How DEPLOY composes the three-axis actuarial substrate

What the three-axis substrate is

Actuarial analysis of physical AI requires three substrates, not one. Incident counts without exposure data over-state risk. Exposure data without financial counterparty context mis-prices counterparty risk. Both without component-level failure mode mapping flatten supply-chain risk into a single maker layer.

We compose the substrate from three axes. Each axis carries its own verification posture; each axis's cap-flag triggers operate independently; each axis ships as a framework-in- action narrative on the editorial surface. The substrate as a whole composes when the three axes are read together.

Axis 1: incident outcome and deployment exposure (the numerator)

The first axis is incidents and exposure. Incident outcome class names the per-incident severity at primary-source verification depth (FDA Class I, Class II, Class 3; NHTSA recall scope; NTSB findings). Deployment exposure_hours names the operational substrate against which the incidents are counted: hours of operation, miles of operation, procedures performed, totes handled. The two together produce the numerator-and-denominator for per-unit-of-exposure incident rates.

The canonical worked example is the Phase 3 Dim 1 actuarial substrate. Incident outcome class is verified at FDA primary- government-record source-quality across the surgical cohort; NHTSA primary-source across the AV cohort. Deployment exposure_hours is verified where the operator or customer of record discloses; honest-absence cap-flag at 128 entries where the column remains null because no primary source has surfaced exposure data.

The 128-nulls pattern is canonical. We do not estimate exposure_hours from announcement-stage funding or contract framing. The null is the verified state for those 128 records. The substrate ships honestly at full-population scale; the verification posture is honest-absence at the column layer, not at the record layer.

For the canonical worked example at narrative depth, see How DEPLOY tracks incident outcome_class and deployment exposure_hours.

Axis 2: manufacturer financial state (the denominator)

The second axis is manufacturer financial state. Revenue, revenue_basis, lifecycle stage, counterparty risk class, going-concern flag, and cash-runway-and-basis fields together describe the financial counterparty context against which the incident-rate numerator is read. A 1-in-100,000 incident rate on a manufacturer at counterparty-low parent-backing reads differently from the same rate on a growth-stage private at counterparty-moderate runway depth.

The canonical worked example is the Phase 3 Dim 2 manufacturer financial state substrate. Three SEC-anchored mature public-filers (Intuitive Surgical FY2024 $8.35B per 10-K; Stryker nine-month 2024 $16.16B per 10-Q at narrower- precision cap-flag; Zimmer Biomet FY2024 $7.68B per 8-K). One parent-backed mature private (Boston Dynamics counterparty-low via Hyundai approximately 80% parent- backing). Three growth-stage privates (Apptronik plus Figure AI plus 1X Technologies at counterparty-moderate via funding- stage proxy; revenue and cash-runway honest-absence across all three).

The Stryker narrower-precision cap-flag pattern is canonical. The verified figure is nine-month per the Q3 10-Q; the full- year is not yet captured. We use the SEC-verified narrower- precision figure rather than extrapolate. The cap-flag is explicit; the verified posture stays.

The Boston Dynamics counterparty-risk-by-parent-backing pattern is canonical. Standalone revenue is honest-absence; counterparty-low is verified at parent-backing depth.

For the canonical worked example at narrative depth, see How DEPLOY tracks manufacturer counterparty risk at financial-state-axis granularity.

Axis 3: supply-chain component failure mode (the scalar selectivity)

The third axis is supply-chain component failure mode. Each incident on Axis 1 chains to a component and to a failure mode class on the component. The third axis selects scalar: which component, which failure-mode class, which severity, which remediation pattern. The selection refines the incident-and-exposure picture from the maker layer to the component-and-supplier layer.

The canonical worked example is the Phase 3 Dim 3 supply- chain component failure mode substrate. Five safety-critical components across the surgical cohort. Six failure modes chained to six real recall incidents via the component_failure_incidents junction. Software_fault concentrates at three of six substrate rows (50% of the surgical-cluster Dim 3 substrate), confirming the cross- cohort software-defect root cause class observable at incident-layer in Phase 3 Dim 1 and at component-layer here in Phase 3 Dim 3.

Source-inheritance is canonical. Each component_failure_mode inherits the linked recall incident's primary source as its evidence. The recall that attests the failure attests the failure mode. Zero new-URL fabrication; the third axis composes against verified evidence already anchored on Axis 1.

For the canonical worked example at narrative depth, see How DEPLOY tracks supply-chain component failure modes chained to recall incidents.

How the three axes compose

The three axes compose into a single actuarial view per maker per deployment per component. A reader looking at a specific deployment encounters Axis 1 (the incidents that have fired on the deployment; the operational substrate against which they fired), Axis 2 (the manufacturer's financial state and counterparty risk at the deployment date), and Axis 3 (the components implicated in the incidents and their failure mode classes).

The composition is read at the lowest-verification-depth axis. An incident-rate calculation against an Axis 1 numerator that is verified plus an Axis 1 denominator at honest-absence resolves at honest-absence on the rate. We do not infer a rate when the denominator is honest-absence; the rate is itself honest-absence. Aggregator coverage frequently publishes per-unit incident rates against estimated denominators that no primary source supports; the framework rejects the extrapolation.

The same composition discipline applies to counterparty risk. Counterparty-low on a manufacturer at SEC-anchored Tier 1 financial state plus Axis 1 incident outcomes at FDA Tier 1 primary government records resolves at verified composition. The same counterparty-low classification on a manufacturer at private_reported with revenue honest-absence plus Axis 1 incident outcomes at FDA Tier 1 resolves at cap-flagged composition because the financial-state axis carries honest- absence on a substantive field.

Verification posture per axis

Each axis carries its own verification posture, set at the per-claim layer per the discipline at /methodology/verification-posture.

Axis 1 verification posture sits primarily at Tier 1 primary- government-record source-quality. FDA recall records, NHTSA Part 573 records, NTSB findings, ITC dockets anchor incident outcome class. Deployment exposure_hours sits at counterparty- disclosed Tier 2 when the customer of record discloses; at maker-disclosed Tier 4 or Tier 5 when only the maker discloses; at honest-absence when no source surfaces.

Axis 2 verification posture sits primarily at verified_ir posture per the verification-posture taxonomy. SEC 10-K, 10-Q, 8-K filings anchor revenue and lifecycle stage for public filers. Private_reported records anchor at Tier 4 or Tier 5 maker-disclosed source-quality; counterparty risk classifies via funding-stage proxy or parent-backing depth where applicable.

Axis 3 verification posture inherits from Axis 1. The component_failure_mode record inherits the linked recall incident's primary source. The verification posture on Axis 3 is structurally coupled to Axis 1's posture; a software_fault failure mode on a recall verified at FDA Tier 1 source- quality is itself at Tier 1 source-quality through inheritance.

Cap-flag uniformity across the three axes

The cap-flag construct operates uniformly across all three axes. Honest-absence on Axis 1 deployment exposure_hours (the canonical 128-nulls pattern). Honest-absence on Axis 2 revenue_disclosed_usd plus cash_runway_months for private makers (validator-aware null-when-figure-null discipline). Honest-absence on Axis 3 failure-mode-class=other when the root cause stays under investigation at the primary source. The cap-flag flows through composition; an axis at honest- absence resolves the composed view at honest-absence on the dependent claim.

The transparency is the brand position. See /methodology/cap-flag-as-trust-signal for the standalone treatment of cap-flag as the published verification state and the honesty-as-strength editorial framing the three-axis substrate operates under.

The framework applies to us

When a primary source surfaces an exposure-hours figure for a deployment that previously sat at honest-absence, we add the figure with the source attribution; the column moves from null to the verified value. When a manufacturer files a full-year 10-K that resolves the nine-month narrower- precision cap-flag, we update the revenue field; the cap- flag is removed; the prior cap-flag state stays in the source history.

The composition shifts as the underlying axes shift. The three-axis substrate is not static. We refresh the substrate on a recurring cycle; new primary-source disclosures resolve cap-flagged claims; new corrections re-anchor records that resolved against incorrect prior sources.

The corrections are logged at /corrections. The substrate evolution is preserved.

Where to go next

The framework-in-action narratives that anchor each axis: Phase 3 Dim 1 incident outcome and deployment exposure, Phase 3 Dim 2 manufacturer counterparty risk, Phase 3 Dim 3 supply-chain component failure modes chained to recall incidents.

The methodology canon supporting the meta-pillar: /methodology/what-verified-means, /methodology/verification-posture, /methodology/cap-flag-as-trust-signal.

The Project B methodology canonical reference, where the three-dimension actuarial framework operationalizes alongside Dim 4 (cross-jurisdictional regulatory clearance state) and related Project B essays: How DEPLOY thinks about robot insurance.