ASTRAScience Network
Set up a team investigation

Enter the team name, research question and optional reviewer once to prepare this workspace; nothing is shared or charged.

Next: add evidence, sign in to save a private project, then preview and share a review version. Assisted service availability is separate from self-service access.

Literature monitoring

PUBLIC BETA · RESEARCH WORKSPACE

Your investigation.

Question → Evidence → Connect → Decide → Outcome → Review.

How this investigation is stored

Your investigation is the work you edit here. A private project stores its account copy; a decision record is its reviewable output. Use the same browser draft throughout. Source mappings and measurement links require human interpretation. Complete JSON preserves the investigation; limited summaries are not full backups.

Save & share

No project open

How saving and review work
  1. Create a project, then save the browser draft.
  2. On return, select a project to open its saved working contents and autosave subsequent edits; recover any pending device copy first.
  3. Copy current fields, selected references and the optional measurement/dependency attachment into the packet preview; inspect all contents before granting reviewer access.
  4. The named reviewer opens the link and submits an attributed review. Reopen project metadata to inspect feedback.
  5. Export the frozen snapshot with review, or download a full private backup. Edit the draft and save a new version; earlier review snapshots remain unchanged.

Saving is private. Reviewers see only the version you share and can add feedback. Read security and access limits.

Sign in

1. Choose a project

Create a new project
Open an existing project

Projects created before the directory was added can still be opened by ID.

Open a project to continue; saving never shares your investigation.

2. Save your work

Autosave keeps this project private. Sharing is separate.

Autosave off until a project is opened or saving is enabled.

Load a saved version or restore a backup

These actions replace work. Export your current draft first.

Cloud version history

Latest five previous distinct drafts (older copies rotate out as you save); restores create a new version and revoke reviewer and feed access.

Separate backup
Keep downloaded backups in a separate trusted location; new backups include the current private project record and up to five retained previous drafts. External source files and expired versions are not included. A missing version or size limit stops export rather than returning an incomplete backup. Automatic offsite backup is not configured.
Recovered history is an archived original record, not active reviewer access. Older backups may lack prior draft contents.
3. Share for review
Review the fields below before granting your named reviewer seven-day access.
Shared: these fields, reference extracts and any measurement/check attachment; other project notes and later edits stay private.
Reviewer access and revocation
A reviewer signs in with the specified email and uses this version’s link; revocation stops future reads but cannot recall downloaded copies.
Prepare a review version before sharing.
Choosing a reviewer does not grant access; only Share this version does.
Shared attachment: included measurements contain original raw values, saved condition/transfer checks and next-test text.
Inspect or edit raw attachment formats
Shared attachment: raw measurements, saved condition/transfer checks and next-test text; clear it to exclude these from this version.
Lab metadata connection & scoped sensor feed
Owner-only, explicit setup. Benchling requires an operator-configured read-only tenant key bound to your account; none is installed by this interface. LabArchives and bidirectional ELN writes remain unavailable. Sensor flags are operational review requests, not validated scientific limits or safety controls. Do not send confidential or regulated measurements.
Sensor feed: accepted operational criterion
Token shown once, never stored in the browser draft. Keep it on the sending device; never put it in a URL or support request. Up to three active feeds and 20 retained grants; maximum one new sample per minute per feed; latest 20 samples retained. No email or background literature monitoring. Restore revokes feeds; deletion removes feed contents.
Connection setup, request format and operational limits →
Verify review history
Check private review snapshots and hash-linked reviews; retain a receipt separately to compare future history.
Receipt includes reviewer identities, review text and shared snapshots. Hashes are tamper-evident, not an immutable ledger or proof of reviewer expertise.
Keep a checkpoint outside ASTRA to detect later rewrites. This is not independently timestamped or immutable storage.
No integrity check run in this session.
Report an integrity problem
Access history and latest save
Retention and deletion
Reviewer access expires after seven days and can be revoked earlier. Expiry stops access; it does not delete retained packet content. Content remains until explicitly deleted; no timed purge is promised. Deletion removes the server project draft and packet contents but retains a minimal audit tombstone. Provider backups, browser copies and downloads are separate; do not assume immediate deletion from those systems. Do not upload regulated or highly sensitive pre-patent data; independent assessment is pending. See security, processing, retention and assessment status.




What happens in each stage?
  1. Question: define what you need to find out and search for sources.
  2. Evidence & measurements: inspect papers, import data, and read quality warnings.
  3. Check & decide: examine assumptions, record a decision, and plan the next test.
  4. Review, save & export: save your project, prepare a version for review, and download your records.





Define the question




One working project, saved in this browser.







Browser draft. Save to your account before switching devices.


About this step
This is a private working draft on this device, not a published Astra claim. Changing the question makes existing reviews need renewal. Sign in and save an account copy before switching browsers or domains.


AFE: source → proposed audit → project manifest
Stage an automated falsification audit
AI/PDF processing is not configured. This local intake accepts a structured audit you supply, previews it, and applies accepted suggestions to decision criteria and counterarguments. It does not prove claims false or verify competing literature.
AI reviewer assistance: scope and validation requirements

ASTRA should offer dependable AI assistance. But it should earn that description through testing. Its current AFE intake does not establish that capability. The tasks below are intended capabilities, not currently available automated processing.


Start with narrow tasks researchers can check:


    

  • Extract claims with exact quotations and page references.
    • 

  • Flag missing experimental details, such as controls, units, or starting conditions.
    • 

  • Suggest possible conflicts, while checking whether the studies used comparable conditions.
    • 

  • Draft review questions and test criteria, clearly separating reported values from suggestions.
    • 



Every output should show its source, identify uncertainty, and remain a proposal until a researcher accepts it. Missing information should stay missing; the AI must not invent thresholds or treat a confound as proof that a claim is false. Researcher acceptance records a decision; it does not independently validate the suggestion.


Before calling it dependable, evaluate it on domain-reviewed examples and measure unsupported claims, missed issues, false alarms, and whether it saves researchers time. That evaluation has not yet established dependable automated reasoning in ASTRA.


The strongest role is an AI reviewer assistant inside a traceable research workflow. That strengthens ASTRA’s value while keeping scientific decisions accountable to people.


Only filename, size, and digest are retained after acceptance; PDF parsing is unavailable. Files stay on this device.
 
No text is sent to an AI service. Review external tools’ privacy and cost terms before using them.
3. Export complete project (JSON) includes accepted audit snapshots and your current editable criteria and objections. Account saving remains an explicit separate action.



Search a specific source →

Already have sources? Go to your decision brief →






Save evidence and its limits


Open each original source and record what it supports or leaves uncertain.
Source notes and record types
Save sources from your search, open the original record, and write a note in your own words. Each note stays attached to its source and review history. Funding records are kept separately from scientific evidence.


Import a Mendeley BibTeX/RIS export →


Import a literature shortlist from CSV
Export a CSV from your literature tool, including Elicit where its export is available. Supported columns: Title or Paper title, DOI, URL, Authors, Year. This is a file handoff, not a direct API connection. Imported metadata is not inspected evidence.
Select a file to preview before adding references.














Removed sources (0)

Removed sources retain their notes and review history here and in your JSON export.









Restore a project from JSON

Choose an exported project. You can inspect its question and source count before restoring it.












Browser recovery copies

ASTRA keeps up to three dated session copies in additional browser storage as you work. Reopen this website in the same browser, download a copy below, then use “Restore a project from JSON” to inspect and restore it. Nothing is published or saved to your account automatically.


Copies may miss the last changes if the page closes before a save finishes. Clearing site data, private browsing or losing this device can remove them. Export a separate backup for important work.


Checking browser recovery…







Your browser decides whether to protect local copies from automatic eviction.






Write your next decision or test →



Your shortlist and worksheet stay in this browser until you save to your research account. Export a copy to keep or share them. They are not public Astra findings.






THE OUTPUT / PRIVATE DRAFT


Decide what to do next


Record your question, criteria, reviewer and provisional decision.
When to use the full decision form
Start with four essentials. Open the full version when consequences are substantial or difficult to undo. Both are working drafts, not approvals.









Link saved evidence
Which saved sources inform this decision?
Read the original before relying on a note. A source title is not evidence.







Test this with 2–3 past decisions · Software controls: current behavior and implementation plan


Preview this brief




Make this decision reusable


Save the evidence and reasoning behind your decision.
What a decision record establishes
Save a structured record so another researcher or AI can inspect each source, its limits, the objection and the test rule. Source notes and review labels are provided by users; no result is automatically verified.




Pre-review your experiment plan
Inspect suggested gaps before exporting your plan.
How these local checks work
Local checks suggest missing controls, criteria, competing explanations, and relevant operating conditions as you edit. These are heuristic questions, not AI reasoning or scientific approval. Nothing is sent to an external model.
 
Inspect the brief before using an external AI tool. It can contain private notes; that tool’s access and charging terms apply.
Prepare evidence search from this hypothesis. Search submission is explicit; search results are candidates, not proof of support or contradiction.


Methodology review and Jupyter handoff
Inspect structured candidate objections, saved source-check coverage and recorded dependency holds. These local rules do not infer a validated causal graph or statistical power. English keyword checks can miss flaws or flag already addressed conditions.
The AI brief uses the same bounded evidence snapshot; it does not call a model. No global PubPeer feed, independent timestamp, zero-knowledge proof, hardware control or reviewer payment is connected.
Exports may contain private protocol text and source passages. Inspect before sharing. No account identifiers or unrelated project fields are included.
Copy notebook if downloading fails
Save these exact contents as astra-review-handoff.ipynb. Save the exact review JSON field below as astra-methodology-review.json; the formatted preview above uses different bytes.
When copying, add no trailing newline or other changes to the review JSON. A changed file is rejected.
 
Use both files together in your own Python 3 environment. The notebook verifies exact snapshot bytes and requires explicit human assessments before proceeding. It does not run your experiment, verify current online status, or prevent someone bypassing it. Re-export after every revision. Hashes are unsigned; retain the notebook separately.






Export to your review team or lab notebook
Download a review copy for your team or lab notebook.
File format and notebook support
Download the Markdown handoff and attach or paste it into your notebook where Markdown or attachments are supported. This is a portable file, not automatic notebook synchronization.










Sharing requires an invited research account. Check the record for private information before sharing.


Inspect linked decisions and their challenges → · Invited library access required.


Compare evidence across sources

Follow the ten-step research method →


Prepare a nine-section analysis from your question, selected records and supplied material. This does not run AI or search all five providers automatically.


Check original passages and facts as well as the reasoning. Multiple databases can list the same study; that is not independent confirmation.







Waiting for your question and supplied evidence. No data is sent to an AI service.


Preview / manually copy prepared output




Challenge a prior decision


You can also open a decision record from the invited document library.













Account draft & archives
Your research account


Sign in with ChatGPT to keep a private draft across devices. The shared repository is available to invited researchers. Account drafts are separate from published evidence and do not establish scientific review.



Checking account…



Open shared document repository →







Record a human review


Read the source, then record your assessment.
What a recorded review means
Read each source and its notes here, then record your decision. A named person must attest that they read the original source and checked the note. This records a self-reported review; it does not verify identity, independence or scientific validity.









Continue to project brief →



Optional study and funding plan
Other tools, product plans & background
Adoption capabilities, pilot criteria, and integration gates →
Version 129: independent governance and citation-indexing strategy →
Version 128: trust, academic credit, and local-first compute architecture →


Run bounded Python with explicit inputs →
Open the living-paper workbench: sections, corrections, protocol receipts, and local runs →


Optional full project record


Your brief includes the question, source identifiers, notes, limitations, review decisions and unresolved work. Add optional study or funding details below. Missing fields remain “Not supplied”; unreviewed notes remain clearly labeled.



Astra’s current stage: research program taking shape. A legal funding recipient, budget, delivery dates, scientific team and independent reviewer are not yet confirmed. Read the proposed work packages.




















Interface prototype — separate from the investigation demo
Open the multilingual interface prototype — fictional data, eleven language options, mobile disclosures and a lightweight layout. Your research workspace is separate.
How we will test whether ASTRA improves a research decision

Use the same bounded question, source set, date and time limit in ASTRA and a standard search-and-notes workflow. A qualified reviewer checks source locations and proposed tests without seeing which workflow produced them. Compare missed evidence, citation errors, test quality, researcher time and reviewer disagreement. Record disagreements and costs. No researcher comparison has been completed, so no effectiveness or time-saving claim is made.








Provider coverage and access limits

Source providers supply leads for investigation. Open the original records and check their methods and terms before use.






SageMathCell · math computation


Run SageMath calculations in an optional third-party workspace. Results are not proof of a theorem or a reviewed Astra finding. Code is sent to SageMathCell only when you load the tool.






Crossref


Broad publication metadata search without signing in. A matching title or DOI is a lead, not a reviewed finding.






OpenAlex


Optional DOI-based research connections on source cards. Keyless access can be limited; missing context does not invalidate a source.







Europe PMC



Publication metadata and links to original records. Read the methods, limitations, corrections, and retraction notices before relying on a study. The absence of a warning here does not establish reliability.






ClinicalTrials.gov


Registered study records and reported status. Registration is not proof that a study is complete, peer reviewed, safe or successful.






ENCODE


Functional genomics experiment metadata. ASTRA cannot search ENCODE directly while its server returns 403 to ASTRA requests; use the official portal link. Existing saved records remain in your private draft.






NASA Earthdata


Earth-observation collection discovery through NASA’s public metadata system. Search results describe available collections; downloading or analyzing a dataset is a separate step.







DataCite



Dataset and software DOI metadata. Verify file availability, version, methods and reuse terms at the repository.







Crossref, Unpaywall & OpenAlex



Use Check on a DOI-bearing paper to find recorded updates, open copies, and research context. These checks are live lookups; record important notices in your own evidence note. Missing metadata does not establish reliability, and citations or related work do not establish quality.







Grants.gov



Posted and forecast U.S. federal opportunities. A forecast is not an open call. Verify the official notice, amendments, deadline, eligible entity types and registration requirements. A search match does not establish Astra’s eligibility.






Results load in batches of 12. Publication filters use provider metadata; excluding a publication type does not establish study quality. Search settings are retained with saved sources. This is exploratory search, not an exhaustive or reproducible systematic review. Records can be cached for up to 10 minutes; their retrieval time is shown.






Open connected decisions, protocols and results → · Inspect the example decision record