Trace one record through the loop

Each curation candidate enters Vela as source material, not truth. For this record, the source is a registered Phase II trial readout. The frontier records its locator, scope, and provenance separately from the finding's assertion.

source id

vs_389ee6e7b8c090d0

title

TRAILBLAZER-ALZ Phase II (NCT03367403, Eli Lilly, Phase II): Established proof-of-concept for donanemab and introduced tau-based patient stratification.

locator

doi:10.1056/NEJMoa2105073

doi

10.1056/NEJMoa2105073

A mechanical curation pass spotted that the evidence atom attached to this finding had no locator, while the parent source already carried a stable identifier. The agent emitted one proposal of kind evidence_atom.locator_repair, asking to copy the identifier down to the atom.

proposal id

vpr_05116687194cef48

kind

evidence_atom.locator_repair

actor

agent:vela-curation-bot-2026-05-08

created

2026-05-08T13:20:28.519685+00:00

reason

Mechanical evidence-atom locator repair derived from the parent source record. The source's locator is canonical; this proposal copies the value down so the atom carries an inspectable pointer.

{
  "id": "vea_0f0583c879503cf2",
  "source_id": "vs_389ee6e7b8c090d0",
  "finding_id": "vf_9a4c5010e77212f3",
  "locator": null,
  "caveats": [
    "missing evidence locator"
  ]
}

{
  "id": "vea_0f0583c879503cf2",
  "source_id": "vs_389ee6e7b8c090d0",
  "finding_id": "vf_9a4c5010e77212f3",
  "locator": "doi:10.1056/NEJMoa2105073",
  "caveats": []
}

The repair touches one field on one evidence atom. It does not change the finding's assertion, confidence, or human-review state. That is the substrate property the doctrine guards.

Mechanical proposals (locator-only, identifier prefix doi:, pmid:, nct:, pmc:, openalex:) auto-accept. Anything that touches assertion text, confidence, evidence-span text, or human-review state stays queued for a human reviewer.

status

applied

reviewer

agent:vela-curation-bot-2026-05-08

reviewed at

2026-05-08T13:20:28.541040+00:00

The accepted proposal lands as one canonical event in .vela/events/. The event hash chains from the atom's pre-repair canonical bytes to its post-repair canonical bytes, so a chain validator can confirm exactly the named atom changed and exactly the named repair was applied.

event id

vev_4a0f740b34440e9c

kind

evidence_atom.locator_repaired

target

evidence_atom · vea_0f0583c87950…

actor

agent:vela-curation-bot-2026-05-08

timestamp

2026-05-08T13:20:28.541058+00:00

before_hash

sha256:…3f6201cd

after_hash

sha256:…dcca19b9

After acceptance, vela frontier materialize regenerates frontier.json, vela.lock, and the proof artifacts under proof/. The proof packet records the post-repair event log and snapshot hashes; a fresh clone can replay the chain and arrive at the same materialized state.

./target/release/vela proof verify projects/alzheimers-bbb-dysfunction-delivery --json

The proof's event_log_hash commits to all 545 accepted events on the BBB frontier, including the 200 locator repairs landed in this curation wave.

The finding now carries a stable source pointer. A reader running vela impact against this finding sees the downstream dependents declared by typed links: which other findings cite, depend on, or contradict this one. Impact reports stay read-only and do not propagate confidence automatically.

./target/release/vela impact projects/alzheimers-bbb-dysfunction-delivery vf_9a4c5010e77212f3 --json

If this finding's confidence weakens later, the same impact traversal tells a reviewer which downstream records to audit. Vela records dependency-aware state. It does not settle Alzheimer biology, recommend treatment, or convert agent confidence to truth.

Impact reports on the BBB frontier

Twenty-five typed dependency links have been declared across the BBB accepted-core findings (Block A4). Eight impact reports are committed under projects/alzheimers-bbb-dysfunction-delivery/exports/impact/; each shows what supports, depends on, or contradicts a single accepted-core finding through the typed-link graph.

• vf_4b7f1b77d9783c58 1 direct · 1 total downstream · 2 events · supports

  Safety profiles of lecanemab: A systematic review and meta-analysis of randomized controlled trials and real-world evidence. — BackgroundSafety profiles of leca

• vf_8108c9642038749a 2 direct · 7 total downstream · 2 events · supports

  Interactive effects of blood-brain barrier breakdown and Alzheimer's disease biomarker status on cognitive decline in older adults without dementia. — Introduct

• vf_9bf64b2a67b97f51 3 direct · 3 total downstream · 2 events · supports

  Focused ultrasound and Alzheimer's disease A systematic review.

• vf_a0f9919e6d163b8d 1 direct · 2 total downstream · 2 events · supports

  Reducing anti-TfR antibody affinity for transferrin receptor enhances receptor-mediated transcytosis and brain uptake; bispecific anti-TfR/BACE1 antibody lowers

• vf_a9e324457a7a027a 2 direct · 8 total downstream · 2 events · supports

  Efficacy and <i>APOE</i> ε4-stratified risk of donanemab in Alzheimer's disease: A systematic review and meta-analysis of randomized clinical trials

• vf_ba4f834d6dc506d1 2 direct · 2 total downstream · 3 events · supports

  APOE4 carriers show accelerated, BBB-specific dysfunction predicting cognitive decline; the broader pericyte-loss/BBB-breakdown narrative is concentrated in APO

• vf_c8730c33b092abbe 1 direct · 4 total downstream · 2 events · supports

  The glymphatic system clears amyloid beta and tau from brain to plasma in humans. — Dysfunction of the glymphatic system has been proposed as a mechanistic link

• vf_dd4a7ec9ef34cf1c 2 direct · 4 total downstream · 2 events · supports

  DCE-MRI in a large aging cohort demonstrates that hippocampal and parahippocampal BBB permeability is elevated in cognitively impaired individuals and correlate

Reproduce locally with ./scripts/build-bbb-impact-examples.sh.

Impact reports on the anti-amyloid frontier

The anti-amyloid translation frontier (vfr_5076e7b3ff8e6b0f) carries reviewer-declared supports, depends, contradicts, and replicates links between accepted findings. Eight more impact reports are committed under projects/anti-amyloid-translation/exports/impact/.

• vf_03875dbe6be66b00 2 direct · 2 total downstream · 5 events · supports

  Verubecestat failed to improve cognition or function in mild-to-moderate Alzheimer's disease despite BACE1 target engagement.

• vf_0767356cc73a6387 0 direct · 0 total downstream · 2 events

  Lanabecestat phase 3 trials were stopped for futility, adding class-level pressure against broad BACE1 inhibition as an Alzheimer's disease treatment strategy.

• vf_29671540d388fc5d 0 direct · 0 total downstream · 5 events

  The minimum clinically important difference for CDR-SB in early Alzheimer's disease remains a decision-critical threshold for interpreting anti-amyloid trial ef

• vf_3d75299a343a2fc8 0 direct · 0 total downstream · 2 events

  Atabecestat development was halted after liver toxicity signals, separating BACE1 target engagement from acceptable translational safety.

• vf_6aff53d375659bcf 1 direct · 1 total downstream · 4 events · depends

  Anti-amyloid programs can produce large amyloid biomarker changes while producing modest, absent, or discordant cognitive effects.

• vf_7b4bfdf919b22af5 3 direct · 3 total downstream · 4 events · contradicts, depends

  Donanemab slowed clinical decline in early symptomatic Alzheimer's disease in TRAILBLAZER-ALZ 2, with efficacy interpreted against amyloid and tau-selection cri

• vf_9ded3333a52f3e58 5 direct · 11 total downstream · 4 events · contradicts, depends, supports

  Lecanemab slowed decline on CDR-SB in early Alzheimer's disease in CLARITY AD, with a modest absolute effect that requires patient-level risk stratification.

• vf_afddbeabc19efa09 3 direct · 6 total downstream · 3 events · contradicts, depends, supports

  Lecanemab reduced amyloid burden and shifted downstream biomarkers in CLARITY AD, but biomarker movement remained larger than the observed clinical effect.

Reproduce with ./scripts/build-anti-amyloid-impact-examples.sh.

The same seven steps drive a different frontier. The early-AD biomarker calibration frontier (vfr_a22c9022674a2304) carries eight reviewer-grade draft findings calibrated against the anti-amyloid translation MCID anchor. One of them, the donanemab CDR-SB effect from TRAILBLAZER-ALZ 2, runs the same loop end to end. Nothing here is a clinical recommendation; it is a record of state that a reviewer can audit.

step 1 · packet

source id

vs_2c5a3d3bd78a01d4

title

Donanemab in Early Symptomatic Alzheimer Disease: TRAILBLAZER-ALZ 2 (Sims et al., JAMA 2023)

locator

doi:10.1001/jama.2023.13239

doi

10.1001/jama.2023.13239

step 2 · proposal

A reviewer-grade manual deposit. Unlike the BBB locator repair, this proposal touches assertion text, so it does not auto-accept; the curation bot applies it locally with the --apply flag and the proposal carries a caveat that evidence review is still required.

proposal id

vpr_c322c1e3251c7878

kind

finding.add

actor

agent:vela-curation-bot-2026-05-09

created

2026-05-08T21:47:17.953594+00:00

reason

Manual finding added to frontier state

step 3 · diff

{
  "id": "vf_1731753d816fb7d0",
  "status": "absent",
  "assertion": null,
  "evidence": null
}

{
  "id": "vf_1731753d816fb7d0",
  "assertion": "In TRAILBLAZER-ALZ 2, donanemab slowed CDR-SB progression by 0.67 points at 76 weeks vs placebo in early symptomatic Alzheimer disease participants with low or medium tau pathology (CDR-SB difference -0.67, 95% CI -0.95 to -0.40, P<.001).",
  "evidence_span": {
    "section": "abstract",
    "text": "LSM change in CDR-SB score at 76 weeks was 1.20 (95% CI, 1.00-1.41) with donanemab and 1.88 (95% CI, 1.68-2.08) with placebo (difference, -0.67 [95% CI, -0.95 to -0.40]; P < .001) in the low/medium tau population"
  },
  "confidence": { "score": 0.7, "basis": "operator-supplied frontier prior; review required" }
}

The finding lands with one verbatim evidence span quoted from the JAMA abstract, plus a typed depends link to vf_29671540d388fc5d@vfr_5076e7b3ff8e6b0f on the anti-amyloid translation frontier. The link is the calibration: this CDR-SB effect is read in reference to a separately declared MCID anchor.

step 4 · review

The proposal is applied locally on creation, but the finding carries a caveat: Manual findings require evidence review before scientific use. The review queue keeps the record visible until a reviewer-grade verdict is appended.

status

applied

reviewer

agent:vela-curation-bot-2026-05-09

reviewed at

2026-05-08T21:47:17.962115+00:00

step 5 · event

The accepted proposal lands as one finding.asserted canonical event. The full finding payload travels inline so a fresh replay_from_genesis reconstructs the finding from the event log alone.

event id

vev_2c19f1a5bd7801b8

kind

finding.asserted

target

finding · vf_1731753d816fb…

actor

agent:vela-curation-bot-2026-05-09

timestamp

2026-05-08T21:47:17.962115+00:00

before_hash

sha256:…256:null

after_hash

sha256:…d9a82b1d

step 6 · proof

After acceptance, the same materialize step regenerates the early-AD frontier's frontier.json, vela.lock, and proof packet. The proof's event_log_hash commits to the full accepted-event history including this manual deposit.

./target/release/vela proof verify projects/early-ad-biomarker-calibration --json

frontier id

vfr_a22c9022674a2304

snapshot_hash

sha256:…df8e0070

event_log_hash

sha256:…e27e3324

proof_freshness

fresh

step 7 · impact

The donanemab finding has no declared downstream dependents on this frontier; it is itself the dependent, calibrated against an upstream MCID anchor on the anti-amyloid translation frontier. The cross-frontier depends link makes that relationship inspectable: a reader running vela impact on the upstream anchor sees this finding listed.

./target/release/vela impact projects/early-ad-biomarker-calibration vf_1731753d816fb7d0 --json

direct dependents

0

total downstream

0

accepted events

1

cross-frontier link

depends → vf_29671540d388fc5d@vfr_5076e7b3ff8e6b0f

Same shape, different domain. Two frontiers, one substrate contract: every state change is a proposal, every accepted proposal is one canonical event, every event is replay-checkable from the proof packet.