Diversion Pipeline Tracker

Cascade graph — OEM → intermediary → end-customer

Each edge is one corporate-filing or trade-record-derived relationship in the alleged Culper-NVDA chain. Edge color encodes whether the relationship is primary-source corroborated (forest green), single-sourced via a Tier-3 analyst citation chain (warm amber dashed), or absent from FactSet’s supply-chain disclosure topology (warm-stone dotted). Toggle the “fade Tier-3 edges” control to render only the structural skeleton that survives at primary-source corroboration.

Cross-source disclosure tensions · 8

Where public primary-source disclosure (FactSet 2026-05-14 pull, SEC filings, A-share + HK + Bizfile / SSM filings) diverges from the Tier-3 narrative reading of the same entity. Each card juxtaposes the Tier-1 measurement against the Tier-3 interpretation in visually distinct boxes — the page never presents a Tier-3 reading as primary-source corroboration, and the cards make no claim that either layer is wrong. Both can be simultaneously correct.

Falsification framework · what would refute each major claim

Four primary-source-anchored signals, each anchored to a specific filing or program where the signal would surface if it surfaced at all. Refutation strength on each card (partial / substantial / load-bearing) is Scrutica’s analytical positioning of how decisively the named signal would refute the parent claim; observability window names the cycle in which a researcher could check.

Comparable cases · short-thesis genre context

Three comparable diversion- or cross-jurisdictional-corporate-topology short theses from the past five years, listed for genre orientation — outcomes noted. The panel describes how each case differs in evidence-type mix; it does not position the Culper thesis against these cases.

Three comparable short-thesis reports from the past five years, listed for orientation — they differ from the Culper · NVDA thesis on evidence-type mix (operational deception vs Chinese-listing topology vs offshore-vehicle topology vs Tier-3-trade-record + cross-listing + BGP) and on post-publication regulatory resolution. The panel is descriptive of the short-thesis genre; it does not position the Culper thesis against these cases.

• Hindenburg · Nikola CorporationHindenburg Research · 2020 · target: Nikola Corporation (NKLA)

  Operational-deception thesis anchored heavily in on-the-record former-employee interviews, internal emails, and a particularly damaging video of a non-functional truck rolling down a hill. Low cross-jurisdictional component; almost no trade-record dependence.

  Outcome. Trevor Milton (Nikola founder) convicted of securities + wire fraud (2022); Nikola Corp paid $125M SEC settlement (2021). Among the most regulatorily-vindicated short theses of the past decade.

• Muddy Waters · GSX / GOTUMuddy Waters Research · 2020 · target: GSX Techedu (later Gaotu Techedu, GOTU)

  Student-and-teacher account inflation thesis on a US-listed Chinese after-school-tutoring company. Cross-jurisdictional disclosure analysis (PRC operating subsidiary vs US-listed parent) but limited primary-source corporate-filing depth available to outside analysts.

  Outcome. SEC investigation followed publication; share price subsequently declined ~90%, principally on Chinese after-school-tutoring regulatory action (Jul 2021), not the short-thesis-specific allegations. No public enforcement action against GOTU directly.

• Hindenburg · Adani GroupHindenburg Research · 2023 · target: Adani Group (multiple listed entities)

  Offshore-vehicle thesis anchored in Mauritius / Cyprus / UAE corporate-filing analysis. The closest structural parallel to a multi-layer holding-entity diversion thesis: extensive bilateral corporate-filing analysis, light trade-record dependence, no network-layer evidence.

  Outcome. Adani Group market capitalization fell ~$150B in the first six weeks post-publication; SEBI investigation and Supreme Court of India review ongoing at substrate-cutoff. Some disclosure deficiencies confirmed by Indian regulators; broader fraud allegations unresolved.

PeeringDB BGP forensic corroboration

Public IX peering tables disclose which networks are present at which SE-Asian facilities. A direct peering between a Culper-named facility and a Chinese-hyperscaler AS-number is primary-source forensic evidence of network-level proximity — non-narrative-dependent, independent of any Tier-3 analyst interpretation. The True IDC East Bangna ↔ Huawei Cloud (AS136907) finding leads; 12 adjacent disclosures from the same regional cluster + 3 calibrated nulls follow.

Evidence timeline · 7 milestones

Verified-date chronology from the 2023 BIS designation of Inspur Group through the 2026-05-13 Culper publication. Each milestone clicks through to its primary source; authority tier per milestone reflects whether the underlying event is anchored in a Tier-1 filing or a Tier-3 narrative aggregation.

Regulatory levers · statutory authority per reform direction

Four levers, each naming the statutory authority under which reform could proceed and the structural-engineering gap the lever would close. Descriptive of what each lever would mechanically do — not a prediction of legislative likelihood, and not advocacy of any particular reform.

Where the framework ends · explicit out-of-scope

Six exclusions naming where Scrutica’s primary-source-anchored evidence floor stops relative to Culper’s narrative envelope. Each omission is a property of the framework (not a gap in it) — a future workstream may close some, but the floor itself is part of what makes the framework portable to other diversion-thesis cases.

The framework is deliberately narrower than Culper’s narrative envelope. Each omission below reflects a methodological choice about where Scrutica’s primary-source-anchored evidence floor sits relative to Tier-3 analyst material. A future Level-3 workstream may close some of these boundaries as adjacent substrate becomes available; the boundaries themselves are part of the framework, not gaps in it.

• Culper's narrative claims about specific Nvidia executives' personal awareness of the alleged channel, or specific representations made in earnings-call commentary.

  Why omitted. Substrate records only structurally observable corporate-filing topology; personal-knowledge claims would require deposition-level evidence not present in primary-source disclosure. Tier-3 narrative quotation is not Tier-1 fact.

• Culper's employee-interview quoted material — anonymous-source attributions and former-employee characterizations of internal sales practices.

  Why omitted. Single-source, uncorroborated, and not independently verifiable; lowering the page's evidence floor to admit this would weaken the framework's applicability to other diversion-thesis claims that lack adjacent employee-source corroboration.

• Any determination that the disclosed Speedmatrix–Aivres trade flow, or any of the disclosure anomalies, constitutes a specific legal violation under US export-control statutes.

  Why omitted. Scrutica observes structural patterns and renders them at primary-source granularity; legal determination is the proper domain of BIS, DoJ, and adjudicating courts. The page surfaces the *pattern* — not the *adjudication*.

• Causal claims that specific shipments documented in Tradesparq aggregates reached specific Chinese end-customers via specific Megaspeed-financed lease arrangements.

  Why omitted. Tradesparq aggregates document trade flow at the exporter–consignee level (Tier 2); end-customer attribution requires bilateral disclosure at the consignee end (Malaysian SSM filings of Speedmatrix's revenue concentration) that Scrutica has yet to ingest. The aggregate pattern is in scope; the per-shipment causal chain is not.

• Pre-publication NVDA short-interest patterns, options-flow analysis, or any read of the financial-market signal embedded in Culper's timing.

  Why omitted. Outside Scrutica's domain (AI-compute infrastructure intelligence). The framework intersects with diversion-thesis *claims*; analyzing the *trading position* of the entity making the claim belongs to financial-market microstructure research.

• Active network measurement — traceroutes, BGP looking-glass queries, real-time AS-path observation — that would corroborate the PeeringDB-disclosed peering relationships at packet-flow granularity.

  Why omitted. PeeringDB discloses *declared* peering relationships (operator-attested); active measurement would corroborate that the declared peering is *actually carrying* Chinese-hyperscaler traffic at the named facility. Substrate is conservative — declared peering is sufficient for the structural-signal layer; active measurement is a future workstream (Level 3+).

Forward watchlist · structural signature, no published allegation

The framework is extensible: any entity that surfaces the structural-pattern signature — foreign holding chain, opaque parent, AI-compute capacity buildout in a jurisdiction with hyperscale-DC private cross-connect topology — becomes a watchlist candidate. Watchlist status is explicitly NOT allegation; it identifies where the framework’s evidence-rendering surface would apply if a Tier-3 claim were to surface in the future.

Methodology

The Tracker assembles six substrate files into one auditable view of a diversion thesis. Cascade edges come from corporate-filing primary sources (Singapore Bizfile UEN-traceable, Malaysian SSM, Hong Kong corporate register, FactSet entity trees), trade-record secondary sources (Tradesparq aggregations cited through Culper Research), and direct shipment records (single June 2024 Aivres → Speedmatrix shipment reported by the New York Times). Each evidence row carries an authority_tier — Tier 1 (primary disclosure), Tier 2 (research database / derived metric), Tier 3 (analyst report / press), Tier 4 (Scrutica-inferred). The 8 disclosure anomalies are formalized as rows in organization_data_quality_flags via migration 20260605000000 and surface here in feature-page-ready form.

BGP corroboration runs against PeeringDB's public peering tables — the same data ISPs use to coordinate IX interconnection. A direct peering between a Culper-named facility and a Chinese-hyperscaler ASN (Huawei Cloud Global AS136907, Alibaba Aliyun AS45102, China Mobile International AS58453, ByteDance / TikTok AS396986, China Unicom Global AS10099) at a public IX is forensic evidence of network presence that does NOT depend on Tradesparq or analyst narrative. Calibrated nulls (YTL Green DC, Bridge DC, Singapore Equinix flagships) are surfaced explicitly because hyperscale DCs use private cross-connect topology invisible to PeeringDB — absence is neither corroboration nor refutation.

The magnitude studio (Section 1) is built on Scrutica's FLOP Estimation Engine: per-card BF16 PFLOPS from Nvidia spec sheets (Tier 1) multiplied by a chip mix and ASP multiplier (Tier 2 derived metrics), with a model-FLOPs-utilization slider (25–50% band) and a Tier-1-only counterfactual toggle that strips the Tradesparq trade-volume input. Output scales linearly with the trade-volume input; the framework is most useful for understanding the magnitude band, not for asserting point-precision capacity. See the Compute Visibility Index for the parent framework on attributing capacity to specific jurisdictions, and Export Controls for the BIS Entity List designation context behind the Inspur Group rebrand narrative.