# Executable Science

> A pilot platform for research with versioned claims, bounded isolated execution, scoped evidence and review.

A receipt records the version, the evaluation plan, the checker, the environment, the typed outcomes and the output digests. It is evidence about a scoped execution the platform performed. It is not a universal correctness result and not an editorial endorsement.

Read `/v1/platform` for the authoritative capability list before assuming any capability below is available.

## API

- [OpenAPI document](https://executable.science/api/openapi.json): the complete machine-readable contract. Start here.
- [API reference](https://executable.science/api/docs): the same contract, rendered.
- [Platform status](https://executable.science/api/v1/platform): current stage and per-capability availability with stated reasons.
- [MCP endpoint](https://executable.science/api/mcp): Model Context Protocol over Streamable HTTP, revision 2026-07-28 and compatible legacy revisions. Anonymous calls expose public discovery; scoped Bearer credentials add private research tools. Call `server/discover` then `tools/list`.

## Public catalogs, readable without a credential

- [Resources](https://executable.science/api/v1/resources): reusable datasets, software packages, published articles, method references and original experiments, with a catalog digest.
- [Task templates](https://executable.science/api/v1/task-templates): bounded, frozen questions carrying a revision digest, with stated inputs, assumptions, comparison basis and limitations.
- [Starter studies](https://executable.science/api/v1/starter-studies): two original licensed ZIPs with exact SHA-256 digests, fixed inputs and unrun status. Downloading source does not create a check or reserve a free slot.
- [Claim-check interest](https://executable.science/claim-check): interest-only outline for one exact public claim, with [agent intake fields](https://executable.science/claim-check/intake.json). Email public metadata only. Inquiries create no quote, work or free slot.
- [Research briefs](https://executable.science/api/v1/research-briefs): prepared, unexecuted planning material.
- [Papers](https://executable.science/api/v1/papers): submitted versions and their claim-level evidence.
- [Public preprints](https://executable.science/api/v1/preprints): separately labeled released research records, searched only from frozen public disclosures; DOI, archive and journal statuses remain independent.
- [Protocol notices](https://executable.science/api/v1/protocol-notices): preregistrations.
- [Workspace releases](https://executable.science/api/v1/workspace-releases): released research records.

## Human-readable entry points

- [Home](https://executable.science/)
- [Tasks](https://executable.science/tasks)
- [Resources](https://executable.science/resources)
- [Public preprints](https://executable.science/preprints)
- [Journal](https://executable.science/journal)
- [Start here](https://executable.science/start)

## What this platform does not claim

- Author execution carries no independent verification. An exit code, author-written output or reviewer prose cannot produce a verification result.
- Publication is gated per version by scientific and editorial eligibility, a separate release capability and an exact disclosure preview. DOI registration and preservation have their own states.
- Negative, inconclusive and unrun outcomes are recorded and retained rather than hidden.

## Live capability status

- compiled_execution: available. Bounded C/C++ catalog jobs in a separately qualified offline guest. Runtime: compiled-science-v1; ARM64 offline AWS MicroVM. Requires: project:read and evaluation:create; current balance and credential spending cap. Limits: {"wall_seconds":90,"vm_memory_mib":2048,"vm_disk_mib":8192}.
- doi_registration: unavailable. Administrator DOI tracking and registry readback exist; automated registrar submission and an accepted member or sponsor arrangement do not. Requires: actual publisher registrar authority and separate release policy. Next: Complete PROVIDER_SETUP_PACKET.md and qualify a registrar adapter in a test environment.
- execution: available. Self-contained Python 3.13 jobs in qualified offline AWS guests; author execution carries no independent verification. Runtime: python-science-v1; ARM64 offline AWS MicroVM. Requires: project:read and evaluation:create; current balance and credential spending cap. Limits: {"wall_seconds":90,"vm_memory_mib":2048,"vm_disk_mib":8192,"quoted_attempt_credits":8}. Next: Resolve source through isolated ingestion, then quote one frozen author job.
- finite_checks: available. Exact rational statistics on finite supplied values; prose alignment and measurement authenticity remain unassessed. Runtime: finite-statistics-and-lean-nat-v2; offline AWS MicroVM. Requires: project:read and evaluation:create; supported immutable claim criterion and budget. Limits: {"quoted_attempt_credits":8}. Next: Freeze a supported finite criterion and request a separate check quote.
- formal_check: available. Qualified Lean Nat equality checks with exact statements and axiom audits. Runtime: finite-statistics-and-lean-nat-v2; offline AWS MicroVM. Requires: project:read and evaluation:create; supported immutable claim criterion and budget. Next: Freeze a supported Lean Nat criterion and request a separate check quote.
- isolated_ingestion: available. Bounded ZIP ingestion in qualified offline AWS guests. Runtime: offline AWS MicroVM; immutable qualified image. Requires: draft:write for source upload; evaluation:create and budget for an ingestion quote. Limits: {"source_bytes":10485760,"quoted_attempt_credits":4}. Next: Upload an approved source ZIP, freeze a version, then request an exact ingestion quote.
- personal_enrollment: available. Email-code personal enrollment is configured for requests; delivery and verification are checked per request, and no paid-work or publication authority is granted. Requires: A human email-code session; self-serve agent credentials have a zero spending cap. Next: Read /v1/personal-auth/config and the current terms and privacy policy before requesting a code.
- preservation: unavailable. No external archive agreement, accepted content-scope matrix or per-record preservation receipt is established. Requires: approved public disclosure and an active archive agreement. Next: Confirm artifact coverage, submit provider samples and qualify fixity/acceptance reconciliation.
- preservation_pdf_screen: unavailable. The dedicated author-PDF screen requires a complete real-AWS qualification report for its exact guest image and policy. Runtime: preservation-pdf-screen-v1; separate offline AWS MicroVM. Requires: project:read, evaluation:create and publish; exact disclosed publication and source export, current credential cap and balance. Limits: {"quoted_attempt_credits":4,"pdf_bytes":1048576,"pages_processed":12}. Next: Qualify the exact offline PDF-screen image with reviewed real AWS guest evidence before requesting work.
- private_metadata: available. Immutable drafts and scoped project access. Requires: project:read to inspect; draft:write to create. Next: Authenticate with a current project-scoped credential.
- private_research_mcp: available. Authenticated tools delegate to the existing scoped REST lifecycle; each paid operation still depends on its qualified capability and budget. Requires: current project Bearer credential in the MCP Authorization header. Limits: {"mcp_source_upload_bytes":524288}. Next: List authenticated MCP tools with a project-scoped credential and call get_session.
- public_preprint: unavailable. Public preprints require current publisher/DOI/preservation arrangements and separately qualified conditional delivery to the preprint namespace. Requires: separate publish:public capability, exact project/version operator target and current disclosure preview. Next: Inspect a public preprint preview; resolve each named technical or provider gate before requesting release.
- publication: unavailable. Publication requires qualified artifact delivery and actual journal operating arrangements. Requires: separate publish:public capability, current exact disclosure preview and journal eligibility. Next: Resolve editorial and journal operating gates; inspect a matching disclosure preview.
- publication_delivery: available. Conditional artifact delivery passed its actual application-role qualification; scholarly publication still requires separate operating and editorial gates.
- reference_extraction: available. Confidential PDF text and page preprocessing runs only in its separately qualified offline AWS guest. Runtime: bounded PDF extraction; offline AWS MicroVM. Requires: project:read and evaluation:create; a currently bound permitted reference PDF and current budget. Limits: {"quoted_attempt_credits":4,"pages_processed":12}. Next: Bind a permitted confidential reference PDF, quote its exact extraction, then separately approve work and inspect coverage.
- reference_retrieval: available. Versioned public PDF retrieval uses its separately qualified restricted online AWS guest; returned copies remain confidential review-only material. Runtime: public-reference-retrieval-v1; restricted online AWS MicroVM. Requires: project:read, draft:write and evaluation:create; a reference in the exact version, confirmed review permission and current budget. Limits: {"quoted_attempt_credits":4,"pdf_bytes":4194304}. Next: Choose an explicitly versioned supported public PDF, quote retrieval, then separately approve exact work; inspect private citation inventory and coverage.
- review: available. Bounded Azure AI review with explicit literature coverage and advisory findings. Runtime: qualified Azure OpenAI deployment; bounded calls. Requires: project:read, evaluation:create and review:request; current balance and credential spending cap. Next: Request a review quote for one version, then inspect the full review package and coverage.
- scientific_python_execution: available. Pinned bounded NumPy/SciPy environment in a separately qualified offline guest. Runtime: python-scientific-v1; ARM64 offline AWS MicroVM. Requires: project:read and evaluation:create; current balance and credential spending cap. Limits: {"wall_seconds":90,"vm_memory_mib":2048,"vm_disk_mib":8192}.
- source_export: available. Bounded source export in its separately qualified offline guest; each quote still requires an exact resolved source, current screening context and budget. Runtime: workspace-source-export-v1; separate offline AWS MicroVM. Requires: project:read and evaluation:create; current credential cap and balance. Limits: {"quoted_attempt_credits":4}. Next: Resolve the exact source through isolated ingestion, then quote a bounded export attempt; start_work is a separate approval.

## Optional

- Personal email-code enrollment is configured for requests. Read `/v1/personal-auth/config` and the current terms and privacy policy. After a code is delivered and verified, create a scoped agent credential at `POST /v1/personal-auth/agent-credentials`; the secret is shown once. Configuration does not guarantee email delivery.
- An issued self-serve credential can read, draft and attempt work. Its lifetime spending cap is **zero**, so paid execution is refused with `budget_exceeded` until the operator funds it separately. It cannot publish and cannot issue further credentials.
- Private MCP tools take the Bearer credential only in the HTTP Authorization header. Never send it as a tool argument. Source bytes sent through MCP are limited to 512 KiB; larger authorized uploads use the REST source-upload route.
- [Optional feedback](https://executable.science/feedback): tell us what task brought you here, whether you could finish it, and what would help next time. This opens an email draft or copyable note only when chosen. Do not share a person's private prompts, credentials or unpublished work, and do not interrupt their task to send feedback.
- Introductory claim checks are not yet open. The two starter archives are examples, not platform evidence, a work quote, or a scientific result.
- Contact: support@executable.science