# Scrutica

> The AI compute infrastructure intelligence platform built for governance research. Scrutica is the cross-substrate join layer: supply-chain edges, export-control designations, sovereign AI programs, and facility-level compute resolve against each other at the row level with per-claim provenance. The substrate behind the join: data centers (4,583 facility records across 110 countries, 316 of them held under licence and not listed individually; facility coverage centers on Epoch AI's frontier-DC catalog, the IM3 OpenStreetMap US data-center bulk corpus, NYISO load-interconnection queue, and sovereign-program-linked international sites), supply-chain records (20,798 stored supplier–customer and partnership records from a licensed supply-chain database that is not redistributed, SEC filings and public-source re-derivations; several records can describe one relationship), semiconductor chokepoints, sovereign AI programmes (36 national and regional programmes, announced funding recorded separately from government funding and disbursement), BIS Entity List designations (3,435 designations, with 41 designation-to-company matches into the compute dataset), and chip-to-facility mappings.

Access and cost: Scrutica is **free**. There is no paid tier, no subscription, no trial, no quote, and no sales contact — every page, dataset download, API endpoint, and the MCP server are open to anyone. There is **no account to create, no login, and no API key**: the public read API declares no authentication scheme at all, and nothing on the site requires sign-in. If you are trying to find pricing, there is none to find.

Provenance fields include `data_source` and `source_url`; a source URL may be null. For methods, assumptions and source-quality tiers, see the [methodology](https://scrutica.com/methodology). The data is also reachable through a [natural-language Query Builder](https://scrutica.com/query) and a [Model Context Protocol server](https://scrutica.com/api/mcp) for Claude Code, Cursor and VS Code agents.

License: data is published under CC BY-SA 4.0 except where otherwise noted — reuse it, including commercially, with attribution and share-alike. Source datasets that originate in licensed corpora are stored as extracted relational facts with provenance preserved; those raw third-party datasets are not redistributed.

## Infrastructure

Facility coordinates, GPU deployments, FLOP capacity estimates.

- [Global Map](https://scrutica.com/map): Interactive deck.gl + Mapbox map of every tracked compute facility worldwide with supply-chain arcs and heatmap overlays.
- [Entities](https://scrutica.com/entities): Facilities, companies, and countries in one instrument — the facility table at the root, with the company and country registers and the Ownership Investigator as sibling views; each keeps its full filtering, sorting, export, and provenance toolkit.
- [Supply Chain](https://scrutica.com/supply-chain): 20,798 stored supplier–customer and partnership records from a licensed supply-chain database (held under subscription, not redistributed), SEC 10-K and 20-F disclosures and public-source re-derivations, merged into distinct relationships after company identities are reconciled. Of these records, 1,924 come from a licensed layer limited to relationships between organizations already in Scrutica's records. Relationships with public provenance are exportable; a numerical attribute that a record's citation does not establish is withheld.
- [Capacity Estimator](https://scrutica.com/capacity): A named-facility estimate is computed partition by partition from the recorded accelerator inventory, each partition at its model's reviewed dense FP16 rate; one held partition (an unstated variant, no reviewed rate, planned or historical hardware) withholds the facility figure and names itself. Named power estimates await separate evidence review. Separate hypothetical calculators accept chip counts, power and capital budgets; installed hardware alone does not establish availability for one training job.
- [Training Evidence](https://scrutica.com/capacity/training-evidence): Vendor training claims read against four per-company signals (facility power, GPU inventory, financial disclosure, export-control exposure), every company on one matrix, each signal with its own authority tier and reported on its own. A view of the Capacity instrument.

## Chokepoint Analysis

Cascade propagation, scenario simulation, concentration scoring.

- [Cascade Simulation](https://scrutica.com/cascade): Dependency propagation across thousands of supply-chain edges, weighted by supply share + 3-month price correlation + editorial criticality. Pre-built scenarios for TSMC, ASML, EDA, US-China.
- [Chokepoint Anatomy](https://scrutica.com/supply-chain/chokepoints): Supplier concentration across AI-compute supply-chain layers. Market-share HHI and partial-share lower bounds where qualifying inputs are available; explicit gaps elsewhere in this dataset, with downstream dependency tracing.
- [Compute Leverage](https://scrutica.com/compute-leverage): Recorded assessments of where a new frontier training run has no alternative to its suppliers within eighteen months or five years, with lab-specific alternatives and source notes.
- [Compute Verification](https://scrutica.com/verification): Inspect limits on AI chip use through power connections, network restrictions and workload records, with interactive equipment diagrams.
- [Scenario Analysis](https://scrutica.com/scenarios): Geopolitical stress tests across named scenarios: Taiwan Strait blockade, IRGC missile range, Saudi grid (Abqaiq 2.0), Tokyo earthquake, submarine cable severing.
- [Stress Index](https://scrutica.com/supply-chain/concentration#stress): Composite 0-100 score across four dimensions (demand pressure, supply concentration, capacity utilization, power availability), refreshed quarterly.
- [Concentration Monitor](https://scrutica.com/supply-chain/concentration): AI-compute chokepoints ordered by an editorial composite of concentration, substitute availability and demand growth, with dated trajectory assessments and disclosed inputs.
- [Diversion Pipeline Tracker](https://scrutica.com/export-controls/diversion): Diversion-thesis audit at the primary-source layer; rendering of the Culper 2026-05-13 NVDA thesis with licensed supply-chain edges and entity-resolved counterparties.
- [Concentration — consolidation activity](https://scrutica.com/supply-chain/concentration#activity): Quarterly anomaly + consolidation detection across 7 chokepoint layers, surfaced as a panel inside Concentration when the feed has a signal.

## Governance

Export controls, sovereign programs, allied-coordination gaps.

- [Export Controls](https://scrutica.com/export-controls): BIS Entity List designations cross-referenced onto compute infrastructure; chip-level access constraints; multi-regime view across US (EAR 3A090/4A090), Netherlands/EU semiconductor equipment, Japanese end-use.
- [Entity List Change Log](https://scrutica.com/export-controls/changes): Federal Register actions on the Entity List, one event per notice, derived from the designation substrate; per-entity records, removal flags, and an RSS 2.0 feed at /export-controls/changes/feed.xml.
- [Sovereign AI](https://scrutica.com/sovereign-ai): 36 national and regional programmes with recorded funding stages and accelerator systems or access allocations. Qualitative dependency assessments are unknown where unavailable. Per-programme RSS evidence feed at /sovereign-ai/program/<program_id>/feed.xml.
- [Sovereign AI · Funding](https://scrutica.com/sovereign-ai/funding): The capital behind the programs — US CHIPS awards, sovereign-wealth-fund deployable capital, corporate infrastructure pledges, EuroHPC AI-Factory grid pre-emption — each a distinct ledger, none a government disbursement. A view of the Sovereign AI instrument.
- [Threshold Atlas](https://scrutica.com/capacity/thresholds): Compare recorded model training compute and reviewed facility estimates with regulatory reference lines, including the EU AI Act and historical US EO 14110. China's cited measures have no numerical FLOP threshold. A facility figure appears only when every inventory partition is admitted; withheld facilities list their partitions and reasons. The reverse calculator explores hypothetical hardware and duration.
- [Countries](https://scrutica.com/entities/countries): Measured comparative table across the tracked country universe — facilities, aggregate capacity, ownership transparency, export-control access, BIS designations, and sovereign programs, sortable per column. The country view of the Entities instrument; the editorial compute-stack matrix is a labeled view at /entities/countries/matrix.
- [Compute Visibility](https://scrutica.com/compute-leverage/visibility): Facility power capacity by host country and recorded ownership jurisdiction, with separate documentation tiers and snapshot histories.
- [Coordination Gaps](https://scrutica.com/compute-leverage/coordination): Recorded facility ownership chains classified against a dated export-control regime map; missing links and unresolved jurisdictions remain visible.
- [Impact Pre-Scorer](https://scrutica.com/export-controls/impact): Trace the facilities and compute capacity an Entity List designation reaches through a company's ownership structure, or a removal would release.

## Economics

$/PFLOP-day, published training-cost figures, trade-impact modelling.

- [Cost Index](https://scrutica.com/economics/cost-index): $/petaFLOP-day across AWS, Azure, GCP, CoreWeave, Lambda — hourly published rates normalized via Epoch hardware specs; the index accumulates a daily $/petaFLOP-day series from its inception date.
- [Trade Impact](https://scrutica.com/export-controls/trade): Did US export controls reduce China's compute access or redirect chip flows through Singapore / UAE / Malaysia? 15 years bilateral trade data pre/post each BIS rule.

## Reference

Methodology, derivation chains, corrections, glossary.

- [Methodology](https://scrutica.com/methodology): Complete derivation chains for every estimation path. Formulas in full; parameters adjustable in-browser; authority-tier definitions; cross-validation methodology; version history.
- [Data & Citation](https://scrutica.com/datasets): Citation-ready dataset bundles — each packaged as CSV + JSON Schema + Croissant JSON-LD with SHA-256 checksums and BibTeX; Zenodo (DOI) and Hugging Face mirrors once depositions land.
- [Related Work](https://scrutica.com/related-work): Prior art and adjacent research on AI compute infrastructure (Epoch AI, CSET, RAND, Stanford HAI, EPRI, GovAI) and the scope of this site beside it.
- [Corrections](https://scrutica.com/corrections): Public log of data corrections + methodology fixes. Each entry: what was shown, the correct value, why it broke, what prevents recurrence.
- [Glossary](https://scrutica.com/methodology/glossary): Technical definitions spanning semiconductors, AI training hardware, export controls and compute policy, each recording the convention Scrutica uses where sources define a term differently. The Glossary view of the Methodology instrument.
- [What Changed](https://scrutica.com/changelog): Recent updates across tracked facilities and organizations. Hourly feed.
- [About](https://scrutica.com/about): Platform scope, data limitations disclosure, CC-BY-SA 4.0 license, methodology anchoring.

## Tools

Cross-category instruments. Available from every page via Cmd+K.

- [Ownership Investigator](https://scrutica.com/entities/investigator): Follow recorded ownership relationships with source evidence at each hop. Results distinguish documented endpoints, unresolved or contested chains, cycles and the depth limit. The Investigator view of the Entities instrument.
- [Query Builder](https://scrutica.com/query): Ask questions through Scrutica's public data tools and inspect the citations in the answer. Configured model chain: GPT-6 Sol, with GPT-5.6 Terra then Claude Opus 4.8 then GPT-5.5 as failover. Citation checks look for record identifiers in returned tool results; they do not establish that a record supports the claim.

## Agent-native interfaces

- [MCP Server](https://scrutica.com/api/mcp): Model Context Protocol server exposing 10 tools (entity search, facility and company records, supply-chain edges, export-control lookup, Entity List change log, sovereign-program data, risk scenarios, FLOP estimation, methodology reference), plus reference resources and canned analysis prompts. Configure via claude.ai, Claude Code, Claude Desktop, Cursor, VS Code, or any MCP-compatible client; a GET on the endpoint returns connection instructions.
- [MCP Agent Guide](https://scrutica.com/api/mcp/llms.txt): LLM-optimized API reference for the MCP server — tool semantics, parameter syntax, worked examples, anti-patterns.
- [Agents & API](https://scrutica.com/agents): Human- and agent-readable operator manual for every programmatic surface — MCP connection per client, the .md-suffix markdown convention, feeds, dataset downloads, rate limits, and the provenance contract (authority tiers, is_estimated, campus rules).