JavaScript is switched off in this browser. The purchase links still work. The email box below will not respond until scripts are enabled.

Free reference, no email

The 47 Documentation Failure Modes

On this page1 to 6 Fabrication · 7 to 11 Omission · 12 to 16 Attribution and certainty · 17 to 21 Identity, PHI and the persistent record · 22 to 27 Numbers, units, sides and soundalikes · 28 to 32 Structure and template · 33 to 37 Verification and signature · 38 to 42 Workflow and handoff · 43 to 47 Patient-facing language · What it is for

A catalogue of the ways AI-assisted clinical documentation goes wrong: notes, portal replies, discharge instructions, prior authorization letters and handoffs drafted with a general purpose model or an ambient scribe. I collected them from drafts I reviewed on service. It is for US physicians who sign what these tools produce. Each entry names the failure, its mechanism, a check you can run before signing, most of them in under a minute, and the habit that prevents it. Read it once, then keep it beside the draft. Educational material, not clinical or legal advice.

Download the PDF →Read it here

MODE 01 · THE INVENTED EXAM FINDING · NOTE EXCERPT

Draft returned for your reviewdermatology follow-up, rash

O: Skin: [AS EXAMINED] Neuro: cranial nerves intact, reflexes symmetric no reflex hammer was picked up

Catch it: name the maneuver, instrument or position behind each objective line. The rule: delete every exam line you cannot place in the room.

Fabrication

Modes 1 to 6

01The invented exam finding

Looks like: A dermatology follow-up for a rash comes back with a full neurological exam in the objective section, cranial nerves intact and reflexes symmetric, and you never picked up a reflex hammer.
Why: The exam section has the most predictable shape in the note, so each organ system the input does not mention receives its usual normal line.
Catch it: Take each organ system line in the objective section and name the maneuver, instrument or position you used to produce it. A line with no maneuver behind it is deleted before you read the rest.
The rule: Delete every exam line you cannot place in the room.

02The phantom lab

Looks like: An admission note written at three in the morning cites a normal cardiac enzyme in the assessment while the results tab shows the sample collected and not yet run.
Why: Once a diagnosis is named, the model supplies the workup that usually accompanies it, and a plausible value is easier to generate than a blank.
Catch it: Open the results tab and, for each named value in the draft, find the resulted entry with a collection time earlier than the note's timestamp. No entry, no value.
The rule: No number enters the note unless you can point to it in the record.

03The expanded review of systems

Looks like: You dictated "review of systems otherwise negative" in a brief urgent care visit for a sore throat, and the subjective section now lists every organ system by name with a specific denial under each, including questions you never asked.
Why: A summary phrase is expanded into the full structure the model has seen most often, and each expanded line inherits the word "denies."
Catch it: Count the systems listed in the review against the questions in the scribe transcript or your own dictation, then delete every system with no question behind it and restore the one-line summary you actually said.
The rule: The review of systems lists only the systems you asked about, and says so.

04The confabulated medication

Looks like: On the discharge medication section, an inhaled controller appears under "continue" for a patient with asthma in the history who has never been prescribed one, and the outpatient pharmacy fills it.
Why: The model associates conditions with their common treatments and writes the association as a fact about this patient.
Catch it: Place the draft list beside the reconciled list in the chart and tick each draft line against a reconciled line. Any unticked line in the draft is removed before you do anything else.
The rule: The medication list is copied from reconciliation, never regenerated.

05The invented conversation

Looks like: The plan of a hospital day three progress note reads "discussed with nephrology, agree with current management" on a day nephrology had not rounded and no page was placed.
Why: These phrases are the connective tissue of a plan, and the model inserts them wherever a plan of this kind normally contains them.
Catch it: Search the draft for "discussed with," "spoke to," "updated" and "per." For each hit, write the person and the time in the margin. If you cannot, delete the sentence.
The rule: Every documented conversation names a real person, role and time you can recall.

06The fabricated patient quote

Looks like: The chief complaint of an ED note carries a vivid quotation, "it feels like an elephant sitting on my chest," and the patient said their chest hurt.
Why: Quoted speech makes a history read as persuasive, and the model generates speech that sounds like the complaint rather than reproducing what was said.
Catch it: Search the draft for quotation marks. Each quoted phrase must match the transcript or your memory word for word, or the marks come off and the sentence becomes reported speech.
The rule: Quotation marks are reserved for verbatim speech; everything else is reported.

Omission

Modes 7 to 11

07The dropped pertinent negative

Looks like: In the history of present illness for a chest pain visit, your dictated "denies exertional component, denies shortness of breath" is gone, and the note reads as though you never asked.
Why: Summarization keeps positive statements and trims denials, which carry less weight in the model's compression of the encounter.
Catch it: Before generating, write your three most important negatives on the encounter's scratch line. After generating, search the draft for each word.
The rule: Your pertinent negatives are written down before the draft is generated, then checked against it.

08The vanished allergy

Looks like: The patient mentions a reaction to a contrast agent while the scribe is recording the medication history, and the pre-procedure note, the allergy field and the order set all stay silent about it.
Why: Allergies are often stated once, in passing, outside the structured part of the conversation, and a scribe transcribes the flow of the visit rather than its consequences.
Catch it: Search the transcript for "allergic," "reaction" and "can't take," then compare what you find with the allergy field in the chart and the allergies in the draft. All three must agree.
The rule: Any allergy spoken in the room is entered in the structured field by you, not by the draft.

09The silent pending result

Looks like: A Friday afternoon discharge summary says "workup complete" while a blood culture drawn the night before is still incubating, and no clinic knows to look for it.
Why: A model summarizing an encounter treats an ordered test as a done test, because the text it learned from rarely distinguishes the two.
Catch it: Filter the results list for the admission by status. Every item without a final result is copied into the summary with the name of the person who will follow it.
The rule: Every pending result is written into the summary with the person responsible for it.

10The lost condition

Looks like: Discharge instructions turn "return if the fever persists past the weekend" into "return if fever," and the same compression strips the "if the pressure drops overnight, then" line from the evening signout, so the night resident inherits the problem without the plan.
Why: Conditional clauses are compressed away as low value detail, leaving the main verb and a shorter, blunter instruction.
Catch it: Search the plan, the instructions and the signout for "if," "unless," "until" and "when." For each instruction and each active handoff problem that has none, restore the condition you dictated.
The rule: Every instruction and every handoff problem carries its if-then line.

11The shrinking problem list

Looks like: The problem list in today's progress note draft is shorter than the one in yesterday's note, and the problems that went missing are the chronic, stable ones: the hypothyroidism, the treated glaucoma, the remote stroke that still shapes the discharge plan.
Why: A draft built from the day's conversation keeps the problems that were talked about, and a stable problem that nobody mentioned on rounds has nothing in the input to keep it alive.
Catch it: Open yesterday's note beside the draft and count the problems in each. Every problem present yesterday and absent today is either restored or documented as resolved in your own sentence.
The rule: A problem leaves the list only by a sentence you wrote, never by omission from a draft.

Attribution and certainty

Modes 12 to 16

12The hardened hedge

Looks like: Your dictated "possibly consistent with early pneumonia, cannot exclude atelectasis" appears in the assessment as "consistent with pneumonia," and the plan follows the certainty.
Why: A hedge reads to the model as noise in an otherwise fluent sentence, and cleaning the sentence removes the doubt.
Catch it: Search the draft for "consistent with," "shows," "demonstrates" and "confirms." Each one must match the hedge you dictated; where it does not, put your words back.
The rule: The note is never more confident than the clinician who signs it.

13The trajectory nobody measured

Looks like: A progress note's assessment says "renal function improving" and "pain well controlled" when a single creatinine has resulted this admission and no pain score was recorded overnight.
Why: Direction words are the normal grammar of an assessment, so the model attaches one to each problem whether or not a second data point exists.
Catch it: Search the assessment for "improving," "stable," "worsening," "resolved" and "controlled." Each must have two dated values in the chart behind it, or it is rewritten as a single observation.
The rule: No direction word without two measurements.

14The borrowed voice

Looks like: "Patient reports the swelling has gone down" in the subjective section becomes "swelling reduced" in the objective section of the same note, and a report is now a finding.
Why: Shortening a sentence removes the attribution first, because it is grammatically the least essential part of the clause.
Catch it: Read the objective section line by line and, for each statement, name the instrument or exam that produced it. Any line that came from the patient's mouth moves back to the subjective with "reports" attached.
The rule: Subjective content keeps its speaker attached.

15The outside record as verified

Looks like: In a new patient visit, "history of seizure disorder" from a referral letter appears in the past medical history as established, with no source, and it reaches the driving line of the plan.
Why: A referral letter, a prior discharge summary and a confirmed result all arrive as text of the same shape, and the draft reproduces each with the same grammar of fact.
Catch it: For each past diagnosis in the draft, find its source in the chart: a prior note, a result or a problem list entry. A diagnosis with only a letter or the patient's word behind it gets "reported" in front of it.
The rule: History carries its source until you have confirmed it yourself.

16The consultant's suggestion as an order

Looks like: The consult note says "could consider anticoagulation once the bleeding risk is clarified," and the primary team's plan reads "per cardiology, start anticoagulation."
Why: When the model merges a recommendation into a plan, it rewrites it in the plan's imperative grammar, and the consultant's conditional verb does not survive the move.
Catch it: Open the consult note and compare the verb of its recommendation to the verb in your plan. "Consider," "may" and "discuss" do not become "start" or "will" without your own decision written in between.
The rule: A consultant's recommendation is quoted with its own verb, and the decision that follows it is yours by name.

Identity, PHI and the persistent record

Modes 17 to 21

17The wrong patient's paragraph

Looks like: The second patient of the afternoon clinic has a history section that opens with the first patient's knee replacement, because the scribe session was never closed between rooms.
Why: A conversation window or a scribe session that was not cleared carries content forward, and the model blends the two encounters into one coherent note.
Catch it: Read the first two lines of each section for age, sex and chief complaint against the header. On any mismatch, discard the whole draft rather than editing it.
The rule: One patient, one fresh session, every time.

18The identifier in the prompt

Looks like: A discharge summary pasted into a general purpose tool to shorten it still carries the record number, the date of birth and the admission date in its first three lines.
Why: Copy and paste moves the whole block, and the identifiers sit in the header where the eye stopped seeing them months ago.
Catch it: Before pasting, search the text for digits, for capitalised words that are not the start of a sentence, and for the word "at" followed by a place. Read each hit and remove or generalize what points at one person.
The rule: Nothing leaves the record for an unapproved tool until the identifiers are gone and you have looked.

19The diagnosis that became permanent

Looks like: A model-suggested "possible heart failure" from an admission note's differential sits on the problem list at discharge, and three clinic notes later it is in the past medical history without a confirming study.
Why: Problem lists are copied forward by every tool and every template, so a hedge that reaches the list once is repeated as a fact by every note after it.
Catch it: Every new problem list entry must point at the sentence in the note where you made the diagnosis. An entry you cannot trace to your own sentence is removed.
The rule: Nothing enters the problem list that you did not put there in your own words.

20The sex or pronoun swap

Looks like: The header says female, the history says "he," and a template line about a prostate exam has followed the pronoun into the objective section.
Why: Pronouns are inferred from a name or a typical pattern, and once one appears the model stays consistent with its own mistake.
Catch it: Read the first sentence of the history against the header for sex and pronoun. If either is wrong, search the whole draft for the wrong pronoun and read every line it touches.
The rule: Sex and pronouns are confirmed against the chart before anything else is read.

21The third party's history

Looks like: The daughter in the room describes her own thyroid surgery, and the past surgical history of the patient, her mother, now lists a thyroidectomy.
Why: Ambient scribes hear speech, not speakers, and they attribute everything to the encounter's subject unless told otherwise.
Catch it: When the transcript shows more than one speaker, read each item in the past history against the transcript and confirm who said it.
The rule: With a second voice in the room, the past history is read twice.

Numbers, units, sides and soundalikes

Modes 22 to 27

22The soundalike that changed the meaning

Looks like: You dictated "hypokalemia" in the assessment of an overnight admission and the draft says "hyperkalemia," or a medication with a phonetic neighbour appears in the medication section under its neighbour's name, and the plan follows the wrong word.
Why: Speech recognition chooses between soundalikes by frequency, and a prefix or a syllable that differs by one sound is resolved to whichever word appears more often in the text the model learned from.
Catch it: Search the draft for every "hyper" and "hypo" and every drug name, and read each hit against the transcript or your dictation. A soundalike passes the eye; it does not pass the ear.
The rule: Every prefix pair, side and drug name is read against what was said, not against what looks plausible.

23The unit swap

Looks like: A weight entered in kilograms at triage appears in the ED note in pounds with the same digits, and a weight-based calculation two sections later inherits it.
Why: Units are short tokens that the model treats as decoration on a number rather than as part of its meaning.
Catch it: For every number with a unit, read the unit against the source field, then ask whether the magnitude is possible for the patient in front of you.
The rule: A number is not verified until its unit is.

24The computed number

Looks like: The draft states a body mass index in the assessment, a clearance estimate in the medication section and "post-operative day five" in the header, all calculated by the model, and two of them are wrong.
Why: Language models produce numbers that look like answers rather than computing them, and the error is invisible because the format is perfect.
Catch it: Recompute every derived value with the chart's calculator or a calendar and overwrite the draft's figure with the result.
The rule: The model never does the arithmetic; the calculator does.

25The recycled date

Looks like: A surgical clinic note carries last year's operation date into "date of injury," and the follow-up line names a date that has already passed.
Why: Dates in the input are reused as anchors, and the model has no sense of today unless you supply it.
Catch it: Read every date in the draft against a calendar and against the event it marks. The follow-up date must be in the future and the onset date before the visit.
The rule: Every date is typed by you or checked by you; none are inherited.

26The reinterpreted frequency

Looks like: In the discharge medication section, the interval between doses you dictated has been normalized to a more common one, and the sig no longer matches the signed order.
Why: Frequency phrases are spoken fast, abbreviated inconsistently, and normalized by the model into whichever form is most common in its training text.
Catch it: Compare route, frequency and duration of every medication line to the signed order, word by word, with the order open.
The rule: The sig in the note is copied from the order, never dictated a second time.

27The laterality flip

Looks like: A pre-operative note for a right knee arthroscopy says "left" in the exam, the consent draft and the plan, consistently, because one early error was carried faithfully through the document.
Why: Left and right are frequent, symmetrical tokens, and the model's drive for internal consistency propagates a single early error everywhere.
Catch it: Search the draft for "left" and "right" and confirm each instance against the imaging report and the signed consent.
The rule: Laterality is checked in every place it appears, not in the first.

Structure and template

Modes 28 to 32

28The unfilled wildcard

Looks like: The placeholder in the assessment field of a clinic template was replaced by a generic paragraph about the diagnosis, so the gap disappeared without ever being answered.
Why: A wildcard reads to the model as an instruction to produce something, and producing something is always easier than leaving a mark.
Catch it: Search for your template's placeholder characters, then read each field a placeholder once occupied and confirm it contains something specific to this patient: a value, a date or a finding.
The rule: A wildcard is filled by you or deleted by you; the model does neither.

29The cloned note

Looks like: Three consecutive urgent care notes for upper respiratory complaints have identical history, exam and plan paragraphs, down to the order of the sentences.
Why: The tool fills each section with the sentences most often seen in that section, so a common complaint produces the same paragraph for every patient who has it.
Catch it: Open your previous note of the same type beside the draft and count the identical sentences in the assessment and plan. If the count is more than zero, search the draft for this patient's chief complaint in their own words and for today's specific finding; if either is absent, type it in.
The rule: Every note contains at least one sentence that could only be about this patient.

30The section slip

Looks like: A progress note's assessment contains "will repeat imaging in the morning," the history of present illness contains "on exam, tender at the right costovertebral angle," and the plan contains yesterday's exam finding, so the reader who goes to the plan for decisions finds descriptions instead.
Why: The model infers structure from headings but assigns content by resemblance, and a sentence about tomorrow's action resembles an assessment as much as a plan. This is a structural failure; the related error of a patient's report re-voiced as a finding is mode 14.
Catch it: Under the assessment heading, search for "will" and "plan to" and move each hit to the plan. Under the history heading, search for "on exam" and move each hit to the objective section. Under the plan heading, read for any sentence with no verb of action.
The rule: Actions live in the plan, findings in the objective section, and nothing is signed under a heading it does not belong to.

31The template that outranks the encounter

Looks like: The general appearance line of an ED note says "no acute distress, resting comfortably" for a patient who was pacing and holding their flank when you walked in.
Why: Default text in a template is treated as evidence by the model and retained unless the input contradicts it in the same words.
Catch it: Keep a list of your template's default phrases and search the draft for each. Every one that survived unchanged is confirmed against the visit or deleted.
The rule: Template defaults are overwritten or deleted; they are never left because nothing contradicted them.

32The bloated note

Looks like: A two-problem follow-up produces four screens, with the small talk about the patient's holiday transcribed in full and the medication change buried in the ninth paragraph.
Why: A scribe transcribes everything said in full sentences, and a model asked to summarize has no way to tell which of those sentences carried the decision, so it keeps nearly all of them.
Catch it: Find the sentence that states your main decision and move it to the first line of the plan. Then count the paragraphs in the draft against the problems in the assessment; every paragraph that does not attach to a problem or to the decision is deleted.
The rule: The decision is in the first line of the plan, and everything that does not serve it is cut.

Verification and signature

Modes 33 to 37

33The skim signature

Looks like: At the end of a clinic session with twelve unsigned notes, the clinician reads the first line under each heading, recognizes the shapes and signs all twelve in nine minutes.
Why: A fluent draft looks finished, and the eye reads fluency as accuracy. The better the tool, the stronger the pull to skim.
Catch it: Read the assessment, the plan and any patient-facing paragraph aloud, under your breath if you must. Then look away from the screen and state today's decision. If you cannot state it without looking, you have not read the note, and you do not sign it.
The rule: No signature on a sentence you did not read.

34The code status that drifted

Looks like: The code status line of an admission note reads "full code" for a patient whose active order says otherwise, or the goals of care paragraph records that the family agreed to comfort measures when the family asked for time to think.
Why: Code status and goals of care are summarized like any other conversation, and a summary resolves an open discussion into the outcome that most often follows it.
Catch it: Read the code status line against the active order and against the last documented conversation. All three must say the same thing, or the draft is corrected before anything else is checked.
The rule: Code status and goals of care are typed by you from the order and the conversation, never accepted from a draft.

35The unpropagated correction

Looks like: You corrected the wrong side in the exam of a pre-operative note, and "left" is still in the assessment, the plan and the patient instructions on the same screen.
Why: Edits are made to one line, and generated documents repeat facts across sections with no link between the copies.
Catch it: After any correction, search the whole draft for the old value and again for the corrected one, and read every hit.
The rule: Every correction is followed by a search for the thing you corrected.

36The stale draft

Looks like: An afternoon draft says "imaging pending" and you sign it in the evening after the read has posted with a finding the plan does not mention.
Why: Drafts are produced in seconds and signed in hours, and nothing in the workflow flags that the record changed in between.
Catch it: Compare the draft's timestamp to the results and orders tab. Anything posted after the draft was generated is read before you sign.
The rule: A draft older than the newest result is regenerated or edited, never signed as it stands.

37The attestation the draft wrote for you

Looks like: The bottom of a teaching service note reads "I personally examined the patient and was present for the key portions," the total time statement names a duration, and a cross cover note carries the primary team's authorship line, none of it typed by the person signing. And the same note carries no statement that a tool drafted it, though your organization's policy requires one.
Why: Attestation language lives in templates and prior notes, and the draft reproduces it as boilerplate; the disclosure line lives in a policy document that the tool has never seen.
Catch it: Read every sentence beginning with "I" at the foot of the note and verify each against who was actually in the room and for how long: examined, present, total time, author. Then search for your organization's required disclosure phrase and add it if policy requires it.
The rule: An attestation is typed by the person it describes, and the disclosure line is part of the signing checklist, not a memory.

Workflow and handoff

Modes 38 to 42

38The level of service the note did not earn

Looks like: The highest visit level sits on a brief follow-up because the draft added a comprehensive exam and a total time statement, or a critical care time attestation appears in an ED note for a patient who was never critically ill.
Why: Templates and prior notes in the context carry the elements of a high level visit, and the model reproduces them as the normal furniture of a note rather than as claims about this encounter.
Catch it: Read the billing line first, then find the sentence in the note that supports each element the level requires. Any element with no sentence, or a sentence you cannot vouch for, lowers the level.
The rule: The visit level is chosen from what you did, never from what the draft says you did.

39The paraphrased criterion

Looks like: A prior authorization letter argues against a smoothed restatement of the payer's criterion instead of the criterion itself, so the wording in the letter does not match the payer's own policy and the reviewer has to go looking for the criterion you meant.
Why: Models restate what they are given in more fluent language, and a restated criterion is no longer the one the reviewer is scoring against.
Catch it: Open the payer's policy and place their sentence beside the one in the letter. They must be identical, in quotation marks, before your answer.
The rule: The payer's criterion is quoted, never paraphrased, and the answer sits directly under it.

40The letter that outran the chart

Looks like: A medical necessity letter for an imaging study asserts a failed course of conservative treatment and a documented functional limitation that appear nowhere in the clinic notes.
Why: Asked to write a persuasive letter, the model supplies the facts a persuasive letter would contain, whether or not they were in the input.
Catch it: For every factual assertion in the letter, find the note, result or order that supports it and write its date in the margin. Anything without a date is removed or documented first.
The rule: A letter contains nothing that the chart does not already say.

41The copied-forward paragraph

Looks like: An AI-generated assessment from hospital day two is still in the progress note on day six, describing a state that has changed, and each day's draft has preserved it word for word.
Why: Prior notes are fed to the tool as context, and the model reproduces them because they are the strongest signal of what today's note should look like.
Catch it: Open yesterday's note beside today's draft and, for each problem, look for an identical assessment sentence. Every identical sentence is rewritten for today or deleted.
The rule: Yesterday's assessment is input for thinking, never text for today's note.

42The recording that ended early

Looks like: The plan changed in the last two minutes of a clinic visit, at the door, after the ambient recording stopped, and the note carries the earlier plan into the after-visit summary.
Why: Scribes document what they hear, and the conversation at the door, in the hallway or on the follow-up call is outside their window.
Catch it: Before signing, compare the last plan item in the draft with the last thing you and the patient agreed, and check the after-visit summary matches the plan.
The rule: The plan is checked against the last thing decided, not the last thing recorded.

Patient-facing language

Modes 43 to 47

43The reply that promised what nobody did

Looks like: A drafted portal reply tells the patient "your refill has been sent" and "we will call you tomorrow with the results" when no refill was signed and no call has been assigned to anyone.
Why: Replies in the training text end with an action, so the model closes the message with the action that usually follows the request, whether or not anyone has taken it.
Catch it: Search the reply for "has been," "will," "sent" and "scheduled." Each promise must match a signed order or a task assigned to a named person, or it is removed.
The rule: A reply promises only what is already ordered or assigned.

44The simplification that changed the meaning

Looks like: Asked to rewrite discharge instructions for a lower reading level, the tool turns "hold the medication until your visit" into "stop the medication," and a pause becomes a discontinuation.
Why: Simplification replaces precise clinical verbs with common ones, and common verbs carry different meanings.
Catch it: Place the simplified version beside the original and, for each sentence, compare the verb. Hold, stop, continue, reduce and skip are not synonyms; any verb that changed is changed back.
The rule: Every simplified instruction is checked against its original for the action it tells the patient to take.

45The reply to a different question

Looks like: A patient asks through the portal whether they can take a second medication alongside the first, and the drafted reply explains what the first medication is for.
Why: Models answer the question they are most confident about, and a general explanation is safer to generate than a specific answer.
Catch it: Reread the patient's message after reading the draft, underline the question, and check that the first sentence of the reply answers that question and no other.
The rule: The first line of every portal reply answers the question the patient actually asked.

46The unauthorized reassurance

Looks like: A portal reply about a new symptom ends with "this is nothing to worry about" and "you will be fine," and you had made no such judgment when you read the message.
Why: The tool has learned that a medical reply usually ends by comforting the reader, so it supplies comfort whether or not you formed the judgment behind it.
Catch it: Search the draft for "nothing to worry," "very common," "normal," "fine" and "no need." Each one is deleted unless you would say it to the patient's face knowing what you know.
The rule: Reassurance is a clinical judgment; the model does not make it, and it is deleted wherever it was not yours.

47The reversed instruction

Looks like: The after-visit summary says "take this on an empty stomach" where the plan said "do not," and the sentence reads as naturally as the original.
Why: Negation is a single short token, and dropping or inserting one produces a sentence that is just as fluent as the original.
Catch it: Search the instructions for "not," "never," "avoid" and "do not," read each hit against the plan, then read the instructions that contain none of them and ask whether one should.
The rule: Every negated instruction is read twice, once for the words and once for the direction.

What this catalogue is for

Read it once, then keep it where you sign. Every entry ends in a check that fits before a signature, and the catalogue is the list the Clinician Suite was stress tested against, so you can audit that claim yourself instead of taking it on faith.

The six protocols, free The Clinician Suite, $99

Educational material for qualified clinicians. Not a medical device, not clinical decision support, not medical advice. Nothing here makes an AI tool compliant with any data protection regime. Edition 1, September 2026. Written by a practicing clinician of thirteen years.