Connections that preserve research boundaries

1. Direct ELN read: Benchling foundation

Inside an owner-controlled isolated project, preview up to 20 explicit entry IDs, then accept selected metadata. Acceptance re-reads the entries and rejects changed revisions. Existing draft notes and research choices are not overwritten; three prior metadata snapshots are retained. Only ID, title, modified time and same-tenant source link are accepted. Notes, attachments, results tables and provider writes are excluded. Missing entries are retained rather than deleted.

Not connected by default. An authorized operator must configure BENCHLING_TENANT (tenant name only), BENCHLING_API_KEY (read-only personal key) and BENCHLING_OWNER_ID (the permitted ASTRA account ID) in server-side secrets. Do not paste keys into the website, project exports, URLs or support requests. Verify provider API entitlements and costs first; this update neither purchases access nor installs credentials. Other ASTRA accounts and packet reviewers cannot use that tenant key.

The adapter uses Benchling’s selected-entry API with bounded metadata fields and documented authentication. It is read-only and on demand, not full ELN synchronization, unattended polling or bidirectional editing. Real-tenant acceptance tests remain pending. LabArchives is a candidate, not an implemented connection.

2. Machine-to-project sensor ingestion

An isolated-project owner can explicitly accept an operational threshold and create a seven-day, single-project token. The bearer token permits sample submission only: no project reads, edits, grant changes or backups. It is shown once; the server stores its SHA-256 digest, not plaintext. Keep the token on the sending device. It is not an authentic scientific signature.

Use the exact endpoint shown after token creation. Send one JSON sample per request:

{
  "id": "unique-sample-id",
  "variable": "EXACT_CONFIGURED_VARIABLE",
  "value": 1.0,
  "unit": "EXACT_CONFIGURED_UNIT",
  "observedAt": "REPLACE_WITH_CURRENT_ISO_8601_UTC_TIME"
}

The displayed value is a format example, not a physical threshold. Configure the sender to report real observations with explicit provenance. Use the optional stdlib Python sender:

python3 send-project-sample.py --endpoint YOUR_EXACT_ASTRA_ENDPOINT --token-file OWNER_ONLY_TOKEN_FILE --sample SAMPLE_JSON_FILE

The sender rejects other hosts and redirects. Do not put credentials in command arguments. This helper sends one sample only; it does not attach hardware, schedule polling, buy cloud services or configure your lab network.

• Up to three active feeds and 20 retained grants per project; one new sample per minute per feed; latest 20 samples retained. Rate limits are not an infrastructure billing cap.
• Exact variable and unit required. No implicit unit conversion or invented threshold. Samples older than 24 hours, out of order, or over one minute into the future are rejected.
• Recent sample IDs are idempotent. Reusing an ID for different content fails. Concurrent project writes reject rather than overwrite; retry the same sample ID.
• Threshold crossings append an operational reassessment event and update feed status. Returning inside the criterion does not resolve an objection, validate a result or change a research decision.
• Revocation, expiry, project deletion and restore stop token use. Restore does not reactivate grants. It restores the research draft only; current ELN metadata, sample history and audit events are not rolled back. Latest feed samples are not a complete instrument archive.
• Flags appear when the owner reopens the project. No email, push notifications, safety interlocks or unattended literature search. Actual sender deployment and production verification remain pending.

3. Domain preparation templates

General science, energy/infrastructure, clinical-trial preparation, machine-learning reproducibility and psychological meta-analysis checklists are available in Research tools. Applying a template fills only empty method/criterion fields with explicit unknowns and prompts; existing research is preserved. These are editable templates, not validated ontologies, clinical advice or permission to upload participant data. The numerical condition engine remains limited to its documented units and contexts.

4. Repository and funder preparation

Prepare an OSF registration handoff, not an “official ASTRA preregistration.” Follow OSF’s registration process, supply a full protocol, permissions and contributors, and obtain the actual repository confirmation. No OAuth, upload, partnership or independent timestamp is established by an ASTRA export.

The NIH worksheet points to the updated 2026 DMS elements. Use the official required format and institutional review. Horizon Europe prompts cover general FAIR data stewardship under European Commission open-science guidance; they are not a grant/call-specific compliance mapping. No submission, financial commitment or acceptance is automated.

5. Open schema, private application

Portable Decision Record JSON Schema 1.0.0 is published under a schema-only MIT license. Anyone may implement compatible tooling under those terms. The application source remains private, and the schema license does not license researcher data, trademarks or third-party sources. A compatible file is not verified science, permission to publish, an immutable log or proof of compliance.

The portable format is a working-draft subset, not a complete project backup: keep native project JSON for full notes, accepted audit/condition histories, pilot preparation and recovery. Portable exports can be previewed through the existing project-file import control and restored as a new local draft. Original source IDs are retained, source-reading must be reattested, and no account identity, sharing grant or registration is restored. This importer accepts version 1.0.0 only and rejects unsupported versions or passages over its 4,000-character import limit. Keep the original file. Version policy: retain existing 1.x field meanings; incompatible changes require a new major schema ID. An unknown major version must be rejected or preserved uninterpreted, never silently converted. No public source repository or standards-body endorsement is claimed.

6. Connected research package

In the workspace, open Save & share → Prepare a connected research package. Preview the protocol, exact source notes, original measurement CSVs, structural warnings, descriptive calculations, objections, responses and next test. Private protocol and reviewer history is opt-in and requires a matching saved project. Inspect names and sensitive text before downloading or sharing.

The envelope contains payloadText (the exact UTF-8 JSON string) and its SHA-256. The matching Python 3 replay checks payload and CSV hashes and recomputes per-batch, same-unit descriptive summaries. Run python3 astra-replay-measurements.py astra-research-package.json. It does not pool experiments, propagate uncertainty or establish a scientific conclusion. Older draft bodies and external files are excluded; retain separate native backups. This package is not accepted by the native project restore control.

Protocol history now supports independently retained prefix checkpoints: later events may be appended while earlier commitments must still match. This is a linked hash history, not a Merkle transparency log, signature or witnessed timestamp. Retain checkpoints outside ASTRA. Frozen registrations and amendments can be inspected in the protocol panel.