Knee symptom demo · guided reasoning

A clearer way to check knee injuries, instability, and red flags.

Describe one knee problem in plain language. The engine keeps what it can justify, asks a short follow-up round, and stops with a shortlist, fallback, or safety escalation.

Open guided check Read the reasoning loop

Registry 53 symptom signals

Observation-level evidence rather than broad symptom buckets.

Question bank 28 follow-up prompts

Only the next useful questions are surfaced each round.

Outcome model Three bounded endings

Shortlist, fallback, or safety escalation.

Current scope Knee-only MVP

ACL tear, meniscal tear, patellofemoral pain syndrome, and knee osteoarthritis.

Reasoning atlas

This atlas shows how a knee complaint becomes evidence, questions, and a result.

Complaint in, evidence mapped, follow-up chosen, result out.

01 Start with the complaint story

Mechanism, swelling, pain location, timing, locking, giving way, fever, or trauma.

02 Convert it into structured evidence

Only mapped symptom keys enter the registry. Ambiguous details stay unresolved.

Knee focus

Evidence is shaped around the knee, not guessed outside the registry.

03 Ask the next useful questions

The selector chooses the smallest set of prompts that best separates the remaining candidates.

04 Stop with a bounded result

Shortlist, fallback, or safety escalation. The system does not pretend certainty it has not earned.

Free text still matters. The first complaint shapes the session instead of forcing a giant checklist upfront.

Safety sits outside ranking. Red-flag patterns can interrupt normal candidate scoring early.

How you can use it

Three knee complaint patterns it handles best.

The strongest sessions usually start with one of these complaint shapes.

Acute twist or sports injury

Useful when the complaint starts with a pivot, awkward landing, or sudden turn and the story includes a pop, fast swelling, instability, or mechanical symptoms.

Twist or change of direction
Pop, swelling, or giving way
Clarifies ligament versus meniscal fit

Front-of-knee pain over time

Helpful when the story is gradual, worsens with stairs or prolonged sitting, and needs separation between common non-traumatic patterns and a too-weak history.

Stairs, sitting, kneeling, squat load
Gradual onset instead of one big event
Captures when evidence still stays too thin

Red-flag screening

Designed to interrupt normal ranking when the story suggests something urgent such as deformity, a hot red knee with fever, or major trauma with severe inability to bear weight.

Deformity or severe trauma
Hot, red, feverish joint
Escalation before false reassurance

What the engine does

How the engine moves from complaint to result.

Engine loop

Parse only what is justified.

Mapped signals become structured evidence. Unclear facts remain unresolved.

Score candidate nodes against the live registry.

Supports, penalties, contradictions, and stage rules all affect the shortlist.

Ask the next useful questions.

The selector chooses follow-up prompts that best separate the remaining candidates.

Stop without over-claiming.

The engine either earns a shortlist, falls back cautiously, or escalates for safety.

Three endings

The only three endings.

Shortlist

Ranked fit scores with the strongest supporting and penalizing features when evidence is strong enough.

Fallback

A deliberate stop when the structured evidence is still too weak or too ambiguous to trust.

Safety escalation

A direct escalation message when red-flag logic overrides the normal ranking loop.

Knee workspace

Continue on the knee page.

The homepage only captures the complaint. The interview, shortlist, and ledger open on the knee page.

Entry handoff

Start a knee check

Open empty page

Dedicated page

What opens on the knee page

/knee/

01 Live interview state

The page opens with the current question batch, session metadata, and the next best prompts.

02 Ranked candidates and bounded result

The shortlist, fallback, or escalation stays visible while the session updates.

03 Inspectable ledger

The append-only session trail remains on the knee page, where the reasoning flow actually runs.

If you land there directly with no complaint yet, the knee page also works as a blank workspace.

Coverage right now

Narrow scope, same reasoning loop.

Condition nodes in this build

ACL tear
Meniscal tear
Patellofemoral pain syndrome
Knee osteoarthritis

Where the design is strict on purpose

Medical logic lives in versioned registries
Unknown evidence is not silently treated as false
Question selection is value-driven, not exhaustive
Safety rules sit outside disease ranking