Tools chosen: kalshi_direct kalshi_related polymarket_related wikipedia claude_news code_execution

Sub-questions (Fermi decomposition)

Planner reasoning

This is a pure low-probability base-rate question: no M8.0+ earthquake has ever been recorded in California (1857 Fort Tejon and 1906 San Francisco were ~7.9), and USGS UCERF3 gives very low annual probabilities for M8+ on California faults. The key work is anchoring on the Kalshi price, retrieving USGS/UCERF3 probability estimates, and computing a Poisson base rate over the ~3-year window.

# Event Will an earthquake of magnitude ≥8.0 with epicenter in California (or its territorial waters) occur before Dec 31, 2028? # Outcomes to forecast - Yes - No # Kalshi market anchor **KXEARTHQUAKECALIFORNIA-28 — current YES price: 13.00%**. 7-day change: -1pt; 30-day change: -3pt (trending down). Avg daily volume: 142 contracts. Price range over observed period: 4%-18% (32 data points) — notably volatile for a low-probability tail event, suggesting thin liquidity and possible mispricing rather than new information flow. # Sub-question answers 1. **Kalshi YES price/volume?** — 13.00% YES, ~142 contracts/day avg, down 3pts over 30 days [kalshi_direct]. 2. **Historical M8.0+ CA epicenters since ~1800?** — Zero confirmed instrumental-record M8.0+ events with CA epicenter; largest were 1857 Fort Tejon (M7.9) and 1906 San Francisco (M7.9) [Wikipedia]. This is a genuine tail event with no direct historical precedent in ~200+ years of record. 3. **UCERF3 30-year M8.0+ probability?** — UCERF3 (2015) estimates ~7% for M≥7.8 in 30 years (up from 4.7% in UCERF2); this figure is often cited loosely as "M8+" but technically covers M≥7.8, meaning true M8.0+ probability is somewhat lower [claude_news]. Annualized rate ≈0.0016-0.0024/yr under different interpretations. 4. **Poisson probability by end-2028?** — Model synthesis: ~0.2%-1.0% central estimate for the ~3-4 year window (now through 2028), with fat-tail scenarios up to ~3-4% if using upper UCERF3 bound uncritically [code_execution]. Point estimate: ~0.5-0.8%. 5. **Elevated seismic activity/advisories?** — Active swarms: San Ramon/East Bay (Nov-Dec 2025, up to M4), Calaveras fault swarm (Nov 2025, 40+ quakes, largest M3.8), M7.0 off Humboldt Peninsula (Dec 5, 2025, tsunami warning canceled), Southern CA saw record swarm activity in 2024 (15 sequences) but sharp drop in 2025 (5 sequences) [claude_news]. No USGS/CA OES official statement has raised M8+ probability estimates due to these swarms. 6. **Comparable market pricing consistency?** — KXEARTHQUAKECALIFORNIA-35 (before 2035, longer horizon) prices at 24% YES; KXEARTHQUAKEJAPAN-30 (before 2030) prices at 40% YES (Japan has much higher M8+ base rate — subduction zone). CA-28 at 13% is internally consistent with CA-35 scaling up for time, but both appear high relative to physical models [kalshi_related]. # Key facts (high-confidence, factual) 1. [Wikipedia] No M8.0+ earthquake with CA epicenter recorded in instrumental history (~200 yrs); largest were M7.9 (1857, 1906). 2. [claude_news] UCERF3 official benchmark: ~7% in 30 years for M≥7.8, still cited as of Dec 2025. 3. [claude_news] Southern San Andreas is the fault segment most likely to produce a very large event. 4. [code_execution] Poisson/Bayesian modeling from both UCERF3 rates and empirical zero-event base rate converge on ~0.2%-1.0% probability for the 2025-2028 window. 5. [claude_news] Multiple 2025 swarms (San Ramon, Calaveras, Humboldt M7.0) have not triggered any official upward revision of M8+ probability. # Cross-market signals - Kalshi related: CA-2035 market prices 24% YES (vs 13% for CA-2028) — roughly consistent scaling for longer horizon. Japan-2030 M8+ market at 40% reflects genuinely higher subduction-zone base rate, not directly comparable. - Polymarket: No matching earthquake markets found [polymarket_related] — no cross-check available. - Sportsbook implied: N/A (not applicable to this category). # Analyst opinions and speculation - code_execution synthesis explicitly flags the Kalshi price as likely overpriced: "a market price above ~3-5% would appear overpriced relative to physical seismic-hazard modeling," and 13% is well above even the pessimistic model range. - Media coverage of 2025 swarms sustains public "Big One" salience, likely inflating retail-driven market pricing despite no corresponding scientific probability increase. # Directional lean per outcome - **No**: Strongly favored. Zero historical M8.0+ CA epicenters in 200+ years; UCERF3 implies well under 1% probability for a 3-year window; no current advisory raises this estimate despite active swarms. - **Yes**: Weak support — only the general "elevated swarm activity" narrative and the possibility that UCERF3's M≥7.8 threshold underestimates true M8+ tail risk, plus market pricing (13%) reflecting some fat-tail/uncertainty premium. # Gaps / unknowns - No official USGS M8.0-specific (vs M7.8+) probability figure directly available; conversion introduces uncertainty. - Unclear how "territorial waters" epicenter definition might include offshore Cascadia-adjacent events not fully modeled here. - No Polymarket cross-check available. # Calibration anchors - Kalshi current YES price: 13% (anchor, but likely overpriced vs. models). - Physical/statistical models (UCERF3 + Bayesian empirical): ~0.2%-1.0% (central ~0.5-0.8%) for the window to end-2028. - No historical precedent of M8.0+ CA epicenter ever recorded instrumentally.

1. Both forecasts land at 3-4% Yes, well above the brief's stated central model estimate of ~0.5-0.8% and even above the "fat-tail upper bound" of ~3-4%; neither justifies why they anchor near the top of the model range rather than the central estimate, effectively giving substantial weight to the "market may know something" hypothesis without evidence for it. 2. Neither forecaster explicitly addresses the "territorial waters" resolution ambiguity flagged in the gaps section — offshore Cascadia-adjacent or Humboldt-type events could resolve YES even if they're not "California epicenter" in the strict sense, which could meaningfully raise true probability above the onshore-only historical base rate of zero. 3. Both cite the CA-2035 market's 24% price as a "consistency check" validating some fat-tail premium, but the brief explicitly flags this scaling as "internally consistent... but both appear high relative to physical models" — using one likely-overpriced market to justify partially trusting another is circular reasoning neither forecaster interrogates. 4. Neither forecast discusses the Southern San Andreas segment specifically (flagged in the brief as "the fault segment most likely to produce a very large event") — no discussion of whether current locking/strain data on this segment shifts probability within the window versus a flat Poisson assumption. 5. Both forecasters converge suspiciously close (4% vs 3%) via nearly identical reasoning chains, suggesting anchoring on the same brief narrative rather than independent model derivation; neither explores whether the true answer could be even lower (e.g., 1-2%, closer to the stated central model range) given the near-zero historical base rate over 200 years. 6. Neither forecast quantifies how much weight is being placed on "market may reflect resolution-criteria information" versus pure physical modeling — this hedge appears to be doing most of the work pushing both estimates 4-8x above the stated central physical estimate (~0.5-0.8%), and neither justifies that multiplier explicitly.

2% (74% confidence)

Yes: 2% No: 98%

Physical hazard modeling is the dominant evidence: zero M8.0+ California-epicenter events in 200+ years of record (max M7.9 in 1857 and 1906), and UCERF3's ~7%/30-yr figure for M≥7.8 annualizes to only ~0.2-1.0% over the remaining ~3-year window. The 13% Kalshi anchor sits on a thin, volatile market (142 contracts/day, 4-18% range) with obvious retail 'Big One' salience premium, and the CA-2035 market at 24% is not an independent validation but the same bias scaled — so the critique is right that both forecasters over-weighted the market. I do allow a modest uplift above the ~0.5-0.8% central estimate for genuine model uncertainty, the M8.0-vs-M7.8 threshold conversion, and the 'territorial waters' resolution ambiguity that could capture an offshore Cascadia-adjacent rupture off Humboldt (a segment that genuinely can produce M8+), which the strict onshore base rate excludes. The 2025 swarms triggered no USGS/CA OES probability revision and warrant essentially no update. Net: ~2.5% Yes, below both forecasters and well below the market, but above the pure onshore Poisson central estimate.

Total pipeline time: 112.8s