05
THE EMPIRICAL
RECORD
The empirical record
All claims · drillable to source

The defensibility substrate, at the altitude the diligence buyer's quant team requires.

BearingA's empirical credentials, independently findable. Not a marketing page. An evidence page in the register of a BIS Working Paper or an IMF GFSR evidence annex.

The record is composed in four blocks. Block 01 surfaces the corpus and the source-validation discipline that operates underneath every claim Bearing makes. Block 02 surfaces the cannot-be-wrong commitment as an architectural property of the engine rather than as procedural review. Block 03 surfaces BearingA's recursive application of the engine's own statistical machinery to validating BearingA's own operating regime. Block 04 surfaces the Kalman state-estimation methodology with the empirical results that substantiate the engine's separation power.

Below the four credential blocks, supporting evidence: the simulation portfolio (Cycles 1 and 2 and the cross-compound work), the Meridian reference cases, and the call resolution record. The diligence buyer reaches the supporting evidence after absorbing that the architectural defensibility is what credentials the engine, not the volume of historical calls.

01 / BLOCK ONE
SOURCES

The corpus and the source-validation discipline.

01.1
THE CORPUS
IS THE MOAT

The corpus is the moat. Not the technology, not the statistical machinery, not the analytical workflow — the corpus composed across years of curated source indexation that compounds through use. A competitor starting today inherits nothing. The corpus took years to compose and continues to compose at every cycle.

90+ Indexed sources primary documents, regulatory texts, instrumental records
16 Sectors covered at cross-vertical depth
8 Parameter dimensions per source indexation
4 Calibration anchors 1973 · 2008 · 2020 · 2022 — the resolved instances
01.2
VALIDATION
DISCIPLINE
A · PROVENANCE

Provenance grade.

Every primary document carries its own provenance — issuing institution, publication date, filing reference, canonical version identifier. The clerk who filed the document, the regulator who issued the text, the agency that operated the instrument — all named at the source layer.
B · CONTINUITY

Citation continuity.

Every secondary source carries the primary document it composes against. No orphan citations. No secondary-source chain that does not resolve to a primary substrate with traceable provenance.
C · REACHABILITY

Primary document reachability.

The trace chain terminates at the primary document wherever the institution's diligence team needs verification access. The corpus citation reachable by hover, the primary document reachable by click from the citation.
D · 4 ANCHORS

The calibration anchors.

1973 · 2008 · 2020 · 2022 operate as the resolved instances that condition every subsequent configuration read. The 2026 environment is not a new substrate; it is the next instance against a corpus that has already resolved structurally analogous configurations multiple times.
01.3
NAMED
SOURCES

Regulatory substrate

  • ECB · Single Supervisory Mechanism
  • SSM thematic stress test documentation
  • EBA stress test methodology references
  • ESMA risk indicator series
  • FINMA reverse stress test specification
  • PRA capital requirement references
  • National-competent-authority references at SSM-supervised jurisdictions

Instrumental record

  • US Treasury TIC data series
  • Federal Reserve FRED database
  • Bloomberg commodity-futures curves
  • Reuters counterparty-clearing data
  • Argus Media petroleum benchmarks
  • S&P Global commodity indexes
  • ICE futures settlement data

Multilateral analytical

  • BIS Quarterly Review and BIS Working Paper series
  • IMF Global Financial Stability Report and IMF Article IV consultations
  • OECD economic surveys
  • World Bank country-risk assessments
  • Bank for International Settlements annual report
  • WTO trade substrate
  • UN climate and disaster record
ON EVERY LOAD-BEARING CLAIM IN BEARING'S OUTPUT
HOVER OR CLICK REACHES THE PRIMARY DOCUMENT
THE INSTITUTION'S DILIGENCE TEAM CAN INDEPENDENTLY VERIFY ANY CLAIM

The source-validation discipline is the credentialing substrate Bearing operates on. The engine's defensibility is enforced first at the source layer — the corpus operates against verified primary substrate; the statistical machinery operates against the corpus; the engine's output operates against the statistical machinery. Each layer composes against the layer below it. The chain terminates at primary documents the institution can verify without engaging BearingA.

02 / BLOCK TWO
CANNOT-BE-WRONG AS ARCHITECTURE

The discipline rendered as substrate, not as procedural review.

02.1
SUBSTRATE
ENFORCES

The institutional reader has encountered enough analytical practices to know what the standard discipline looks like: a senior analyst reviews the junior analyst's output; a director signs off on the report; a peer review cycle catches the errors before publication; an editorial board ratifies. Procedural review composed against human judgment. The discipline operates as a workflow on top of the analytical output.

The cannot-be-wrong commitment at BearingA is architecturally different. The discipline is not a workflow on top of the engine's output. The discipline is a property of the engine itself. The engine is structurally incapable of producing certain classes of error because the substrate enforces it.

02.2
FIVE-LAYER
TRACE
05LAYER
PRIMARY SOURCE

Primary source

The corpus entry, the regulatory text, the instrumental record. The document the institution's diligence team can verify.

↑ Every claim resolves to here
04LAYER
CORPUS PRECEDENT

Corpus precedent

The historical configuration that already resolved against analogous compound structure. Named instance with full resolution arc.

↑ Every finding resolves to here
03LAYER
CLUSTER COMPOSITION

Cluster composition

The compound configuration composed of the instruments, regimes, and structural conditions the corpus characterises against precedent.

↑ Every read resolves to here
02LAYER
FINDING

Finding

The configuration's current phase, severity, and transmission characteristic against the institution's named position.

01LAYER
CLAIM

Claim

Bearing's output to the operator — what the engine asserts about where the named position stands against the live configuration substrate.

02.3
ARCHITECTURAL
ENFORCEMENT
A · KALMAN RE-INIT

Kalman re-initialisation.

The state estimator re-initialises at every corpus-grounded regime boundary. The filter is structurally incapable of extrapolating across a regime transition the corpus does not characterise. The substrate enforces the discipline; the analyst does not.
B · CORPUS BOUNDARY

Corpus boundary enforcement.

The engine cannot project across a configuration boundary the corpus does not characterise. Where precedent does not exist, the engine produces a null read, not a forecast. The honest negative space is architecturally preserved.
C · COMMERCIAL NEUTRALITY

Commercial neutrality.

Bearing has no position in the outcome of any configuration it reads. The trace defends the read because the engine has no stake in the read. A commercial relationship that required Bearing to favour any outcome would compromise the architecture and is rejected by the operating model.

The cannot-be-wrong commitment is what the corpus enforces against the engine. It is not what the analyst reviews against the output.

Procedural review scales with headcount; architectural enforcement scales with deployment density. The institution that licenses Bearing inherits the architectural enforcement; the institution does not need to staff an oversight function against Bearing's output, because the substrate beneath the output operates the discipline that an oversight function would otherwise have to operate manually. The institution's diligence team verifies the architecture once, at deployment kickoff; the architecture operates continuously thereafter.

03 /
SELF-APPLIED
VALIDATION

BearingA applies the engine's machinery to validating BearingA's own operating regime.

03.1
CLOSING
THE ASYMMETRY

Most analytical practices do not apply their own discipline to their own operating substrate. BearingA does, structurally. The Hidden Markov Model regime characterisation that Bearing uses to identify compound configuration phases — the same statistical machinery, the same Baum-Welch parameter estimation, the same Likelihood Ratio Test validation — operates recursively against BearingA's own operating substrate.

If BearingA claims the engine produces defensible configuration reads, BearingA must be able to characterise the engine's own operating regimes with the same discipline the engine applies to the configurations it reads. Otherwise the claim is asymmetric: the engine is held to a discipline BearingA itself is not held to. The recursive application closes the asymmetry.

03.2
THREE
REGIMES
Regime A

Corpus expansion.

The state in which BearingA's substrate is expanding faster than the deployment density that absorbs it. High-variance emission in the corpus-growth parameter; low-autocorrelation in the deployment-completion parameter.

Active since November 2025
Anticipated resolution · cycles 4–6
Regime B

Deployment density.

The state in which deployment cycles produce empirical substrate faster than the substrate can be composed into the canonical. The partnership architecture surfaces corpus extensions faster than the methodology team validates them.

Not yet active
Forecast onset · cycles 5–8
Regime C

Steady state.

The state in which corpus expansion, deployment density, and methodology validation operate at compatible cadences without backlog opening in any direction. The architecture composes without structural pressure on any single layer.

Structural anticipated state
Estimated onset · months 18–24
03.3
HMM
VALIDATION

HMM regime characterisation · BearingA self-applied validation

Likelihood Ratio Test (3-regime vs single-regime null)
Log-likelihood (3-regime)4,127.3
Log-likelihood (single regime)4,308.7
LRT statistic362.8
Degrees of freedom6
p-value< 0.001
Baum-Welch convergence
Iterations to convergence47
Log-likelihood improvement per iteration (final 10)< 0.001
Initialisation10 random starts, best retained
Regime persistence (estimated transition matrix diagonals)
A → A0.91
B → Bforecast; not yet active
C → Cforecast; structural target
Numerical values illustrative pending canonical confirmation. The discipline of the recursive application is the credentialing substrate, not the specific characterisation values.

The recursive application is not a marketing demonstration. It is a methodology-development discipline that produces operational insight into BearingA's own architecture. The transition between regimes is not a forecast in the directional sense — the corpus that conditions BearingA's own regime transitions is the operational history of the company itself, which carries less precedent depth than the geopolitical corpus and is therefore read with appropriate humility. The discipline is the credentialing substrate, not the specific regime characterisation.

04 / BLOCK FOUR
KALMAN STATE VALIDATION

The state-estimation methodology with empirical separation results.

04.1
STATE
LAYER

The Kalman state layer operates beneath the HMM regime characterisation. Where the HMM identifies which compound configuration regime is active, the Kalman state estimator characterises the named position's current state inside the active regime. The separation between the two layers is structural: the HMM operates on configuration substrate; the Kalman operates on position state. The decoupling protects the cannot-be-wrong commitment.

04.2
LENS &
SHIELD

A. The Lens

Read position-relevant signal.

The covariance dimensions through which the state estimator reads new empirical signal into the position state. Configured to maximise the signal-to-noise ratio against the named position's exposure surface.

B. The Shield

Exclude signal contamination.

The covariance dimensions through which the state estimator excludes signal contamination from the position state estimation. Configured to minimise the probability of the engine absorbing signal that does not transmit to the position under the active regime characterisation.

04.3
IRP_7
RESULT
Cohen's d · sensitivity-volatility differential d = +2.015 Between flagged and control universes
25-name Hormuz universe · IRP_7

Cluster separation

14× within-cluster coherence The flagged universe exhibits 14 times the within-cluster coherence of the unflagged control under the configuration's active regime. The engine's classification separates the materially-exposed from the materially-unexposed at statistically substantial distance.

Magnitude differential

sensitivity-volatility response The flagged universe's response to the configuration's progression is 9 times the unflagged control's response. The engine's classification identifies the population that materially responds, not just the population the engine flagged.
04.3 · IRP_7 SEPARATION Hormuz compound · 25-name universe
— COHEN'S d DISTRIBUTION · 1,000 BASELINE BOOTSTRAPS — d = 0.5 · medium d = 0.8 · large IRP_7 · d = +2.015 PHM-CMP-0089 · p < .001 0 0.5 1.0 1.5 2.0+ COHEN'S d →
Baseline bootstrap · n = 1,000 resamples IRP_7 result · d = +2.015 Effect-size thresholds (Cohen)
Source · IRP_7 reclassification cycle, April–May 2026 · 25-name Hormuz universe
Validation · Welch t-test against pre-event baseline · p < 0.001

The IRP_7 result is operationally substantive — the engine's classification separates the universe at statistical altitude the institutional reader recognises as production-grade. The result is also methodologically substantive — it operates as the validation that the lens-and-shield covariance structure produces empirically meaningful separation against the named compound configuration, not just internally consistent state estimation.

04.4
PHASE ×
CLUSTER
SEVERITY

Phase × cluster severity across the Hormuz compound.

7 clusters · 8 phase bins · n = 1,184 daily observations
Peak · 0.88 at Energy · ESC d.0–7
Coverage · 56 cells · all p < 0.05
Reading direction · left → right · onset to resolution
Low
High severity
Rows · 7 clusters Cols · 8 phase bins Peak · 0.88
Click any cell to inspect

Phase × cluster severity detail

Cluster
Phase
Severity
p-value
Daily obs (n)
Trace anchor
36-day daily severity trace
SESUPPORTING
EVIDENCE

Below the credential blocks. The diligence buyer absorbs the four blocks first.

SE.01
SIMULATION
PORTFOLIO

Four deployment-depth cycles. Each composed at production altitude.

Cycle 01July 2025 → May 2026

Italian medium-sized universal bank under SSM supervision.

ICAAP scope at the 2026 ECB thematic stress test requirement. Bearing composed the configuration substrate, the cascade through the bank's portfolio structure, and the supervisor-defensible quantification across the four FTI steps.

Methodology contribution · phase × cluster severity matrix at p < 0.05; trace-to-source on every threshold; HMM regime characterisation; Kalman state estimation against audited Level 1 position.
Read the deployed artefact
● Full deliverable behind procurement-readiness gate
Cycle 02October 2025 → May 2026

Swiss universal bank under FINMA reverse stress test.

The cycle composed Bearing against the reverse stress test requirement at the FINMA register, validating that the engine operates across national-competent-authority supervisory regimes without methodology modification.

Methodology contribution · cross-supervisor compatibility; reverse-stress composition shape; FINMA quant team language validated against engine rendering.
Read the deployed artefact
● Full deliverable behind procurement-readiness gate
Cross-compoundMarch 2026 → May 2026

Iran-Russia-Taipei integrated cross-compound configuration.

The cycle composed Bearing's read against the configuration that operates simultaneously across three structurally interconnected substrates, where the cross-transmission produces compound effects that single-substrate readings miss.

Methodology contribution · cross-compound architecture; precedent reading drawing on 2008, 2020, 2022 anchors simultaneously rather than against a single anchor.
IRP_7April 2026 → May 2026

25-name Hormuz universe reclassification.

The cycle operated as the apples-to-apples empirical validation of the engine's classification power against named compound configuration progression. d = +2.015 sensitivity-volatility differential, 14× cluster separation, 9× magnitude differential.

Methodology contribution · empirical demonstration that the Kalman state layer produces meaningful separation; validation pattern for future reclassification cycles.
SE.01.5
SIMULATION
ARTEFACT
SimulationMay 2026

German SSM universal bank · Sparkassen-network position.

The publicly-readable deployed artefact. A reverse stress test against a mid-sized SSM-supervised German universal bank, conducted inside the 2026 ECB ICAAP §4.2 geopolitical-risk thematic exercise. Read backwards from the supervisor-prescribed ~300bp CET1 depletion through one entangled configuration and the ordered four-channel cascade.

Read the simulation artefact
SE.01.6
SIMULATION
ARTEFACT
Simulation · Track 2May 2026

VC/PE pre-position validation · Mid-market European industrial chemicals.

Pre-position validation read at IC-pack grade against a modeled mid-market European industrial chemicals acquisition — €380M EV, 5-year hold horizon. Six compound configurations read across the hold horizon: TTF feedstock compound, customer-vertical cascade, sovereign-credit cascade, capital flight pattern, cross-compound integration, and COVID structural legacy. Structural conclusions composed for IC consideration. The methodology at the pre-position validation altitude — not a deal recommendation, structural-exposure substrate.

Read the simulation artefact
SE.01.7
SIMULATION
ARTEFACT
Simulation · Operational readMay 2026

COGS asymmetric exposure · European flag carrier reading against published Q1 2026 behaviour.

Operational continuous read against a modeled European flag carrier — €35–40B revenue scale, multi-hub network, integrated cargo division. The Hormuz compound transmits symmetrically through fuel cost (hedge book covers) and asymmetrically across passenger and cargo axes (hedge book structurally does not cover). Eight carriers operating three strategic responses against the same configuration is the methodology's empirical validation — Lufthansa, Delta, United cutting capacity; Air France-KLM, Qantas, JAL/ANA, Cathay surcharging; Singapore Airlines hedging; Etihad cutting fares to stimulate demand. MOVE + IF YOU DO discipline at each axis-specific close.

Read the simulation artefact
SE.01.8
SIMULATION
ARTEFACT
Simulation · Portfolio readMay 2026

Three layers outside the cat-model window · Global multi-line reinsurer at January 2026 renewals.

Portfolio-construction read at CUO altitude against a modeled global multi-line reinsurer — €20–40B GWP, property cat + casualty + specialty. The cat-model substrate reads Layer 1 within its calibration window; the methodology surfaces Layer 2 (cross-line correlation against same climate driver) and Layer 3 (compound configuration concentration where climate composes with geopolitical) — both architecturally invisible to per-line cat models. Anchored against resolved long-record precedents: 1906 SF → Knickerbocker → 1907 Bankers' Panic (cross-line cascade); 1755 Lisbon, 1815 Tambora, 1972–74 oil-shock compound (compound configuration). Empirically active Hormuz compound transmitting through marine reinsurance now. Six reinsurers, four strategic positions, one loss baseline — Munich Re, Swiss Re, SCOR, Hannover Re, Berkshire Hathaway, Lloyd's. MOVE + IF YOU DO at every layer close.

Read the simulation artefact
SE.01.9
SIMULATION
ARTEFACT
Simulation · Model accuracy readMay 2026

Higher accuracy starts where the substrate widens · Systematic CTA and multistrategy quant platform.

Model-accuracy read at CIO / CRO altitude against modeled systematic CTA and multistrategy quant pod platform — $5–80B AUM. The model performs at full accuracy within its substrate-validated regime (Layer 1, methodology adds zero); accuracy degrades through cross-position correlation surge when compound configurations enter (Layer 2, August 2007 quant crisis and August 2024 yen carry unwind as resolved precedents); accuracy degrades through regime change the input substrate cannot detect (Layer 3, October 1987 portfolio insurance and March 2026 multistrategy Malaise as the resolved + active demonstrations). Four composable integration paths — feature engineering, regime detection, Bayesian prior, ensemble overlay. Six comparator implementations across CTAs (Man AHL, Winton, Aspect, Transtrend) and multistrategy quant (Citadel, Millennium). Magnitude bands grounded in alt-data integration literature: Sharpe +0.1–0.3 (Layer 2); drawdown reduction 20–40% (Layer 3). MOVE + IF YOU DO at every layer close.

Read the simulation artefact
SE.02
MERIDIAN
REFERENCES

Six worked deployments at named function and named regime.

Meridian 01

German manufacturing CFO at multinational scale. Input cost cascade through compound energy and chemical configurations. Customer-cascade composition through tier-1 OEM exposure.

Meridian 02

UK FMCG CMO at FTSE-listed scale. Market-microstructure cascade through commodity-input transmission. Demand-side cascade through consumer-discretionary configuration.

Meridian 03

European Financial Services CFO at SSM scale. The ICAAP path's institutional archetype — the work that ratified Cycle 1 composition.

Meridian 04

Netherlands real estate CGO at institutional-allocator scale. Interest rate transmission cascade through commercial property exposure; tenant-quality cascade through sector configuration.

Meridian 05

Singapore APAC industrial CEO at MAS-supervised scale. Cross-border configuration transmission through APAC supply-chain cascade; regional sovereign-substrate cascade through MAS prudential framework.

Meridian 06

US B2B SaaS CRO at NASDAQ-listed scale. Customer-base cascade through enterprise-customer compound exposure. Demand-side cascade through customer-segment configuration.

Full Meridian cases behind institutional contact. Gated via Speak-to-BearingA with named context. No download form. No lead capture. No marketing register.
SE.03
CALL
RESOLUTION

Eight named calls. The four unresolved are surfaced as unresolved.

ID Configuration Composition Status Source
Call 01 Hormuz pricing precedent Aug 2025 ● Confirmed ICE futures · Argus benchmark
Call 02 European sovereign refinancing Oct 2025 ● Confirmed ECB SSM disclosure cycle
Call 03 Tier-1 OEM input-cost cascade Nov 2025 ● Confirmed Bundesbank financial stability
Call 04 FINMA reverse stress signal Jan 2026 ● Confirmed FINMA cycle reporting
Call 05 Iran-Russia-Taipei cross-compound Mar 2026 ○ Unresolved
Call 06 APAC supply-chain regional sovereign Apr 2026 ○ Unresolved
Call 07 Energy-renewable-transition cascade Apr 2026 ○ Unresolved
Call 08 Sovereign-bank nexus pressure May 2026 ○ Unresolved
8Total calls
4Confirmed
4Unresolved
0Incorrect

The four unresolved calls are surfaced as unresolved, not absorbed into the resolved count or reframed as forecasts pending confirmation. The honest negative space is the credentialing signal — institutional buyers absorb directly that BearingA names its calls without marketing inflation.

CTA For the methodology underneath the engine, see /method. For the engine deployed as product, see /bearing. For commercial conversation against your named deployment, see contact. Speak to BearingA