Alzheimer's BBB delivery.

Correct under agentic load.

BBB curation now exercises the state-integrity path directly. Source and runtime activity can create packets and proposals, but accepted state is still event history. Reviewers can inspect structural integrity and downstream impact before treating a record as core frontier state.

vela integrity projects/alzheimers-bbb-dysfunction-delivery --json
vela impact projects/alzheimers-bbb-dysfunction-delivery vf_ba4f834d6dc506d1 --json

Why this frontier.

Alzheimer's BBB delivery is narrow enough to hold in one head, translational enough to matter clinically, and contradicted enough to exercise every primitive in the substrate — Findings, EvidenceObjects, Corrections, typed Links, and confidence drift across replications.

The amyloid hypothesis dominated funding for two decades while contradicting trial results piled up in registries no downstream model read. Lecanemab and donanemab eventually arrived with modest benefit in early symptomatic disease — narrower than most actors had prepared for. The substrate exists so the next concentration takes weeks to detect and adjust against, not twenty years.

A worked Correction.

One signed event, end-to-end. Halliday 2016 → Nation 2019 → Montagne 2020 → Correction event narrowing scope from sporadic AD to APOE4-positive sporadic AD. Three downstream objects (target hypothesis, Phase II inclusion criterion, review article headline) get notified by the dependency graph.

See it: /proof/pericyte-correction. Companion: /proof/contradiction (two reviewers signing opposing findings under the same scope, neither overwrites the other).

Strongest claims, by confidence.

12 of the 86 publication-grounded findings on this frontier, ordered by confidence. Each card is content-addressed; click to read evidence, supporting trials, and any active Corrections. Full frontier in /workbench.

These 86 findings are extracted from 83 foundational papers — a depth-first cut, not a survey. Expansion of the source corpus is on the year-one roadmap; the frontier's growth surface is the open questions below.

Open questions on the active surface.

39 questions, hypotheses, and gaps surfaced from working notes by the Notes Compiler. These are not signed findings — they have no DOI and no extracted evidence span. They are what the frontier is currently trying to answer. A question becomes a finding when a curator anchors it to a paper and signs the resulting evidence object.

Showing 8 of 39, most recently surfaced first. The full set is in /workbench; source files in projects/bbb-flagship/.vela/findings/ .

• vf_5142fb969…

  P-glycoprotein expression at the blood-brain barrier declines with age and is further reduced in Alzheimer's disease, contributing to amyloid-beta accumulation by impaired cerebral efflux; this is the same transporter that excludes most oncology drugs from glioblastoma at the infiltrative margin.

• vf_155dd090f…

  Claudin-5 and occludin expression is reduced in cerebral microvasculature of early Alzheimer's disease post-mortem brain, paralleling the tight-junction breakdown observed in glioblastoma peritumoral vessels and consistent with a shared mechanism of BBB compromise across these otherwise unrelated diseases.

• vf_61249fb41…

  Pericyte loss in cognitive-impairment-prone hippocampal regions precedes blood-brain barrier breakdown and is detectable before clinical Alzheimer disease symptom onset; this provides a timing-window for biomarker-driven intervention.

• vf_f62560b9a…

  Which public patient-level dataset can jointly inspect amyloid clearance, tau movement, ARIA events, APOE4 status, and cognitive trajectory after anti-amyloid treatment?

• vf_ac19cd84e…

  A ScienceClaw-shaped runtime proposes that the anti-amyloid interpretation should remain scoped to early symptomatic disease with biomarker confirmation and MRI-capable ARIA monitoring.

• vf_8f771d0c7…

  A ScienceClaw-shaped artifact synthesis suggests anti-amyloid translation should be represented as a bounded benefit-risk state update, not as an unscoped efficacy claim.

• vf_b299e3ad7…

  A4 Study (NCT02008357) preclinical solanezumab arm halted 2020 after EXPEDITION3 + multi-cohort futility analysis; preclinical AD as solanezumab indication ruled out alongside mild-AD.

• vf_10d9f97f1…

  DIAN-TU solanezumab arm (dominantly inherited AD prevention) terminated for futility 2020; soluble-Aβ-targeting MAb framework not viable in preclinical AD even in fully penetrant genetic risk.

Open contradictions.

9 pairs where two reviewers' signed findings disagree under the same scope. The substrate stores both as first-class objects; the dependency graph carries the disagreement until new evidence resolves it.

1. Focused ultrasound blood-brain barrier opening may increase regional amyloid removal during aducanumab treatment, but current human evidence is very small and not sufficient for clinical translation.
2. Anti-amyloid programs can produce large amyloid biomarker changes while producing modest, absent, or discordant cognitive effects.
3. Gantenerumab GRADUATE I and II failed to meet primary clinical endpoints despite amyloid-lowering rationale, reinforcing that target engagement alone does not guarantee clinical translation.
4. Solanezumab did not slow cognitive decline in the A4 preclinical Alzheimer's disease trial, weakening soluble amyloid-beta targeting as prevention monotherapy.
5. Atabecestat development was halted after liver toxicity signals, separating BACE1 target engagement from acceptable translational safety.

Browse by target.

Contribute to this frontier.

Three paths, no permission required:

• Mirror the frontier locally. vela registry pull vfr_06cfcbe7c449d86a --out ./bbb.json — replay the canonical state through any of the three reducers (Rust, Python, or TypeScript).
• Propose a finding. Sign a finding.add event with your own Ed25519 key and POST to vela-hub.fly.dev/publish. Proposals land in the inbox; a curator merges them into canonical events.
• Curate the frontier. The frontier needs one credentialed neurovascular scientist (or small group) to sign Corrections, recruit other signers, and arbitrate scope. Email will.blair0708@gmail.com or open a repo discussion. The first external signer on this frontier makes every downstream adoption argument concrete.