Methodology

FAIR share of housing = residential (dwelling) policies in force ÷ ACS total housing units for the matching ZCTA. Total units — not owner-occupied — because FAIR dwelling policies also cover rentals and second homes. The ratio understates penetration slightly in renter-heavy ZIPs and is the most comparable single number across the state.

• FAIR data: 5-digit ZIP is the floor. No address-, parcel-, or sub-ZIP data exists publicly; a specific property's insurance status is confidential. ZIPs are roughly neighborhood-scale in cities (Pacific Palisades ≈ 90272) — that is the honest limit.
• Voluntary-market data: county is the floor (CDI fact sheets). A ZIP-level FAIR / private / uninsured split does not exist publicly; “uninsured” can only ever be derived (housing − FAIR − voluntary), never observed.
• ZIP ↔ ZCTA: FAIR ZIPs join Census ZCTA polygons directly. PO-box-only ZIPs (and a few non-CA artifacts in FAIR's own files) have no polygon and drop off the map — 7 of 1,683 ZIPs, all with trivial policy counts.

• Growth is recomputed from values. The published YoY column in FAIR's PDFs reports 0% on zero or missing bases; we compute growth from the underlying counts and show nothing when there is no meaningful base.
• Premium per policy / per $1,000 insured are computed from the quarterly premium and exposure files (a different vintage than the 5-year panel — both are labeled where shown).
• Share of housing is capped at 100%. A handful of seasonal-resort ZIPs show more residential FAIR policies than counted housing units (e.g. 32 policies against 15 ACS units). The cause is the denominator, not the numerator: FAIR reports by postal ZIP — a mail service area — while the housing count covers the Census ZCTA polygon, which can exclude outlying cabins the ZIP serves; owner and renter policies on the same unit add a small double-count. The true share in these ZIPs is high but not precisely measurable, so we cap the displayed metric at 100% and flag affected ZIPs (†) wherever the number appears.
• PDF extraction quirks are handled, not ignored: the premium and exposure PDFs use a font whose thin glyphs (commas, i, l) vanish in text extraction; amounts are recovered by dollar-sign grouping and county names are repaired against the canonical 58. Every parsed file is validated against its own printed total row.

The divergence lens compares two independent views of the same ZIP: FAIR's own classification (the share of its policies it places in the High wildfire-risk band, expressed as a statewide percentile) and FEMA's National Risk Index wildfire score (tract-level, aggregated to ZCTA weighted by tract population × area overlap, expressed as a statewide percentile). Divergence = FAIR percentile − FEMA percentile. Caveats: NRI measures expected loss to the whole community, not the insured book; its vintage (v1.20, Dec 2025) reflects modeling that largely predates the 2025 fires; and tract→ZCTA aggregation smooths within-ZIP variation. Divergence flags places worth examining — it does not by itself prove mispricing.

No dataset reports underinsurance directly. We triangulate three proxies: (1) coverage ratio — average FAIR insured value per residential policy ÷ Zillow typical home value. ZHVI is market value (land + structure) while insurance covers structure replacement cost, so the ratio is directional, not a measurement; (2) cap-binding ZIPs — where typical home value exceeds the $3M FAIR limit (unchanged since April 2020), underinsurance is structural; (3) published survey anchors — United Policyholders' year-one LA survey (69% of total-loss respondents underinsured, average gap $247/sq ft) and the peer-reviewed Marshall Fire benchmark (74% underinsured, 36% severely). FAIR-specific drivers (actual-cash-value default, $3M cap) mean FAIR policyholders are structurally more exposed than the admitted market.

Scenario footprints are documented fire-prone clusters defined as whole-ZIP sets — they are illustrations, not predictions. Calibration: the January 2025 fires produced ~$4.05B of FAIR gross loss (FAIR figures, Feb 2025) against the pre-fire (FY2024) FAIR exposure of the affected corridor ZIPs — a corridor-wide ratio of ~37%, applied whole-ZIP-to-whole-ZIP. (Inside the perimeters the realized ratio was ~85%, 2.7× FAIR's own ~31% historical claims-to-exposure average — the “historical-average” severity scales by that ratio.) The loss waterfall uses the documented February 2025 program: $900M retention, tower to $5.78B with FAIR responsible for ~$3.5B of it via deductible + co-participation (≈53% inside covered layers); the 2025–26 treaty raised retention to ~$1.25B, so the modeled structure is, if anything, favorable to FAIR. Retained loss is treated as assessment need (annual premium is consumed by attritional claims and expenses; the plan publishes no surplus). Surcharge math follows CDI Bulletin 2025-4: 50% of the first $1B and 100% above it, recoupable across ~9M residential policies statewide (2023-vintage count).

• No public claim-level microdata and no “dollars demanded” figure exists anywhere — claims-submitted-vs-paid is triangulated from claim counts, paid dollars, and third-party insured-loss estimates ($40–45B for the two fires).
• The FAIR Plan publishes no audited financial statements. The solvency analysis is a capacity/liquidity reconstruction from public disclosures, not a balance-sheet read.
• FAIR headline statistics run ~1% above the sums of the published ZIP tables (the headline includes items outside the tables); we display the tables.
• Historical premium comparisons (Phase 2) inherit their own caveats: averages confound rate changes with coverage-amount growth; NAIC state averages lag about two years; approved rate-change percentages apply to base rates, not any individual's bill.

Manual, quarterly, reproducible: fetch FAIR PDFs → parse (validated against each file's total row) → precompute per (ZIP, fiscal year) → publish. A drop detector watches cfpnet.com for new releases. Source code conventions follow hamindex.com / nyc.hamindex.com.