Notes often name a condition without its type, stage or severity, even when the patient's record holds the
measurements that would let a clinician be specific. The side panel sits on the right of the chart, never freezes the note,
shows the record's own evidence first, offers wording you can place in the draft with one click, and has a Dismiss button.
It covers three conditions: heart failure (type by heart pumping fraction), kidney disease (stage by kidney filtering rate)
and malnutrition (severity from a dietitian's assessment). You decide; it never decides for you.
I am about to sign a note: what do I do with the side panel?
Read your draft as usual. The panel never blocks typing, saving or signing.
Look at the cards marked “Needs a look”. Each says, in one sentence, what the record shows and what the note says.
Open “Evidence from the record” to see the values, dates and who recorded them.
Choose: click one wording under “Options” to place it in your draft (highlighted, with Undo), or click
“Other (I will write it)”, “Clinically unable to determine” or “Not clinically supported”, or press Dismiss.
Sign. If something is still open you are asked once; “Sign anyway” is always there.
I am reviewing a signed note: where do I look?
The panel on a signed note shows the check as it was at signing.
Under “Details” you find the rule versions, the published sources and a link to replay the check.
Every click (choice, dismissal, sign anyway) is in the audit trail with the person's name, the time, the options shown and the evidence shown.
1● Needs a lookHeart pumping (ejection fraction)2Draft · pending review
3The note says CHF; the record has an ejection fraction of 30% on 2026-10-02 (earlier 35% on 2025-06-09). The note does not state the type.
4Note read by: AI reader and rule reader agree · AI reader quoted CHF
5 Evidence from the record
Ejection fraction 30% · echocardiogram 2026-10-02: Left ventricular ejection fraction 30% by biplane method.(cardiology)
6 Options
8Dismiss9Published definition and source
Status chip. What to do: “Needs a look”, “Agrees” (nothing to do), “Not assessable” (the data cannot decide), “Could not read: read it yourself”. Always a word and an icon, never colour alone.
Draft · pending review. The definition behind the card has not yet been approved by your clinical and coding reviewers. It turns into “Reviewed definition” once they approve it in the definition library. It is information, not a warning.
The sentence. What the record shows and what your note says. Your note's own words are in quotation marks.
Note read by. Who read the note's words for this card and the exact words they read; if the two readers disagree, the card asks you to read that part yourself.
Evidence. The values with their dates, quoted from the chart, with who recorded them. Links take you to the result, report or note.
Options. Every wording the published definition allows, always in the same order, each with what it needs and what the record shows (met, not met, not assessable). Nothing is pre-selected or recommended. One click places your choice in the draft, highlighted, with Edit and Undo, until you sign.
Other, unable to determine, not clinically supported. Always offered. They record your answer and insert nothing.
Dismiss. One click; a reason is optional. The card stays dismissed for this note and the dismissal is recorded.
Published definition and source. The source's title, publisher, year and a link, the thresholds in plain words, and any edition differences.
Statuses and what to do
Statuses
Status
What it means
What to do
● Needs a look
The record and the note do not line up: the record meets a definition the note does not state, the note is less specific than the record allows, or the note states something the record does not fit.
Read the evidence; choose an option, write your own wording, or dismiss.
✓ Agrees
The note and the record line up.
Nothing to do.
∅ Not assessable
The data cannot decide (for example, no ejection fraction on file, one kidney result only, or too few nutrition characteristics recorded). This is not an all-clear.
Use your judgement; the card says what is missing.
⚠ Could not read: read it yourself
Part of the note could not be read automatically.
Read the highlighted part of the note yourself.
Could not check: the note is unaffected
The panel itself could not run this time.
Keep working and sign as usual; reload the chart to try again.
Relevant history
Type today's reason for visit in the box at the top of the draft (for example “worsening knee pain”). After three letters the
panel lists related earlier notes, results, imaging, problem-list entries and medications from the whole chart, not only the
last visit, newest first, each with the matching sentence quoted and “Matched on: …” so you can see why it was listed. Text matching with a
small dictionary; when an AI reader is configured it also suggests a few related terms you can remove. It is information only: no
status, no options. Every item listed is a real chart item, quoted from its own text, and “Matched on” says when an AI-suggested
term found it.
What the AI does here, and what it does not
It reads. An AI reader (a language model, when one is configured) reads the draft note and the report text, and
proposes what they state, each with a word-for-word quote: in the note, the type or severity wording, a negation or a history-only
mention; in an echocardiogram report without a structured value, the ejection fraction and its sentence; in a dietitian's narrative
assessment, the documented intake, weight change and physical findings. It may also suggest related search terms for relevant history and,
if staff switch it on, give a second opinion that is only logged.
It is checked. Every quote is matched character by character against the text; the number in a quote must equal the value
proposed; a rule reader (fixed word lists and patterns, no AI) reads the same text and must agree. If they disagree, the card says
“Readers disagreed · read this part yourself” or “could not read the report text”, never “not supported”. If the AI reader does not answer
within a few seconds, the card uses the rule reader and says “AI reader did not answer in time”.
It is labelled and recorded. Each card has a “Read by” line, the quote the AI reader relied on, and, under “Who read the
note”, which model and version produced the reading. A value taken from report text is marked “read from the report text … quote verified”.
Every reading, fault and response is in the audit trail and the provenance record.
It never decides. It never decides a finding (versioned rules do, over structured or quote-verified values), never chooses
or ranks wording, never writes to the note, never signs. Everything works with the AI switched off; the page footer says which AI reader is in
use.
What the software never does
It does not diagnose. The heart failure syndrome, its acuity and malnutrition itself are your clinical judgement.
It does not choose for you: the options are neutral, in a fixed order, with nothing pre-selected.
It does not insert anything unless you click it, and it never signs.
It does not assign codes; codes appear only on the coder view, for coding staff.
It does not use albumin or prealbumin as nutrition criteria.
It does not let a language model decide a finding. Findings come only from versioned rules over structured values or values whose quote was verified and whose two readers agreed; it also works with the model switched off.
What a coder sees
The coder worklist ranks visits by open items. The coder view of a visit shows the computed
category beside the clinician's documented statement, the reference code family for coders (the clinician's documented statement
governs; the software does not assign codes), and a query draft: clinical indicators with where each came from,
several clinically supported options, “other” and “unable to determine”, no money or quality-score language, checked by a linter
for leading wording before it can be recorded.
How to check a source
Open “Published definition and source” on any card, or the definition library: every
definition lists its source's title, publisher, year, link, licence status and review state. Thresholds are written in our own words;
no society table is reproduced. Under “Details”, “replay” re-runs the check from its stored inputs to the same result.
What “synthetic data” means here
Every patient, clinician, note and number in this demonstration is invented. Nothing comes from a real record. Counts shown elsewhere are simulated, not measured.
If you are…
A nurse
Your documented symptoms and weights can appear as evidence on a card, with your role and the date.
The left rail shows allergies, latest vitals, active problems and reminders at a glance.
“Relevant history” finds earlier notes and results for today's complaint.
You never need to act on a card; the signing clinician decides.
A medical coder or documentation specialist
Clinician screens never show codes or money language; your view does.
Each choice, dismissal and “sign anyway” is in the audit trail with the evidence shown at the time.
Query drafts are linted for leading wording and always offer “other” and “unable to determine”.
An unspecified entry does not prove the record could have supported a more specific one: read the evidence.
A physician, nurse practitioner or physician assistant
Read the card's sentence and evidence; choose, write your own, or dismiss.
Inserted wording is highlighted with Edit and Undo until you sign.
Acuity (acute, chronic, acute on chronic) is yours: indicators are shown as evidence only.
Signing is never blocked.
A dietitian
The nutrition card appears only when your nutrition assessment is in the record (or when a note states malnutrition without one).
Your documented findings are quoted with your name and date; weight-loss percentages are computed from the weight history.
Two or more characteristics are required; the card shows which are documented and at which level.
Something looks wrong
Tell your team's clinical lead. Include the demo chart number (for example P29), the card's topic, what you expected and what it
showed; a screenshot helps. Every card's evidence and rule version can be replayed, so the team can see exactly what was shown.
Demonstration software on synthetic data. Not medical advice. Definitions are drafts pending clinical and coding review.