCR or audio). A school can thus define at setup which processing paths are allowed at all: deactivated functions don't even start.

See also: Processing mode

Handling of external AI models

Does real student data go to external AI providers?

No. External model sources are used only along expressly regulated, anonymised paths; real student texts never go to public model platforms. The permitted path is clearly defined instead of being silently assumed. That keeps it controllable what leaves the house at all, and in what form.

For parents: no student text is passed on to outside AI services unasked.

OCR / HTR (text recognition)

How does Laura recognise handwritten text?

OCR (for handwriting: HTR) turns image content into text. Laura runs several recognisers in parallel (local paths first, a cloud fallback only with consent) and reconciles them, instead of relying on a single recogniser. The result is evidence, not a fixed truth.

Concrete models can change; the glossary deliberately describes the multi-recogniser principle.

See also: Tokens (OCR tokens), Confidence cascade (4 stages), Evidence Layer

Tokens (OCR tokens)

What is an OCR token?

A token is the single-word representation of the recognition result: the recognised text plus its position in the image (bounding box), its confidence and its confidence class. Tokens are the smallest unit on which Laura hangs evidence, proposals and decisions.

See also: OCR / HTR (text recognition), Confidence class

Confidence cascade (4 stages)

How does Laura decide whether cloud AI is needed?

The confidence cascade works in four stages: first two local, free recognisers run and compare their result. If they agree confidently enough, the exam is finished for free and locally. Only the contested spots go to two independent cloud processes and are compared again. Whatever stays uncertain after that is not guessed but put to the teacher.

Free and local first, cloud only for the few doubtful cases, and only if the school allows it.

See also: Confidence class, Confidence traffic light

Assessment Object Model (AOM)

What is the Assessment Object Model?

The AOM is Laura's generic data model: Assessment → Submission → Evidence → Proposal → Decision → Output. It's deliberately not German-specific, so further subjects can plug in. It's the blueprint along which every task is processed and documented step by step.

See also: Submission, Proposal (assessment proposal)

Submission

What is a submission?

A submission is exactly one student's hand-in for an assessment: the second level in the data model. Attached to it are the recognised tokens, the proposals and the decisions of that one piece of work.

See also: Assessment Object Model (AOM)

Evidence of a decision

What does Laura store about a teacher decision?

For every decision Laura keeps several layers: the spot in the exercise book, the machine-recognised text, the text confirmed by the teacher, the confidence rating and the type of decision, plus the necessary supporting data. This makes it fully traceable later what a grade rests on and who made it. Nothing is left to memory or guesswork.

Every decision stays fully documented and reconstructable.

See also: Decision type, Traceable trail

Decision type

What kinds of decisions are there?

Every move you make gets a clear label: accept, correct it yourself, unreadable, doesn't belong or look at later. That label tells Laura exactly what you did with her suggestion. And it is the switch that steers learning: only certain deliberate decisions may change Laura's picture of a handwriting, a mere click-through does not.

This way Laura learns only from what the teacher has truly confirmed.

See also: Manual correction (Edit), Safeguard against wrong learning

Manual correction (Edit)

What happens when I correct a suggestion myself?

Sometimes Laura reads a word wrong. You simply type in the right version, and that's it. This one time is enough: Laura saves your correction and keeps an eye out for it when the same handwriting shows up again. All that matters is that something is actually there; from an empty field she learns nothing. So she only improves where you deliberately stepped in.

See also: Decision type, Handwriting learning (writer-adaptive recognition)

Unreadable decision

How do you mark a spot as unreadable?

With the unreadable decision, the teacher marks a token as not legible. A corrected text is excluded here; the decision flows into the handwriting profile as a negative signal. "Not legible" thus honestly stays that way, instead of being guessed.

See also: Decision type

Not-relevant decision

What does "not relevant" mean for a token?

"Not relevant" marks tokens that shouldn't flow into the handwriting profile: such as drawings, numbers or special characters. Laura thus only learns from real handwritten text and not from artefacts that say nothing for handwriting recognition.

See also: Decision type

Escalation (escalate)

What happens with an escalation?

With escalation, the teacher passes a conflict decision on to the school management. A meta event is created without a handwriting-profile update, and a justification is mandatory. Escalation is one of the governed paths when a conflict can't be resolved on the spot.

See also: Conflict resolution, Rationale

Merge-edit strategy

How does Laura resolve two contradictory correction states?

With the merge-edit strategy, the teacher types a new final text that replaces both contradictory states. A new edit decision results from it. No state is silently overwritten: the teacher deliberately sets the authoritative text.

See also: Conflict resolution, Manual correction (Edit)

Conflict resolution

What happens if the same exam was corrected on two devices?

Laura overwrites nothing silently, but asks the teacher which state applies. There are governed strategies: keep the local state, keep the server state, merge or escalate. The resolution thus stays a deliberate decision, with a trace.

See also: Merge-edit strategy, Escalation (escalate), Offline outbox

Tamper-proof chain

How does Laura ensure decisions aren't changed after the fact?

All log entries of a piece of work are linked together like the links of a chain: each entry carries a digital fingerprint of the one before. If someone altered an entry afterwards, the fingerprint would no longer match and the chain would break visibly. So it is immediately apparent whether the decision history has been tampered with.

Like a seal that tears at the slightest attempt to open it.

See also: Unchangeable log, GENESIS hash, Integrity status

Unchangeable log

What is an unchangeable log?

An unchangeable log is a record that can only be added to; changing or deleting is ruled out. Every correction decision is held in it like a stamp in a logbook that cannot be retouched away later. So the history stays honest and complete, even weeks on. What went in stays in, exactly as it was.

Like a logbook: entries are added, but never vanish.

See also: Traceable trail, Tamper-proof chain

Traceable trail

What is traceable in the course of a grade?

The trail is the continuous chain of all decisions, with time, acting person, fingerprint and reasoning. For every grade it can be traced who made it, when, and on what grounds. Nothing happens anonymously or in the dark; every step can be attributed. That is what makes a grade defensible to parents and heads alike.

Who, when, why: readable for every grade.

See also: Unchangeable log, Tamper-proof chain

GENESIS hash

What is the GENESIS hash?

The GENESIS hash is the fixed start value of a hash chain per submission: the "previous hash" of the very first entry. If it's missing, the chain isn't cleanly started; its constancy makes the start of every audit chain unambiguously verifiable.

See also: Tamper-proof chain, Integrity status

Integrity status

How does Laura show whether an audit chain is intact?

The integrity status sums up whether a submission's audit chain is valid or where it would be damaged: such as hash break, missing start, orphaned entry, duplicate identifier or quarantine. It's thus machine-detectable whether the data can be trusted.

See also: Tamper-proof chain, GENESIS hash

Offline outbox

What happens to decisions when there's no network?

The offline outbox is a small queue right in the browser: if the network drops, decisions are stored there and sent on automatically the next time there is a connection. So the teacher can keep working on the train or in the basement with no signal, without a single decision getting lost along the way.

Like an outbox that waits until there's a connection again.

See also: Idempotency, Conflict resolution

Idempotency

What prevents duplicate entries when resending?

Idempotency means: if the same decision is sent several times with the same identifier, such as after a connection drop, only one entry results anyway. A unique index prevents duplicates, so retries are safe.

See also: Offline outbox

Word diff (LCS)

How does Laura show what changed between two states?

The word diff uses the longest common subsequence (LCS) algorithm to make added and removed words visible in colour code. In the conflict view, the teacher thus sees at a glance how two correction states differ.

See also: Conflict resolution

A single source of truth

Why do no two places in Laura contradict each other?

Behind Laura sits the principle of a single source of truth: each kind of data is defined once, centrally, and everything else is derived from it. That way no contradictory copies can arise that drift apart over time. What is defined correctly once holds the same everywhere, one place of truth instead of many slightly differing notes.

Mirrored structure

What is a mirrored structure?

A mirrored structure is an order that looks the same in two places: once as a path through the web interface, and once as a filing in the folder structure. Both reflect the same hierarchy, so navigating and filing match one another. You find your way in the operation just as easily as in the storage.

Handwriting learning (writer-adaptive recognition)

How does Laura get better at handwriting she already knows?

The more often Laura sees a child's handwriting, the better she reads it. For each pupil she gradually builds a feel: which letters that child writes in their own way, where you tend to correct. This feel is fed only by your deliberate decisions, never by unconfirmed guesses of her own. Explained once, it sticks.

Ideally, a quirk only ever needs correcting a single time.

See also: Handwriting memory (WriterAtlas), Writer profile, Safeguard against wrong learning, Manual correction (Edit)

Safeguard against wrong learning

What keeps the handwriting knowledge from learning the wrong thing?

So Laura doesn't pick up bad habits, there's a bouncer at the door of learning. It only lets clean signals through: your real, deliberate decisions. A suggestion nobody checked, or two contradictory entries, don't get in. That way what Laura remembers about a handwriting always rests on what you truly confirmed, and doesn't get watered down over time.

See also: Handwriting learning (writer-adaptive recognition), Decision type, Handwriting memory (WriterAtlas)

Handwriting memory (WriterAtlas)

What does the handwriting memory store?

The WriterAtlas is Laura's memory for handwriting: for each child it records which letters they tend to mix up, which corrections keep coming up with them, and how confident Laura already feels about that script. It fills up solely from your deliberate decisions. That's why, ideally, you only have to explain a quirk once, after that Laura recognises it on her own.

See also: Writer profile, Handwriting learning (writer-adaptive recognition), Safeguard against wrong learning

Writer profile

What is a writer profile?

The writer profile is the page in the handwriting memory that belongs to exactly one child: the quirks of their script, from which Laura derives a reliable reading. It grows only through what you deliberately decide, never through silent guesses. Once learned, it is kept cleanly and applied again by itself on the same child's next piece of work.

See also: Handwriting memory (WriterAtlas), Handwriting learning (writer-adaptive recognition)

Careful technical rebuild

Is Laura being rebuilt in the background without the correction suffering?

Yes, and deliberately step by step. The proven correction engine, already tested on real exams, is being moved bit by bit into a cleaner, long-term structure. Until that move is complete, the tried-and-tested version stays the authoritative one. So it is a controlled relocation rather than a risky all-at-once switch.

Moving house while open for business: only move in once the new home really stands.

Runs right in the browser

Do I have to install anything for Laura?

No. Laura runs straight in the web browser: no installation, no update chores, no extra program on the school computer. You simply open the page and get going. That keeps use simple and fast, and it means Laura behaves the same, predictable way on any device, whether the laptop at home or the machine in the staff room.

No download, no admin appointment with IT: open the page and start.

XMP release metadata

How can you tell a PDF was released by the teacher?

Laura's exported PDFs carry an XMP metadata field that marks the teacher's release once finally confirmed. It's thus machine-verifiable that a document went through the full release path and isn't an intermediate state.

See also: No-Surprises PDF

Inline correction overlay

How do corrections appear on the digitised original?

The inline overlay places correction marks directly onto the digitised original, anchored at the position of the word concerned (bounding box). In adjust mode, the teacher can move markers so the correction sits exactly in the right spot.

See also: Correction marks, Tokens (OCR tokens)

Character Error Rate (CER)

What is the Character Error Rate?

The Character Error Rate (CER) measures the share of wrongly recognised characters out of all characters: the lower, the better. It's a standard metric of text recognition, measured per hand-in and aggregated over several hand-ins.

Standard metric; the named targets are internal goals, not a measured live result.

See also: Word Error Rate (WER)

Word Error Rate (WER)

What is the Word Error Rate?

The Word Error Rate (WER) is the counterpart to the CER at word level: the share of wrongly recognised words out of all words. It shows recognition quality where, for correction, whole words ultimately count.

Standard metric; target values are internal goals, not a sourced live measurement.

See also: Character Error Rate (CER)

Green-word precision (GWP)

How reliable are the words marked green?

Green-word precision measures how often a word classed as "green" (certain) was actually accepted after the teacher's check. It's the quality metric for whether the traffic light really is certain at "green".

Target value per product docs; not a measured, published live result.

See also: Confidence traffic light, Confidence class

Bias check

How does Laura ensure fairness on hard-to-read handwriting?

In the bias check, Laura checks before a rollout whether a single handwriting is systematically treated differently from the rest of the class. Conspicuous cases are reviewed manually, so that no child is disadvantaged just because their handwriting is harder to read.

Curriculum mapping

How does Laura map different curricula?

Curriculum mapping translates the KMK educational standards into federal-state-specific rubric templates. The same core (e.g. a text type) can thus be adapted to different states' requirements, instead of rewriting everything for each state.

Implemented as a BASE+OVERLAY+RESEARCH curriculum layer in the engine.

See also: KMK standards, Federalism mapping

Federalism mapping

Why does school software need a state layer?

Education is a matter for the states: 16 federal states have their own curricula, grade scales and school structures. The federalism mapping is the view onto exactly these state differences, technically the same as curriculum mapping.

See also: Curriculum mapping

KMK standards

What are KMK standards?

The KMK standards are the educational standards of the Standing Conference of the Ministers of Education: the cross-state core. In Laura they form the normalisation layer on which the state-specific curriculum mapping builds.

See also: Curriculum mapping

Substitute package

What's in a substitute package?

The substitute package is a handover for a substitute teacher: e.g. seating plan, support needs and key information. Access is sensitivity-controlled: the substitute only sees what they really need for the day, no more.

Access controlled via capability tokens (data minimisation).

Founding Teacher Cohort

What is the Founding Teacher Cohort?

The Founding Teacher Cohort is a small group of first teachers who use Laura as design partners and give feedback regularly. They shape the product's development in close exchange with the team.

Programme description; concrete participant numbers are planned values.

See also: Sounding Call

Sounding Call

What is a sounding call?

A sounding call is a short, regular teacher interview during the early partner phase. It gathers structured feedback from the real correction routine that feeds directly into further development.

See also: Founding Teacher Cohort

Paste-an-Exam (demo sandbox)

How can I try Laura without uploading real data?

With "Paste-an-Exam", a teacher pastes a synthetic or anonymised exam snippet into the demo sandbox and watches Laura correct live. As a hard boundary: no real student text; the demo is deliberately PII-free.

Privacy boundary: synthetic/anonymised inputs only.

School pack

What does a school management get in the school pack?

The school pack bundles the documents a school management needs for the introduction: e.g. data protection impact assessment, DPA template, rubric templates and an emergency runbook. It's the starter folder for the data-protection review.

See also: Data processing agreement (DPA), Data protection impact assessment (DPIA)

Low-barrier mode

Is Laura usable without a mouse and with a visual impairment?

The low-barrier mode aims for full keyboard navigation, meaningful ARIA labels, high contrast and text scaling, so that assessment flows can be operated without a mouse too. Accessibility is checked continuously, not claimed once.

Target criteria (among them high contrast); conformance is checked by an accessibility audit, not guaranteed across the board.

Rationale

When must a decision be justified?

For no-longer-reversible conflict resolutions, Laura requires a short rationale. It records why the teacher decided that way and makes the decision traceable later in the audit trail.

See also: Escalation (escalate), Traceable trail

Teacher Font Studio

Can Laura correct in my own handwriting?

Yes. The Teacher Font Studio captures your handwriting and turns it into your own font for the PDF export. The corrections in the finished feedback PDF then appear in your familiar hand instead of a neutral computer typeface. For children and parents the feedback feels personal, as if you had written in the margin with your red pen as always.

Your handwriting, set digitally: the feedback stays personal.

Correction assistant

Is Laura an AI teacher?

No. Laura sees itself as a correction assistant: a workplace for teacher decisions, not an AI that replaces the teacher. The AI prepares and proposes; the pedagogical decision stays with the teacher.

Product positioning (MASTERPLAN).

See also: Human-in-the-Loop, The teacher decides