MARMAR Context: three fictional patients for your demo

All names, histories, prescriptions, results, family relationships and confirmation statements in this package are invented for software demonstration. They are not patient records, study data, treatment recommendations or evidence of clinical accuracy.

Use Mariam first. She is the main video story: three ordinary records contain information that becomes more useful when reviewed together. Eric demonstrates restraint; Asha demonstrates human review and clarification.

The examples intentionally use October 2026 dates because this version of the demo fixes its assessment date to 23 October 2026. Do not change the dates casually: lab recency and expected findings can change. All notes are English; names and settings do not imply multilingual validation.

These short personal backgrounds are for narration. The upload files contain the clinical source material the current app can process.

Recording setup: one patient in one step

1. Open https://context.marmar.life/ and sign into the reviewer preview.
2. Click New patient (top right of the Check page, or on the Patients page). Type the patient's name.
3. Click Choose .txt files and select the three files in that patient's folder, for example all of 01-mariam-showcase/*.txt. Each file becomes its own record, titled from its file name ("Clinic prescription", "External laboratory", ...), so every fact points back to the record it came from. Tick the synthetic-records box and click Create patient.
4. In the bottom bar, set Reader to MARMAR-Llama-3.3-70B-Instruct. Tick Fresh model call to show an actual model call, then click Read records.
5. On Review, check the facts against their highlighted source lines. Include all supported is a convenience after reading them, not a substitute. Read Lines not turned into facts and tick the acknowledgement. The bottom bar says in plain words what still blocks the check.
6. Click Run medication check. For Asha, resolve the conflicting records as described below first.

Single-file alternative: in step 3 choose only that patient's *-complete.txt. The facts are the same, but they all cite one record instead of three.

Stronger video: reveal records gradually

Create the patient with only 01-...txt, then add files 02 and 03 one at a time with Add record under the record list, so the audience sees the context arrive. Do not also upload the complete version; that would duplicate the evidence.

Suggested 90-second main story: Mariam

• 0–15 seconds: “This is Mariam, a fictional patient with medicines prescribed across different points of care. A prescription list alone leaves important context out.” Show her clinic note.
• 15–30 seconds: Upload the separate laboratory result and previous care history. “Her kidney result and allergy history are recorded elsewhere.”
• 30–50 seconds: Click Read records with the model. Show the medicine list, eGFR and Penicillin allergy, then click a fact's source link. “Every extracted fact stays linked to the words it came from.”
• 50–70 seconds: Review the facts, acknowledge unmapped notes, then assess. Show the medications-only versus contextual comparison.
• 70–90 seconds: “The reviewed kidney result and allergy record add findings for clinician review. MARMAR brings the evidence together; the clinician makes the treatment decision.” Export the report.

You can also assess the clinic note first and then add the other records. New source material correctly invalidates the old assessment and requires extraction/review again. Do not describe the baseline as “safe”: it has incomplete context.

1. Mariam: recommended main showcase

Files: 01-mariam-showcase-complete.txt, or the three records in 01-mariam-showcase/.

Expected extracted context:

• Metformin 500 mg, one tablet twice daily.
• Amlodipine 5 mg, one tablet daily.
• Amoxicillin 500 mg, one capsule three times daily.
• eGFR 38 mL/min/1.73m2, dated 12 October 2026.
• Penicillin allergy from the previous-care note.
• Diabetes and hypertension as recorded history. Condition-specific checking is outside this prototype's supported coverage.

Observed with the separate records: medications-only had 0 findings and reviewed context had 2. The contextual result includes lab-metformin-egfr-45 and allergy-penicillin. The desired story is two context-dependent findings, not a claim that the AI diagnosed her or proved an adverse event was prevented. Hives and caregiver details can remain unmapped; explain that the prototype does not assess every piece of clinical narrative.

2. Eric: show that more alerts is not always better

Files: 02-eric-family-history-complete.txt, or the three records in 02-eric-family-history/.

Expected behavior:

• Eric's medication is Amoxicillin; his own eGFR is 96.
• His own report is no known drug allergies.
• Penicillin allergy, Warfarin and eGFR 38 belong to his mother. They must not become usable patient facts.
• No Penicillin-allergy rule should be triggered for Eric from his mother's history.

Narration: “Shared notes can mix different people. MARMAR keeps the patient's evidence separate and leaves unrelated material visible for review.”

A zero-finding result means no finding from the supported rules and reviewed data. It does not certify the whole prescription as safe.

3. Asha: reconcile and clarify rather than guess

Files: 03-asha-reconciliation-complete.txt, or the three records in 03-asha-reconciliation/.

The source notes deliberately disagree: clinic Metformin 500 mg versus hospital Metformin 1000 mg. Amlodipine is written as 2,5 mg; do not silently turn it into 25 mg or infer its strength.

First show the conflict. Include the supported facts and acknowledge the original notes; the medication check should block until the two Metformin strengths are reconciled.

Fictional resolution established by this scenario's author: a follow-up with the fictional dispensing pharmacist confirms that the hospital Metformin 1000 mg entry is the current prescription, the clinic 500 mg entry is outdated, and Amlodipine is 2.5 mg, one tablet daily. This is an authored scenario fact, not real pharmacist approval or a recommendation to change a dose.

Then demonstrate:

1. On the withheld Amlodipine line, click Add clarification.
2. Paste this confirmed statement:

Medication: Amlodipine; strength: 2.5 mg; form: tablet; dose: 1 tablet; frequency: 1/day; origin: prescribed

1. Enter Fictional dispensing-pharmacist follow-up in the supplied demo scenario as the confirmation basis. Check the synthetic verification box and add the clarification record.
2. Click Read records again. The original ambiguous note stays intact alongside the clarification.
3. Review all facts. Exclude only the outdated Metformin 500 mg fact from the clinic note. Include the current 1000 mg fact, the clarified Amlodipine 2.5 mg, the eGFR 38 result and other supported patient facts.
4. Acknowledge the original notes and run the check. The renal rule remains relevant for clinician review. The included clarification moves into Reviewed clarifications.

Narration: “When records disagree, MARMAR stops for a person to resolve the evidence. The clarification is recorded without rewriting the original note.”

Recording notes

• Keep “Synthetic patient” visible; do not present these as real beneficiaries or clinical validation.
• Rehearse once before filming. Model wording and latency can vary. Open How these facts were read on the Review step to distinguish actual provider calls from locally parsed fields or cached results.
• Save/export each finished report before switching patient if you want to keep all three. Automatic persistence protects the current workspace in this browser; this is not a patient library or cross-device sync.
• Do not upload the validation JSON through source import; it is a test receipt, not a patient note.