Flagship demo
One question. One inspectable frontier.
Start with anti-amyloid translation in Alzheimer's disease. Show how an
external artifact packet becomes reviewed frontier state, then move from
answer to finding, artifact, event trail, and proof packet.
The moment
The artifact does not become truth. It becomes a proposed change.
This is where the Borrowed Light argument becomes product. A
ScienceClaw-shaped export enters as source material. Vela preserves the
packet, opens proposals, accepts one bounded update, rejects an overbroad
claim, and changes the proof only through canonical events.
The viewer should leave understanding one thing: the field can inherit
what changed, who reviewed it, which evidence moved, and what remains
unresolved.
01 Start with the runtime bridge.
ScienceClaw-shaped and discourse-shaped exports enter as source
material. Vela preserves artifacts, creates proposals, accepts bounded
updates, rejects overbroad claims, and records the decisions in the
proof trail.
Open inbox 02 Ask the field-level question.
What is the current state of anti-amyloid translation? The console
answers with bounded trial benefit, biomarker caveats, ARIA/APOE4
risk, failed BACE programs, delivery constraints, and gaps.
Open overview 03 Click the support finding.
Use the lecanemab finding to show the structured claim, conditions,
evidence span, source locator, confidence, annotations, and event
history.
Open finding 04 Follow the artifact.
The support trail points to ClinicalTrials.gov and FDA label records.
The latest adapter run fetched 10 P0 registry records and accepted
artifact events without accepting clinical interpretation.
Open snapshot 05 Show the correction path.
The review trail keeps the claim narrow: early symptomatic,
amyloid-confirmed patients, modest absolute effect, risk-stratified
interpretation. The caveat is state, not copy.
Open event feed 06 Separate benefit from risk.
Move to Trials and Risks. The point is not a prettier summary.
It is a state object that keeps ARIA, APOE4, anticoagulants,
and eligibility scope attached to the answer.
Return to brief 07 End on the packet and gap.
The imports trail shows how runtime output stays source
material until review. Open questions show what would change the
answer: longer-term functional outcomes, subgroup risk, real-world
monitoring, and delivery constraints.
Open questions Three-minute script
- Open the Inbox and name the runtime-to-state boundary.
- Read the clinical-benefit answer and caveat.
- Open the lecanemab finding and point to conditions and evidence.
- Return to the review trail and show proposal and event IDs.
- Open the ingest run, show registry hashes, then end on Proof and the gap.
What this proves
- The answer is backed by signed frontier state.
- Runtime artifacts stay proposed until review creates an event.
- Artifacts and source locators are attached to the decision path.
- Corrections are visible events, not hidden editorial changes.
- The hub serves the same state that the CLI can verify locally.
Verify locally
vela registry pull vfr_5076e7b3ff8e6b0f \
--from https://vela-hub.fly.dev \
--out /tmp/anti-amyloid.json
vela stats /tmp/anti-amyloid.json
vela check /tmp/anti-amyloid.json
vela decision-brief projects/anti-amyloid-translation --json
vela trial-summary projects/anti-amyloid-translation --json
vela runtime-adapter run projects/anti-amyloid-translation scienceclaw-artifact-v1 \
--input projects/anti-amyloid-translation/packets/runtime-adapters/scienceclaw-artifact-export.v1.json \
--actor reviewer:will-blair-bot \
--dry-run \
--json
vela source-adapter run projects/anti-amyloid-translation clinicaltrials-gov-v2 \
--actor reviewer:will-blair-bot \
--priority P0 \
--dry-run \
--json
vela source-adapter run projects/anti-amyloid-translation regulatory-documents-v1 \
--actor reviewer:will-blair-bot \
--priority P0 \
--dry-run \
--json
vela artifacts /tmp/anti-amyloid.json --json
vela proof projects/anti-amyloid-translation --out /tmp/anti-amyloid-proof
./scripts/test-first-user-release-path.sh
This is a research-state demo. It is not medical advice and does not
recommend treatment for any patient.