METHODOLOGY & DATA PROVENANCE

1 · Reliability Model (S / PF / F)

The launch_events table enforces a nine-state status enum covering every phase of a mission, from scheduled through cancelled. Three of those states describe the outcome of a completed attempt: success, partial_failure, and failure. The distinction between the last two is the core methodological asset of the launch-reliability module.

Why partial_failure matters

A partial_failure is a flight that delivered a payload to orbit but materially below nominal — wrong orbit, degraded mission life, or loss of a recoverable stage after primary separation. A failure is a total loss. Most public trackers collapse these into a single “not success” bucket. They are not the same event for insurance pricing: partial-failures usually drive a partial-loss claim, whereas total-losses drive full hull-and-machinery plus payload claims. A mission-assurance analyst cares too — a partial-failure that still satisfied contract minimums carries a very different program implication than a booster RUD.

Wilson lower bound on small-N variants

Newly-introduced vehicle variants (first ten flights of a new configuration) are scored with a Wilson lower-bound success rate, not a naive fraction. A 3-for-3 start is not “100% reliable,” and the Wilson bound encodes that correctly without requiring editorial override. This matches how an underwriter should price the risk: small-sample new entrants are penalized relative to established fleets with hundreds of demonstrated flights.

Falcon 9 variant handling

Falcon 9 is explicitly split in the reliability query. Flights on or after 2018-05-11 (the Bangabandhu-1 flight, the first Block 5) roll up to the Block 5 variant; earlier flights roll up to Block 1–4. A single pooled Falcon 9 number under-states Block 5 reliability and over-states earlier-block risk. Other vehicles carry vehicle_variant = NULL unless an operator publishes a formally distinct variant line.

Outputs

The reliability tables (vehicle_reliability, operator_reliability) expose total launches, successes, partial-failures, failures, success rate, last-failure date, days-since-failure, current consecutive-success streak, average cadence in days, and crewed-launch count — all cached and rebuilt on every pipeline run.

2 · Space Accessibility Index (SAI)

The AstraVeris Space Accessibility Index is a composite 0–100 score measuring humanity’s progress toward routine space access. It tracks 17 factors across 6 supply-chain tiers, weighted by empirical R² explanatory power and elasticity derived from the 2010–2025 historical series.

Our thesis: space becomes routine when the supply chain is deep enough that no single company or government is a critical bottleneck. Ten new satellite-component manufacturers matter as much as one new rocket — supply-chain depth is what separates a specialist industry from a mature one.

The 17 factors across 6 tiers

Empirical weight derivation

Weights are not set by editorial judgment. For each input/throughput factor we run a bivariate log-log regression against each output factor. The R² gives explanatory power; the coefficient gives elasticity (percent change in output per 1% change in input). Final per-factor weight is the average of the R²-normalized and elasticity-normalized weights, renormalized to sum to 1.0. Factors are also tagged by causal role — input (causes accessibility), throughput (both cause and effect), output (measures outcome) — with role multipliers applied to the raw weights.

Scoring — two-zone log-linear mapping (v3.1)

Each factor is scored 0–100 using a piecewise log-linear mapping. The 0–20 sub-floor band captures pre-commercial and measurement-limited activity; the 20–100 above-floor band captures activity at or above the commercial baseline. Values below the detection threshold (floor × 0.01) still score 0. Above-floor rankings are preserved under an affine transform, so tier composition and per-factor ordering are unchanged by the sub-floor rescaling.

Live vs. static factors

14 of 17 factors are auto-computed from pipeline data on each run. The remaining three — space_workforce, ground_station_sites, and gov_space_budgets — update annually from published reports (BEA, SatNOGS snapshots, NASA/ESA/OECD budget documents) pending automated scrapers. All factors are tagged with a confidence level surfaced through the API and frontend.

Proprietary withholding

The final factor weights, role multipliers, analyst-override envelope, and sub-floor detection thresholds are asserted as SBIR Data Rights. The full audit trail (weights.json) is reproducible by government partners under NDA or SBIR Phase II arrangement.

3 · ARI — AstraVeris Risk Index

The AstraVeris Risk Index is a company-level composite score on a 0–100 scale (higher = lower risk) used for space-insurance underwriting, supply-chain due diligence, and public-market pair analysis. ARI is the closest analog in the stack to a credit score and is the primary driver of the company cards on the public Finance tab.

11 factors across 5 analytical tiers

1. Mission outcomes — launch success rate, partial-failure rate, days-since-last-failure, payload-deployment reliability.
2. Financial — liquidity ratio, debt-to-asset ratio, TTM burn, cash runway in quarters, revenue trajectory.
3. Industrial base — supplier concentration, customer concentration, revenue diversification across government agencies.
4. Capability depth — technology maturity, fleet size, qualified platforms, vehicle variant breadth.
5. Forward pipeline — contract backlog, awarded but not-yet-recognized federal obligations, announced manifest slots, disclosed program milestones.

Category weights in src/finance/ari_scorer.py are 30% Operational, 40% Financial, 30% Market. Factor-level definitions and the stage-prior distributions used for fallbacks are proprietary.

Live-computed vs. stage-prior factors

Per the navbar data audit (2026-04-09, remediated 2026-04-10), 17 of the 20 ARI factors are live-computed from pipeline data on each run. The remaining three fall back to a stage-informed prior because their upstream source is not yet automated: management-stability signals from editorial flow, market-size lookups for nascent sectors, and private-company debt structure where no public filing exists.

Confidence tiers

Every factor emits a FactorResult with one of three confidence tags:

• computed   end-to-end from live data, above the factor’s minimum-N threshold.
• partial   some required inputs present, but a sector median or stage prior was substituted for one or more sub-inputs.
• default   no live data; factor set to a stage-informed prior (seed / Series A / B / growth / public) and flagged.

The composite is confidence-weighted: default factors contribute less than computed factors so that thinly-sourced companies do not drag the average toward the midpoint and mask the signal in well-sourced peers.

Wilson bounds on small-N mission-outcome rates

Several mission-outcome factors are empirical success rates on small samples (a new vehicle variant with a handful of flights, or a young operator with fewer than ten payloads deployed). AstraVeris computes the Wilson score lower bound at a fixed confidence level for every rate estimate below the minimum-N threshold and uses the lower bound (not the point estimate) as the factor input. This is a deliberate conservatism.

Output surface — what ARI emits per company

• Composite score 0–100 with discrete risk label (Very Low / Low-Moderate / Moderate / High / Very High) and color.
• Per-factor breakdown with score, confidence, and human-readable rationale.
• Confidence distribution (share of composite from computed vs. partial vs. default).
• Data-lag indicator — days since the most recent input refreshed for this company, surfaced on the Finance cards so users can see exactly how current each score is.

4 · Debt Maturity Wall

The Debt Maturity Wall is the newest underwriting surface in the AstraVeris stack. For every tracked public space operator, the pipeline extracts the schedule of outstanding debt instruments from SEC filings, surfaces maturity dates and coupon rates, and flags companies whose principal-repayment schedule is compressed relative to their cash runway.

Extraction pipeline

• EDGAR ingest — 10-K, 10-Q, and 8-K filings for tracked issuers are pulled from the SEC EDGAR full-text feed.
• Gemma-based structured extraction — the local Gemma 4 model (26B MoE 4-bit via Ollama) parses long-form debt-and-liquidity disclosures into structured fields. All inference runs locally; no third-party LLM API is called for this surface.
• XBRL cross-check — where the filing carries tagged XBRL facts, the extractor reconciles Gemma’s output against the structured fact to catch misreads.

Captured fields

Runway-pressure flag

The runway_pressure_flag is a derived signal, not a credit opinion: it fires when principal maturing in the next 18 months exceeds current cash & equivalents reported in the same filing. It is intended to surface refinancing candidates for analyst review, not to predict default. Issuers with live ATM programs or active revolvers will frequently carry this flag without distress; the flag is a starting point, not a verdict.

5 · Data Sources & Licensing

AstraVeris is built on a deliberately narrow set of high-quality open-data sources. Source coverage is additive — the platform is designed so that no single vendor outage degrades more than one factor of any index.

Every record carries a source_id and a confidence tag so downstream factor computations can distinguish reported facts from derived or defaulted values.

6 · Export Control / ITAR Posture

AstraVeris ingests only public-source data. The product is classified EAR99. No ITAR-controlled inputs are used, stored, or processed. AstraVeris is U.S.-owned and U.S.-operated. If a customer engagement introduces CUI (Controlled Unclassified Information), we team with a CMMC Level 2 prime rather than taking custody of controlled data on our own infrastructure.

This is a deliberate architectural choice: staying on the EAR99 / public-source side of the line means AstraVeris analytics can be shared across allied-partner settings and commercial underwriting contexts without export-license friction. When a government partner needs analytics that cross into controlled territory, we participate as a CRADA / OTA / SBIR Phase II subcontractor under the prime’s CMMC envelope.

7 · Update Cadence & SLA

Every factor result and company card carries a data_lag_days value measuring days since the newest input record refreshed. It is surfaced to the frontend so users can see exactly how current a given score is.

8 · Data Transparency Tags

Every data point in the AstraVeris stack carries a confidence tag that is persisted end-to-end — from the scraper, through the processors and scorers, to the API response and the UI badge. Tags are surfaced as pills on each public page so a reader never has to guess whether a number is a reported fact or a modeled fallback.

SAI factors

ARI factors

9 · Not Investment Advice

AstraVeris publishes analytical metrics, index values, and editorial commentary for informational purposes only. Nothing on this site is investment advice, a solicitation to buy or sell any security, a credit rating, or a recommendation. AstraVeris is not a Nationally Recognized Statistical Rating Organization (NRSRO), an insurance broker, a registered investment adviser, or a licensed underwriter. Scores and analytics reflect publicly available data and proprietary modeling and may contain errors, omissions, or latency. Readers should conduct their own due diligence and, where appropriate, consult a qualified professional before making any financial, underwriting, or procurement decision based on AstraVeris output.

10 · Changelog

v1.0 — 2026-04-23 — initial publication.