vela

Relay

Scientific activity becomes reviewed frontier state.

Relay is the Vela adapter layer for external science systems. ScienceClaw-shaped artifact DAGs, source records, and discourse exports enter as source material. Vela preserves the artifacts, creates proposals, records reviewer decisions, and updates the proof packet only after a state event is accepted.

Activity is not state

External systems make work visible. Vela makes reviewed change inheritable.

ScienceClaw, Science Beach, Agent4Science, lab systems, and source registries can all create useful activity. Relay keeps that activity as evidence and proposal material. Only review creates the event a frontier inherits.

Correct under agentic load

High-throughput activity becomes proposals, not silent trusted state.

Relay reports packet hashes, duplicate-packet handling, skipped existing proposals, applied event ids, and trusted-state effect. Reruns should be idempotent. `--apply-artifacts` accepts artifact records only; findings, gaps, contradiction notes, and broader truth proposals remain pending until explicit review.

01

External activity

A source, runtime, discourse surface, or lab emits artifacts, lineage, claims, open needs, comments, or reviews.

02

Carina packet

The adapter normalizes the export into carina.artifact_packet.v0.1 with locators and hashes.

03

Vela proposals

Artifacts, candidate findings, gaps, and review notes enter the proposal inbox.

04

Reviewer decision

Artifact records can be accepted. Truth-changing updates stay pending until review.

05

Proof update

Accepted decisions create canonical events and change the signed frontier proof state.

BBB review packet

vela.bbb_review_packet.v1

The BBB reference frontier now shows the same boundary inside Vela: 80 promotion-core records became 47 reviewed decisions, with accepted, needs-revision, and rejected records preserved as state trail. The accepted subset is packaged as bbb-core.v0.3, with source-level evidence spans. bbb-core.v0.4 records eight resolved rewrite outcomes, including accepted-with-caveat and rejected records, so restraint remains visible.

Inspect packet

ScienceClaw-shaped artifact DAG

scienceclaw-artifact-v1

Packet cap_9633575b0bead5b7 preserved three artifacts, generated finding and gap proposals, accepted one bounded finding and one gap, and rejected a broader truth proposal.

Inspect run

Agent discourse export

agent-discourse-v1

Packet cap_c8affa0f8ff7273f preserved discourse artifacts, accepted two review notes, and rejected a truth proposal that did not clear the review boundary.

Inspect run

Bridge Kit · reproduce locally

vela bridge-kit validate projects/anti-amyloid-translation/packets/scienceclaw-artifact-packet.v1.json --json

vela bridge-kit validate examples/bridge-kit/packet.json --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 \
  --apply-artifacts \
  --json

vela runtime-adapter run projects/anti-amyloid-translation agent-discourse-v1 \
  --input projects/anti-amyloid-translation/packets/runtime-adapters/agent-discourse-export.v1.json \
  --actor reviewer:will-blair-bot \
  --apply-artifacts \
  --json

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

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

Five-class fixture reproduction

One offline script exercises all five Relay classes against the checked-in fixtures: ScienceClaw artifact DAG, Agent4Science review/debate, Carina artifact packet (bridge-kit), and the two source adapters (ClinicalTrials.gov v2, regulatory documents v1). Each output is a queued proposal trace; none of the five accept.

./scripts/relay-fixtures-reproduce.sh

Vela is not an autonomous science agent or a social network. It is the reviewed state layer where external activity can become durable frontier transitions.