The Evidence Standard
A school issues credit, a grade point average, a class rank and a transcript. Each one asserts that the work behind it came from the student it is issued to. This is the same forensic read the firm performs on vendors, turned inward: the school’s own syllabi and grade weight breakdowns, read for how much of each grade actually rests on evidence that the work was the student’s own.
Two documents the school already holds. No student data, no student work, and no finding about any student or any teacher.
Not yet in service
In preview. First engagements begin late January 2027.
The method is settled and documented. What is not yet settled is published fees, and the method has not yet been run against a live set of school documents — which is what the pilot is for. Three pilot slots are open for inquiry now.
A school taking a pilot slot should expect to be a first run: the findings it receives are real, and the instrument may be refined as a result of reading its documents. A school that wants a finished product with published rates should wait for general availability.
Inquiries
Open now
Scope conversation, no commitment, no fee.
Pilot engagements
From late January 2027
Timed to the second semester, when the spring syllabi and weight breakdowns exist.
Fees
Published at launch
Pilot terms are set in writing before any work begins, as with every engagement.
What it reads, and what it asks
A course grade is already built out of pieces with percentages attached. The school built that structure. The read borrows it and asks four questions of every piece.
00
The four questions
The instrument
answerable from two documents
1 · Where the work is produced
Does the syllabus state that the graded work is done in class, outside it, or both? Where the syllabus says nothing, the question scores nothing — and the share of the grade whose conditions the school has never written down is itself reported.
2 · Is the developing work graded
Does the weight breakdown carry a line for an outline, a draft, a revision, a notebook? Collecting drafts that carry no points produces no evidence anyone can use, so this question turns on grade weight, which is visible in a document.
3 · Does the student have to explain it
Is there a graded line for a conference, a defense, a seminar, an oral exam? Same reasoning. A two-minute unscored check-in is not evidence.
4 · Could a stranger have done it
Could someone who never attended the course complete this task? A prompt that names a specific thing the class produced — the October seminar, Tuesday’s lab data, the student’s own annotated copy — is anchored. A prompt about a widely available text is not.
What comes out
Per course and per department: the share of the grade awarded for demonstrating knowledge that rests on no connecting evidence, the share whose conditions the school has not documented, where the exposure is concentrated, which assignments are one gradebook line away from holding up, and which teachers already design assessments that do.
The documents requested
The course list with sections and enrollment; the grade weight breakdown per section; and the syllabus for each section, stating where each graded assignment is completed and when it is due. Nothing else is accepted — not student names, not student work, not grades or scores, not access to the student information system, not detection-tool output.
Scope
Courses graded chiefly on text or symbolic work produced outside class — English, history and social studies, science writing, mathematics, world language written work. In practice grades 6 through 12. Courses graded on performance observed in the room, such as studio art, music, theatre and physical education, are out of scope: they score near the top by their nature and the result carries no information.
↓The first reading becomes the baseline every later reading is measured against.
How the engagement runs
Five phases. A school may stop after any one of them, and the first is the only one the others depend on.
01The Evidence Inventory
Start hereone-time · per engagement
What it does
Codes every graded line in every section in scope against the four questions, from the two documents, and reports the result per course and per department.
What the school receives
A findings report ranking departments by exposure; the assignments carrying the most of it; the teachers whose designs already hold up; and a confidential annex on how academic integrity cases in the period were actually decided.
The annex, and why it is separate
Where integrity findings against students rested on detection-tool output, that is a due-process question with a paper trail, not a teaching question. It goes to the head of school or superintendent and to counsel, before it goes to a board.
↓The findings name the specific documents to fix, in the order they cost the school least.
02The Redesign
With the departmentby department · sessions
What it does
Working sessions on the specific assignments the Inventory flagged, in cost order: first what can be fixed by writing down what already happens, then what a rewritten prompt fixes, then what changes the shape of the grade. The firm brings a worksheet showing what each line scores now, what the documents are missing, and what it would score once fixed.
Who writes what
The department authors the content. The firm supplies the standard and the arithmetic, and reads the result. The firm does not write assignments or advise on what a course should ask.
What the school receives
Revised assignment descriptions and syllabus language authored by its own teachers, a revised weight breakdown, a re-score through the same instrument, a change log, and the declined list — what the department chose not to change, and why, in its own words.
↓The teachers whose designs already held up are the ones the instrument is handed to next.
03Calibration
Optionalcohort · then retainer
What it does
Teaches the school’s own readers to code their own courses, and measures how closely their coding agrees with the reference. A reader is calibrated when agreement reaches the threshold on each question, not on average across them.
What it is not
Not instructional coaching, and not a credential. No classroom is observed and no one’s teaching is assessed. Calibration describes agreement with a reference, which is all it claims.
What the school keeps
Its own readers, and the casebook — every judgment call that arose, with the ruling and the reason. The casebook survives turnover, which the readers may not, and it is what the school actually keeps.
↓What the reading found is written into policy that names its own owner and reopens itself.
04The Policy Layer
Board-adopteddrafted · presented
What it does
Drafts what the findings require: an academic integrity standard specifying what counts as evidence, an assessment validity standard, decision rights over assessment, the grading reconciliation, and one AI-use scale adopted school-wide so students in the same course stop facing different rules for the same conduct.
Every provision carries three things
A named owner, a change trigger stating the event that reopens it, and a review date. A provision without a trigger waits for someone to remember it, which is why most adopted policy is stale within two years.
Who drafts, and where the line sits
The firm drafts board policy, as state school boards associations and policy services do. Counsel drafts and reviews any provision bearing on student discipline procedure, liability, or what the law requires. The firm says what evidence should be required; counsel says how a process runs.
↓A year later the same instrument reads the same courses, and the difference is the report.
05The Annual Read
Keep it currentannual · pre-budget
What it does
Re-reads the current documents and reports the movement — on the courses present in both readings only, with what entered and left reported separately. A figure computed across a changed set is not a trend.
Cadence
Lands 90 to 120 days before the budget or renewal cycle, the same clock the firm’s vendor-side annual read runs on, so a school holding both receives one package in one meeting.
And the policy maintenance
Every provision adopted at Phase 04 is checked against its own change trigger, and the amendments the triggers require are drafted — whether or not the number moved. The movement is the finding; the amendments are the deliverable.
Reported against the school’s own choices
The movement is reported beside the declined list. A department that acted is shown next to one that did not. The report is about what the school chose, not about what the firm achieved.
The pilot
Three slots, running from late January 2027. A pilot is a trade, and both sides of it are stated here.
—Pilot engagement
Three slotsspring 2027 · reserved by inquiry
What a pilot school receives
The Evidence Inventory in full across two departments, the findings report, the confidential annex, the Phase 02 worksheet, and one redesign session with a department. Delivered on a stated schedule, in writing, before any work begins.
What a pilot school supplies
The three documents inside an agreed window, a named point of contact with enough standing to get syllabi out of department heads, and a chair willing to sit the session.
What a pilot school accepts
That this is a first run. The findings are real and the arithmetic is fixed, but a reading of live documents may refine the instrument, and a refinement that changes a figure will be reported as a correction rather than quietly absorbed.
What a pilot school influences
The instrument itself. Where a question proves ambiguous against real syllabi, the written condition is rewritten, and the pilot school’s documents are the reason. Those rulings become the first entries in the casebook.
Pilot terms
Set in writing at scoping, below the rates published at launch. Reserved by inquiry in the order received.
Who the pilot suits
An independent school whose transcript is its productA board that wants an outside answerA district with one or two departments to startNot a school needing published rates now
Aggregate data
A pilot engagement includes permission to use de-identified aggregate figures in the firm’s published work. Nothing identifying the school, a course, a teacher or a student is published, and the school is named only with its written consent. Across enough schools these figures are what allow a reading to be reported against a comparison group rather than against a fixed scale, which is the single largest improvement available to the method.
Why the findings hold up
Every figure traces to a document the school produced itself, and the method states its own limits before anyone asks.
From the school’s documents to a figure it can defend
Capture the two documents → list every graded line → separate what certifies from what is practice → ask the four questions → have a second reader check a sample → report with the limits attached.
Capture
The course list, the grade weight breakdowns, and the syllabi. Each figure carries the capture date of the document it came from.
Register
Every graded line with its share of the grade. Where a category holds assignments with different conditions, it is split; where the documents will not support the split, it is coded at its weakest condition and that is recorded.
Separate
Work graded against a standard is separated from work graded on completion. A validity failure only matters where validity was claimed, and homework never claimed it.
Code
The four questions, each answerable from one named source. Where the documents are silent, the question scores nothing and the silence is reported.
Check
A second reader codes a sample independently. Agreement is reported per question, because agreement averaged across questions measures the base rate rather than the reader.
Report
With the scope rule, the undocumented share, the agreement figures, and the limitations attached — in every report, not on request.
What this work is not
It is not an estimate of how many students used AI. No such estimate is possible from these documents, and any that was would rest on detection-tool output, which this method treats as something to be audited rather than relied upon.
It makes no finding about any student, and none about any teacher. Unbacked work is a property of an assessment’s design. The questions describe graded lines and written tasks, not instruction, and the list of designs that already hold up names practice rather than deficiency.
It measures documents, not practice. A teacher who conferences with every student and never wrote it down scores nothing for it. That is a real limit — and it is also the point: a parent, an appeal or an accreditor cannot be answered with a recollection either.
It is not legal advice, and it is not a compliance certification. Counsel advises on liability, drafts and negotiates terms, and interprets what the law requires. A reading describes a school’s documents as of a date. It clears nothing and warrants nothing.
The goal is not a perfect score. A school that drove every course to full backing would have eliminated independent work, which is the wrong problem solved. No report recommends it.
Related services
The same forensic read, pointed at three different sets of documents.
Can your school show that its grades mean what they say?
Three pilot slots, from late January 2027. A thirty-minute call settles scope and holds the slot. Inquiries also by email at jcs@thelanguagefirm.org.
A first conversation carries no fee and no commitment. Scope, schedule and terms are confirmed in writing before any work begins.