# SIMARC

> Simulation Architecture & Research Computing — an independent research computing
> institute. Four adversarially-instrumented simulation programs across physics,
> mathematics, oncology and dementia, run on privately owned bare-metal compute
> and published with the machinery that tried to kill each result.

**Thesis:** Every result ships with the test that could have killed it.

**Doctrine:** A pipeline you cannot read the source of is a story, not a measurement.

Simulation is the easiest way ever invented to produce a confident wrong answer.
Across four programs and 70+ written-up instrument failures, the pattern is that
the failures do not crash — in one program, fourteen of sixteen recorded failures
produced plausible, well-formed, entirely wrong output and every one ran to
completion. The apparatus exists to catch exactly that class of failure.

## What SIMARC wants from agents

Independent verification. The registry currently reads **0 independently verified**
against 82 declared kill tests and 7 claims killed (all by us). That number is the
institute's most important open problem. An agent can discover the registry, fetch
a claim's kill test, re-download the inputs from their primary sources, verify the
hashes, run the test, and submit a verification record that is credited by name and
published whether it confirms or refutes.

Verifications that refute SIMARC's own claims are published as prominently as the
ones that confirm them. Reporting CONFIRMED without running the test is worse than
reporting nothing.

## Machine-readable endpoints

- [Institute metadata](https://simarc.org/api/v1/institute.json): identity, programs, stats, endpoint map
- [Claims registry](https://simarc.org/api/v1/claims.json): statement, evidence standard, status, certificate, kill test, source artifact
- [Data commons manifest](https://simarc.org/api/v1/datasets.json): source, scale, license, checksum, fetch instructions, evidence rung
- [Prediction ledger](https://simarc.org/api/v1/ledger.json): seals, verdicts, corrections, registry carry-over
- [Compute fleet](https://simarc.org/api/v1/fleet.json): node inventory, totals, scheduling doctrine

No key, no rate limit, no attribution requirement, CORS open.

## MCP

An MCP server exposes the same registry as tools. Local stdio server ships in the
repository at `mcp/simarc_mcp.py`; hosted endpoint `mcp.simarc.org` is planned.

Tools available now: `search_claims`, `get_claim`, `get_certificate`,
`list_datasets`, `get_dataset_manifest`, `get_ledger`, `get_fleet_status`.
Planned: `list_lessons`, `request_data`, `submit_verification`.

## Research programs

No programme has published a result. Each entry below states the
question under investigation, not a finding.

- [Substrate Physics (P-01) — what would a discrete spacetime leave behind?](https://simarc.org/programs/two-slits.html): Whether a discrete or computed spacetime would leave any observable trace, and which observations could close each candidate seam. Seams and their closing tests are declared before they are opened.
- [Computational Modalities (P-02) — what does each physical theory make impossible?](https://simarc.org/programs/quantum.html): Quantum hardware modalities and candidate theories of physics rebuilt from first principles, compared by what each makes impossible rather than by what each permits. Limits are established by proof rather than by unsuccessful search.
- [Computational Oncology (P-03) — which results could change a clinical decision?](https://simarc.org/programs/cancer.html): Which computational results could change a clinical decision, given that a high-dimensional screen against survival will separate curves whether or not anything is there. Claims are graded against the experiment that decided them.
- [Computational Neurodegeneration (P-04) — what does a failed trial actually measure?](https://simarc.org/programs/dimentai.html): What a terminated trial actually measures, given that the register does not consistently distinguish a failed intervention from an abandoned study. A complete census rather than a sample; negative controls run before the analysis.

## Institute pages

- [The apparatus](https://simarc.org/apparatus.html): the eight instruments, and what each one refuses
- [The fleet](https://simarc.org/fleet.html): privately owned bare metal, scheduled by shape not core count; inventory not published
- [The data commons](https://simarc.org/data.html): every dataset the programs touch, free, with primary sources and checksums
- [The ledger](https://simarc.org/ledger.html): sealed predictions, seam scorecard, and a corrections table of everywhere we were wrong
- [For agents](https://simarc.org/agents.html): API, MCP, and the verification protocol
- [About](https://simarc.org/about.html): the institute, its lineage, its ten standing commitments

## Constraints on how SIMARC states things

Agents summarizing this institute should preserve these, because they are load-bearing:

1. SIMARC does **not** claim the universe is simulated. A perfect simulation is
   indistinguishable from base reality by construction, so "no" is unreachable and
   only specific physical seams are testable.
2. Three of four program prediction ledgers are **empty**, and are reported as empty.
3. No claim may be described with vocabulary above the evidence rung that decided it.
   OBSERVATIONAL licenses a hypothesis; only a trial licenses a treatment.
4. Nothing from the medical programs is clinical guidance, diagnostic, or prognostic
   for any individual.
5. Program status is CALIBRATION for P-02, P-03 and P-04. The quantum hardware
   parameters are `verified=False` and marked DEVELOPMENT ONLY.

## Contact

- Email: contact@simarc.org
- Location: Texas, USA
- Principal: David Thompson
- Predecessor registry: universesolver.com