Synapse — Exam & Diagnostics tabs

Three design options each, built to react to. Rendered in the real Synapse panel chrome at its actual 380px dock width, using the live sevaro-* palette.

Read this first — what I'm missing

I don't have what you actually laid out with Smriti, Nia, and Pradji. Your note said you discussed the two tabs, not what you landed on — and there's no transcript or meeting doc anywhere in the repos or context. So none of this reflects the meeting. It's built from the OPS prior art plus acute-stroke workflow reasoning. Treat it as strawmen to knock down, and tell me where the meeting went somewhere different.

Two findings that shaped every option below

1. Both tabs are empty today. Synapse's Exam tab renders the string "Exam tab — coming soon." There is no Diagnostics tab at all — just a decorative Imaging | Labs row on the queue screen with no click handlers. This is greenfield, not a refactor, which means we can pick the right shape rather than inherit one.

2. OPS is outpatient. Synapse is a stroke code. This is the load-bearing distinction. OPS optimizes for "document a normal exam in one click" across a 20-minute outpatient visit. Synapse is a physician on live video with a 4.5-hour tPA clock running. Synapse's data model already carries lkw, edArrival, treatmentWindowMinutes, and a timeline[] — the clock is a first-class concept here in a way it never was in OPS. Porting the OPS tabs over verbatim would be the obvious move and I think it would be the wrong one.


Tab 1

Exam

The core question: in a tele-stroke encounter, is the exam a form to fill out, or a procedure to perform? Each option answers that differently.

OPTION A
Guided NIHSS Driver
The exam is the NIHSS. One item at a time, big targets, auto-advance, running score always visible. Scripts the command for the bedside presenter.
✍️Scribe
🩺Exam
📊Scales
📋Evidence
💬Phrases
NIHSS running
12
MODERATE
Item 7 of 11 · Limb Ataxia
Finger–nose–finger, heel–shin, both sides
"Ask them to touch their nose, then my finger. Now the other hand."
0ABSENT
1ONE LIMB
2TWO LIMBS
UNN/T
6b · Right Leg
2
Falls to bed before 5 sec, some effort vs gravity
6a · Left Leg
0
Linear, one item at a time, score always on screen
+Mirrors what the physician is physically doing. No translation step.
+The command script doubles as coaching for an inexperienced bedside presenter — real value on a tele link.
+Running score + severity band makes the tPA conversation immediate.
−Rigid. Physicians who jump around or re-check an item will fight the sequence.
−Only fits stroke. Seizure/AMS/EEG consults get nothing.
OPTION B
Structured Grid (OPS port)
Lift the OPS pattern: accordions per system, normal-defaulted checkbox grid, one-click template fill. NIHSS demoted to a scale chip.
✍️Scribe
🩺Exam
📊Scales
📋Evidence
💬Phrases
⚡ Fill Stroke Exam Template
StrokeSeizure AMSHeadache
Mental Status▾
Awake
Oriented ×3
Follows cmds
Attentive
Cranial Nerves▾
PERRL
EOMI
Face symm.
Fields full
Motor▾
5/5 all ext.
R drift
Tone normal
No asterixis
Sensation▸
Coordination▸
Gait▸
NIHSS
12 · IN PROGRESS
Proven pattern — but built for a 20-min outpatient visit
+Already built and battle-tested in OPS. Fastest to ship, real code to lift.
+Covers every consult type, not just stroke.
+Status dots give an at-a-glance completeness read.
−Normal-defaults is an outpatient optimization. In a stroke code the exam is abnormal — you're documenting deficits, not asserting normals.
−2-col checkbox grid at 380px is cramped; OPS had a full page.
−Clock-blind. Burying NIHSS as a chip inverts what matters.
OPTION CRecommended
Listen-First, Confirm-Fast
The scribe is already listening. It extracts NIHSS items live with confidence, and the tab becomes a gap list — you only touch what's missing or unsure.
✍️Scribe
🩺Exam
📊Scales
📋Evidence
💬Phrases
NIHSS provisional
12
MODERATE
9 of 11 captured
2 items still needed Ataxia and Dysarthria weren't heard in the exam. Tap to score — or say it out loud.
Needed · Item 7 Limb Ataxia
"Touch your nose, then my finger."
0
1
2
UN
✨ Captured from exam
1a LOC — Alert
HIGH0
4 Facial Palsy — R droop
HIGH2
5b R Arm — drifts
MED2
9 Language — mild aphasia
HIGH1
Tap any row to hear the source audio
🔒 Attest & lock score
2 LEFT
Physician confirmation required before the score enters the note
Zero-friction when it works, explicit about what it doesn't know
+Hands stay free. During a stroke code you're talking and watching, not typing.
+Builds on Scribe — the one tab that actually exists.
+The gap list is the honest UI: it says what it doesn't know rather than presenting a confident-looking form.
+Tap-to-hear-source makes verification cheap enough to actually do.
−The hard one: a NIHSS drives tPA. An AI-suggested score carries real medico-legal weight — hence the explicit attest-and-lock gate, and why "provisional" is on screen until you sign.
−Needs an extraction eval before it goes near a patient.
My read on Exam

C, with A's driver embedded inside it — they're not really rivals. C is A with the items you already said out loud pre-filled. Ship A's guided driver first (it's the honest, deterministic core and needs no AI), then layer C's extraction on top as a flag-gated pre-fill. That sequencing means the exam tab is useful on day one and never depends on the model being right.

B I'd argue against as the primary — but its template chips should survive into whatever we pick, because non-stroke consults (Seizure, AMS, EEG) still need somewhere to go and the NIHSS driver has nothing for them.


Tab 2

Diagnostics

The core question: is this a filing cabinet or a decision surface? In an outpatient visit you browse results. In a stroke code you're asking exactly two questions — can I give tPA, and is this a thrombectomy. Every result matters only insofar as it answers one of them.

OPTION A
Time Rail
Everything anchored to last-known-well. Diagnostics as a sequence of events against the clock, not a list of documents.
✍️Scribe
🩺Exam
🔬Dx
📋Evidence
💬Phrases
tPA window
LKW 13:45 · now 15:52
2:23
Last known well
13:45
ED arrival
14:10 · +25
Labs drawn
14:18 · +33
Glucose 112 · INR 1.0 · Plt 244k
CT head non-con
14:32 · +47
No hemorrhage · ASPECTS 8
CTA head & neck
14:51 · +66
L M1 occlusion
Neuro consult
15:40 · +115
You are here
CT perfusion
pending
⏱ Door-to-needle if given now
102 min
Target < 60 min · this encounter is over target
The clock is the organizing principle, not the modality
+Sequence is the clinical question in acute stroke — what came back, when, what's still out.
+Maps onto the existing timeline[] in the data model. Almost free.
+Doubles as the quality metric surface (door-to-needle) the ops side already tracks.
−Weak for browsing detail — where does a full radiology impression live?
−Unfamiliar. Nobody's EHR looks like this.
−Useless for the non-acute consults (Rounding, Outpatient, EEG).
OPTION BRecommended
Decision Gates
Organize by the decision the result unblocks, not by modality. Each gate shows its criteria and what's outstanding. It surfaces — it does not decide.
✍️Scribe
🩺Exam
🔬Dx
📋Evidence
💬Phrases
tPA window
LKW 13:45
2:23
✓ Thrombolysis criteria
ALL CLEAR
No hemorrhage on CT
14:32
Glucose
112
INR
1.0
Platelets
244k
BP
162/88
Within 4.5h of LKW
2:07
◐ Thrombectomy criteria
1 PENDING
LVO on CTA
L M1
ASPECTS
8
NIHSS ≥ 6
12
CT perfusion
PENDING
Pre-stroke mRS
0
⚠ Needs your attention
1
Anticoagulant use unconfirmed
Chart lists apixaban. Last dose not documented — ask.
Criteria checklist · decision support only · not an eligibility determination
Data reorganized around the two questions you're actually asking
+Matches how the decision is actually made. Nobody browses imaging in a stroke code — they're checking a mental list, and this externalizes it.
+The "needs attention" gate is the real win: it surfaces the missing thing (unconfirmed anticoag) that a modality-organized tab structurally cannot show.
+Every criterion stays traceable to its source result and timestamp.
−Automation bias is the genuine risk. A green "ALL CLEAR" invites not looking. Hence: it renders criteria, never a verdict — no "patient is eligible" string anywhere.
−Criteria are institution-specific and drift with guidelines. Needs an owned, versioned rule set — that's a real maintenance commitment, not a one-time build.
−Stroke-shaped. Other consult types need a different gate set or a fallback.
OPTION C
Modality Cards (OPS port)
The shipped OPS ImagingResultsTab: a card per study, impression badge, expand for findings, prior study nested inline for comparison.
✍️Scribe
🩺Exam
🔬Dx
📋Evidence
💬Phrases
Imaging
🧠 CT Head non-contrast
NO ACUTE
14:32 · ASPECTS 8
No acute intracranial hemorrhage. Early ischemic change in the left MCA territory involving the insula and M2. ASPECTS 8.
Prior · 2024-11-03
Chronic small-vessel ischemic change. No acute…
🩸 CTA head & neck
ABNORMAL
14:51 · L M1 occlusion
💧 CT perfusion
PENDING
Ordered 15:38
Labs
🧪 Coags
NORMAL
INR 1.0 · PTT 28 · Plt 244k
🧪 BMP
NORMAL
Glu 112 · Cr 0.9 · Na 139
+ Add study
Familiar, complete, modality-organized — the safe choice
+Already shipped in OPS. Lowest risk, real code, known to work.
+Familiar to any clinician — it's how every EHR does it.
+Prior-study-inline is genuinely good and worth keeping regardless of which option wins.
+Works for every consult type, acute or not.
−Optimized for a problem Synapse doesn't have. OPS needed 8 collapsible study types; a stroke code has three studies and two labs.
−Makes you do the synthesis. All the data's there; the question "can I treat" is still entirely in your head.
−Clock-blind. A timestamp is just a label here, not a constraint.
My read on Diagnostics

B as the default view, A's clock as a persistent header, C as the detail layer underneath. These stack rather than compete: the gate tells you where you stand, the rail tells you how long you have, and tapping any criterion drops you into C's study card for the full impression. One tab, three depths.

The thing I'd flag hardest for the group: B is the highest-value and highest-risk option in this document. A criteria checklist that quietly goes stale against guidelines is worse than no checklist. If we build it, someone owns that rule set with a version and a review cadence — that's a standing commitment, and it should be a deliberate yes, not a side effect of liking the mockup.


Back to you

What I need to formalize this

All at once, so you can answer in one pass — and the first one matters more than the rest combined.

1. What did the meeting actually land on?

This whole document is my inference. If you and Smriti/Nia/Pradji converged on something, that overrides all of it — paste the notes or just talk at me and I'll rebuild.

2. Is Synapse stroke-only, or all five queues?

The top bar has Acute Care / Rounding / EEG / Outpatient / Rescue. Nearly every design call above flips depending on whether these tabs serve a stroke code or all five. If it's all five, the clock- and gate-centric options need a graceful non-acute fallback.

3. Does the 380px dock stay fixed?

It's the binding constraint on every mockup here — it's why B's checkbox grid feels cramped. If the panel can widen or pop out during the exam, option B gets substantially more viable and I'd redraw it.

4. Where do the diagnostics actually come from?

Everything above is mock data. Gates and time rails are only honest if results arrive structured with timestamps. If it's a human pasting a radiology impression, C is the only option that survives contact with reality.

5. How much AI in the exam is too much?

Option C recommends AI-extracted NIHSS behind an explicit attest-and-lock gate. That's a clinical-risk call more than a design one, and it's the one place here I don't want to auto-adopt my own recommendation. If the answer is "not yet," option A stands alone fine and we lose nothing structural.