Your AI knows facts. It can't connect dependencies. Frontier models can summarize filings. They cannot reliably traverse the physical dependencies between materials, suppliers, facilities, policies, and public companies. ForcedAlpha gives agents a supply-chain graph they can query before answering — trace shocks, verify claims, map hidden exposures, find the small-caps everyone downstream depends on, and refuse unsupported relationships instead of guessing. Connect via MCP (claude mcp add --transport http forcedalpha https://forcedalpha.com/mcp) in about 5 minutes. Works with Claude (MCP), Claude Code, REST API, and any MCP-capable agent.
Trace shocks — China restricts tungsten APT → Central Glass WF6/NF3 → NAND/HBM fabs → SK Hynix, Kioxia, Western Digital, Lam Research. Verify claims — Kanto Denka → Tesla? No edge on file. Refuses instead of hallucinating. X-ray portfolios — find hidden upstream chokepoints shared across a basket of holdings. Find the trade — rank small-cap companies by how much market cap sits downstream of them relative to their own size.
Headline: China restricts tungsten APT shipments to Japan. A generic model sees: a tungsten shortage. ForcedAlpha sees: the graph traces the shock to Central Glass Co Ltd's WF6/NF3 gas production, not the tungsten metal itself, transmitting into NAND and HBM fabrication at SK Hynix, Kioxia, Western Digital, and Lam Research. Why it matters: the tradable chokepoint is one hop away from the headline.
More worked examples — captured runs, Jul 4, 2026: Which US defense platforms depend on materials that are byproducts of Chinese-controlled base metal smelting? Gallium and germanium are recovered as byproducts of Chinese bauxite and zinc smelting, not mined directly — a processing chokepoint, not a mining one. The graph traces this down to the F-35 EHA rotor's N48UH-grade NdFeB magnet, which depends on Chinese-separated dysprosium and terbium sourced through the Baotou Rare Earth Processing Complex, and names Proterial, Shin-Etsu Chemical, and JL Mag Rare-Earth as producers of that magnet grade. Who benefits if China restricts indium phosphide substrate exports? The graph identifies AXT Inc. and IQE plc as the only Western-listed InP producer/epitaxy nodes traced in the cascade, with hard breaks named at OSI Systems, Santec Holdings, and Broadcom's Breinigsville fab — 130 downstream nodes across $25,933B in graph-reachable market cap. What happens if China tightens gallium export controls? (captured run, Jul 5, 2026) The graph seeds the trace deterministically on the China gallium production chokepoint and walks it into GaAs substrates, GaN substrates, and trimethylgallium precursor chemistry — naming Lumentum and L3Harris as the most directly exposed listed names, across $36,042B graph-reachable downstream market cap at depth 3.
A live single-lane ask box on this page runs any question through the graph-grounded pipeline only (not a comparison — see forcedalpha.com/compare for the side-by-side against Claude alone).
26 tools grouped by job: trace shocks (trace_dependency, trace_shock, path_between, who_profits_if, co_product_contagion, shock_brief, watchlist_from_shock, find_customers, supply_chokepoints), verify claims (verify_edge, verify_claim, verify_path), x-ray portfolios (portfolio_xray, portfolio_common_mode, portfolio_brief, find_chokepoints, find_critical_nodes, find_structural_asymmetry, find_surprise_dependency, company_dependency_card, company_surprise_brief, explain_asymmetry, stress_test_moat), and utility (search_nodes, capabilities, forcedalpha_brief). Call capabilities() first for the full manifest with your tier's access marked on each tool.
Does Kanto Denka Kogyo supply Tesla? verify_edge(KANTO_DENKA, TSLA) returns NO_EDGE_IN_GRAPH: "No direct edge found between Kanto Denka Kogyo Co., Ltd. and Tesla." Absence of an edge does not prove the relationship doesn't exist — it may be undisclosed or not yet extracted — and the tool states that caveat every time.
Grades: A verbatim quote, confirmed on the source page, B URL present quote pending, C reported/secondary, D inferred no source.
Kanto Denka Kogyo, a Japanese specialty-gas maker, discloses in its EDINET yūhō filing that Samsung Electronics accounted for 21.7% of revenue (¥13,548M), up from 20.6% the prior year. Confirmed on the source: page 24 of document S100W72W, the exact figure Ctrl-F-able on the source PDF. Kanto Denka is also named as a WF6 supplier to TSMC, SK Hynix, Kioxia, and Rapidus, but the only other quantified customer-concentration figure on file is Kioxia at 10.0% of revenue (¥6,494M, FY2025 yūhō) — everything past what the filing states outright is left as inferred, not asserted.
Lasertec Corporation discloses that Intel became its largest customer at 33.3% of FY2025 revenue (¥83,718M), up from 27.7% (¥59,210M) the prior year, overtaking TSMC (¥66,050M). Confirmed on the source: page 19 of EDINET document S100WQ7F, "Intel Corporation 59,210 27.7 83,718 33.3," Ctrl-F-able on the source PDF.
Jurisdictions covered: SEC, EDINET (Japan), DART (Korea), MOPS (Taiwan), and China customs data. Verbatim quotes exist on a minority of edges today — the grade travels with every edge so you always know how much to trust it, rather than presenting every answer with false uniform confidence. The tool refuses on no-edge rather than guessing. Coverage is specialist (physical supply chain dependencies), not generalist.
Free tier: 10 full-fidelity queries a month, same depth and provenance as Pro, then pauses until the monthly reset. Pro ($299/mo) raises that to 1,000/month, unlocks full basket depth, and adds Watch alerts. Get a REST API key at /docs/api/, or see the side-by-side comparison at /compare/.
Your AI knows facts. It can’t connect dependencies.
Frontier models can summarize filings. They cannot reliably traverse the physical dependencies between materials, suppliers, facilities, policies, and public companies. ForcedAlpha gives agents a supply-chain graph they can query before answering — trace shocks, verify claims, map hidden exposures, find the small-caps everyone downstream depends on, and refuse unsupported relationships instead of guessing.
Paste up to 12 tickers, pick a curated scenario, and get a graded, sourced note — not a chat answer. Same book and scenario always produce the same result.
Three curated scenarios, each pre-verified for real citation coverage on the tickers shown. Every run reads live from the graph — nothing here is a canned screenshot.
// small-cap, outsized downstream leverage find_structural_asymmetry( theme: "AI", max_cap_b: 5 )
IQE plc (LSE: IQE.L, market cap <$1B) — sole-scale supplier across every major III-V epitaxy platform (GaAs, InP, GaN, GaSb). Lumentum, Lockheed Martin, RTX, Northrop Grumman, and BAE Systems all carry a graph-traced dependency on IQE. The graph's downstream reach for this one node spans thousands of billions in aggregate market cap — a structural asymmetry ratio in the thousands-to-one range between what IQE is worth and what depends on it. Currently in an active UK takeover offer period. Exact ratios and full basket depth are a Pro feature; the pattern — a sub-$1B company sitting under a market few have mapped — is free to see.
Trace and verify are the trust layer. This is the payoff: rank companies by how much market cap sits downstream of them relative to their own size — the AXTI pattern, found by the graph instead of by hand.
Headline: China restricts tungsten APT shipments to Japan. A generic model sees a tungsten shortage. ForcedAlpha follows the physical dependency path, hop by hop, to the companies it actually reaches — the tradable chokepoint sitting one layer from the headline.
// China restricts tungsten APT shipments to Japan trace_shock(event: "china_tungsten_apt_restriction")
Before an agent asserts a supply relationship, it can check whether that edge actually exists in the graph — and get an honest refusal, with the caveat spelled out, when it doesn't.
verify_edge(from: "KANTO_DENKA", to: "TSLA")
“No direct edge found between Kanto Denka Kogyo Co., Ltd. and Tesla.”
Reveal the upstream chokepoint a whole basket of holdings shares — the risk no single-ticker lookup would surface. Paste a book and a scenario; get a graded, sourced note, not a chat answer.
// book × scenario note portfolio_common_mode( tickers: ["LMT", "NOC", "RTX", "OLN", "TDY"], scenario: "china_antimony_export_controls" )
Per-ticker verdict rows — exposed, indirect, or beneficiary — or an honest no-path verdict that names its cause: nothing there, a path that routes through an overly common hub node, or a linkage that exists only outside what counts as a supply relationship — each with its graph trace, weakest-hop source grade, and citation status. Try it live above with your own book and scenario.
Free-text question, graph-grounded answer only.
Real questions, run against the live pipeline. Captured run · Jul 4, 2026.
A terminal stops at one hop. The graph keeps going.
Call it as tools from an agent harness, or hit the routes directly. Both surfaces read the same graph.
The model doesn't change. Whether it can traverse the graph does.
| — | Claude alone | Claude + ForcedAlpha |
|---|---|---|
| Supplier questions | Confident prose, no sources. | Graded hops with filing quotes. |
| Unknown relationships | Guesses, presented as fact. | Explicit refusal, caveat stated. |
| Freshness claims | Uniform confidence, no dating. | Honest per-edge grading, dated where the source is dated. |
| Book exposure | Generic reasoning per ticker. | Deterministic per-ticker verdicts, same book → same result. |
| Finding small-cap leverage | Can't compute it — no graph to rank against. | Ranked by real downstream dependency, not sentiment. |
← scroll to see the Claude + ForcedAlpha column →
Same model, same question. The only difference is whether the agent can traverse the graph.
Call capabilities() first — it returns this manifest with your tier's access marked on each tool.
1,000 full-fidelity queries/month, always full provenance, cascade depth up to 4 hops, exact scores, full basket depth, Watch alerts.
Unlimited queries, full graph traversal (depth 6), bulk export, dedicated support, SLA.
Free gets 10 full-fidelity queries/month — same depth, same provenance, same exact scores as Pro, then pauses until the monthly reset. Verification tools stay free the whole time.
Worked shock analyses live in our research → forcedalpha.com/research