METHODOLOGY
What this tool is built from, how it scores a firm, what it cannot see, and who to tell if something is wrong. Modeled on the dataset-documentation standard used by Georgetown's CSET Emerging Technology Observatory: exact source and vintage per field, limitations stated plainly, not buried.
Every field on this site traces to one of the sources below. "Live" means fetched fresh at the last data build; "fallback" means a live fetch failed and a curated snapshot was used instead — never silently, always labeled. Pull timestamps are UTC.
| Source | Mode | Pulled | What it feeds |
|---|---|---|---|
| Consolidated Screening List (CSL) | live | 2026-08-10 01:06:49Z | Scanned 25939 CSL entries; 23 seed companies matched via CSL. Unioned 4 curated fallback ID(s) missed by CSL name-match. Unioned 13 DoD 1260H (June 2026) match(es). |
| USASpending.gov — DoD award search | live | 2026-08-10 01:06:49Z | Scanned 1000 DoD semiconductor (NAICS 334413/334419/541715) awards; 4 seed companies matched. "programs" = sub-agency + PE/program tokens from Description. |
| Curated ownership aggregation | fallback | 2026-08-10 01:06:49Z | No new flags this run (parents already covered or no matching edges). |
| Curated China-ecosystem + Nexperia edges (Phase B, CSET ETO + TechInsights + TrendForce + SCMP + BIS Entity List + Nexperia press release) | fallback | 2026-08-10 01:06:49Z | 1 new node(s), 4 new cited edge(s), 2 existing edge(s) provenance-tagged, 11 ownership record(s). Always-on curated data, not fetch-dependent — see file for full per-edge citations. CSET ETO raw data itself contains company-to-technology-category market share, not firm-to-firm edges; used only to cross-check node identity. |
| Wafergraph seed graph | seed | — | The base company/taxonomy/market-share graph (566 firms) — the semiconductor supply-chain map every score in this tool is computed on. Copied into data/source/ and enriched by the live sources above at every data build; edges (who supplies whom) are updated on the wafergraph side, not by this build. |
| Historical-replay ground truth | curated | — | Documented real-world impacts used in the historical replay backtests (Renesas 2011, Nexperia 2025) — hand-compiled from contemporaneous public reporting, each entry carrying its own citation link on that page. Not part of the live data build; updated only when a new replay scenario is added. |
Every firm gets a composite criticality score from 0–100 (higher = more critical if the US lost access to it), built as a weighted average of four factors, each independently scored 0–100:
- Foreign-origin concentration (30%) — how much of the firm's input supply traces to foreign, especially adversary or single-country, origins.
- Sole-source dependence (25%) — how many downstream firms have no qualified alternative to this supplier.
- Sanctions exposure (25%) — proximity to sanctioned or restricted entities, either the firm itself or a direct supplier.
- Downstream criticality (20%) — how essential the firm is to defense programs and agencies, from matched DoD award data.
Deal-killer rule. If a firm — or any company it directly lists as a supplier — appears on a tracked sanctions or restricted list, the verdict is forced to CRITICAL regardless of the weighted composite. A sanctioned node in the immediate supply path is treated as a binary risk: procurement may become legally prohibited or operationally severed overnight, and no favorable factor score can offset that.
Verdict bands. Outside the deal-killer case, the composite maps to a tier: 70 and above is CRITICAL, 40–69 is WATCH, below 40 is NOMINAL / CLEAR. Treat the exact point score as illustrative, not measured — it is a triage ordering built from a heuristic weighting, not a calibrated probability. The tier and the factor breakdown behind it carry the actual signal; lead with those, not the number.
Stated plainly, not as a footnote. Every one of these is a real, current gap — not a hypothetical disclaimer.
- Name-matching false negatives. Sanctions matching is name-based against the Consolidated Screening List. It reliably catches a firm that is directly listed. It will miss a firm that is 50%+ owned by a listed entity but not itself listed (the OFAC 50% rule) and any hidden or indirect ownership structure — this tool cannot see beneficial ownership. Read this as recall-oriented, never as "zero false positives" or complete coverage.
- Graph coverage gap — Huawei / HiSilicon (downstream only). A 2026-08-09 pass added cited upstream supplier edges for Huawei and HiSilicon (SMIC — TechInsights' Kirin 9000s teardown; Sophgo — TrendForce's reporting on the DOC-confirmed TSMC diversion case) and checked CSET ETO's raw chip-explorer data as a possible source of more. That dataset turned out to hold company-to- technology-category market share, not firm-to-firm edges, so it could not supply any. Huawei and HiSilicon still have zero downstream (customer) edges — no citable public source describes a firm that depends on either of them for supply — so any scenario touching them still understates their real blast radius. This is a known, still-open data-layer gap, not a claim that they are unimportant.
- Edge staleness. Who-supplies-whom edges come from the wafergraph seed graph and are not refreshed on every data build — a supplier relationship shown here can lag a real-world change (a firm switching vendors, a new deal) by an unknown margin.
- Semiconductor-only scope. The graph models the semiconductor supply chain. Downstream impact on non-semiconductor industries — automotive assembly lines, for example — is out of coverage by design; the historical replay page reports those firms in an explicit outside-coverage bucket rather than guessing at them.
- No sub-tier / BOM traceability, by design. This tool scores firms and commodities from public data. It does not, and is not meant to, trace individual part numbers through a bill of materials — that is proprietary, often classified, and explicitly out of scope. See the project scope note in the repository's codebase map.
- NOVendor vetting or supplier approval decisions.
- NOCompliance or legal sanctions-screening determinations.
- NOInvestment or trading decisions.
Use a licensed screening tool (OFAC SDN search, a paid compliance platform) for anything with legal or financial consequence.
- YESSupply-chain risk analysis and education.
- YESHypothesis generation — where to look, what to ask next.
- YESWar-gaming and scenario exploration.
Public data only. Every source feeding this tool — sanctions lists, federal spending records, the seed supply-chain graph — is public. No proprietary, classified, or scraped-behind-a-login data is used anywhere in the pipeline.
AI-assisted build. This tool's scoring engine, data pipeline, and interface were built with AI assistance. That is disclosed here rather than hidden, and every analytic claim on this site is meant to be independently checkable against the cited source, not taken on faith.
Reproducible. The scoring logic and data build are open in the project repository; anyone with the same public sources can rerun the pipeline and check the output.
An append-only record of factual corrections to this dataset or its published analysis. Every correction lands here, dated, with what changed.
- 2026-08-09No public corrections have been filed against this dataset or its scoring output. This entry exists so the log is never silently empty: its presence is a claim that we would say so if something were wrong.