Tools chosen: kalshi_direct kalshi_related polymarket_related wikipedia claude_news gdelt_news code_execution

Sub-questions (Fermi decomposition)

Planner reasoning

This is a long-horizon geophysical base-rate question: probability of an M8.0+ earthquake with epicenter in California (or its territorial waters) in roughly a 10-year window. The key work is establishing the historical base rate (California has never had a recorded M8.0+ — 1857 Fort Tejon and 1906 San Francisco were ~M7.9), plus UCERF3 forecast probabilities for M8+ over 30 years, then converting to a decadal Poisson rate. Market price on Kalshi is the anchor; cross-venue and related-market checks add signal.

# Current state Resolution requires an epicenter within California (or its territorial waters) reaching M≥8.0 before 2035-12-31. No M≥8.0 event has ever been instrumentally recorded with a California epicenter (largest are ~7.9). The Kalshi market currently prices this at 26% YES, down sharply (-22pts) over the last 7 days despite no new triggering event reported, suggesting a large repricing/liquidity move rather than new hazard information. # Timeline of key events - 1857-01-09: Fort Tejon earthquake, Mw 7.9, San Andreas rupture ~350km (confirmed, Wikipedia). - 1906-04-18: San Francisco earthquake, M 7.9 (confirmed, Wikipedia). - 2015: USGS/CGS/SCEC release UCERF3, raising 30-yr P(M≥8.0 CA) from 4.7% (UCERF2) to ~7.0% (confirmed, conservation.ca.gov). - 2023: USGS National Seismic Hazard Model (NSHM) update — focused on ground-motion/basin models, did not reissue a new statewide M8+ headline probability; UCERF3's 7% remains the standing figure (reported, seismosoc.secure-platform.com). - 2024–2026: Ongoing reporting (e.g., Univ. of Bern study) that stress on parts of Southern San Andreas may be at historic highs ("highest in 1,000 years"), amplifying "Big One" concern but not producing a revised numeric probability (reported, abc7news.com). - Last 30 days (Kalshi): price flat to -0%, then a sharp -22pt drop in the last 7 days from ~48% to 26% (confirmed, Kalshi data) — no corresponding news event identified in research; likely trader repositioning/thin volume artifact (~130 contracts/day avg). # Event Will a magnitude ≥8.0 earthquake occur with epicenter in California (or its territorial waters) before 2035-12-31? # Outcomes to forecast - Yes - No # Kalshi market anchor KXEARTHQUAKECALIFORNIA-35: **Current YES price 26%**. Down 22pts in 7 days (from recent high), flat over 30 days. Range over 46 days: 7%–65% (highly volatile for a low-probability tail-risk market). Average daily volume only 130 contracts — thin market, prone to noise-driven swings. This is the consensus to beat. # Sub-question answers 1. **Largest recorded CA earthquake / any M8.0+?** No M≥8.0 has ever occurred with a California epicenter historically. Largest are M7.9 (1857 Fort Tejon, 1906 San Francisco) [Wikipedia]. 2. **UCERF3 30-yr M≥8.0 probability / annual rate?** UCERF3 (2015, USGS/CGS/SCEC) estimates 30-yr P(M≥8.0)≈7.0% (up from 4.7% in UCERF2) [claude_news/conservation.ca.gov]. Implied annual λ≈0.0024/yr; scaled to the ~10.1-yr window to 2035, ≈2.4% [code_execution]. 3. **Empirical base rate M≥7.5/7.9 last ~200 yrs?** Two M7.9 events in ~200 years (1857, 1906); zero M≥8.0. Empirical Jeffreys-prior estimate from zero-events-in-200-yrs gives λ≈0.0025/yr, ≈2.5% over 10.1 yrs — closely matching UCERF3 [code_execution]. 4. **Physical capacity of San Andreas for M8+?** USGS notes the San Andreas is 800 miles long but only 10-12 miles deep, making magnitudes >8.3 "extremely unlikely" alone; M8.0+ is plausible but near the top of single-fault capacity, more likely via UCERF3's multi-fault rupture scenarios, with southern San Andreas flagged as most likely source [claude_news/USGS]. 5. **Magnitude-estimate ambiguity effect?** Not directly addressed in research; historically significant CA quakes (1857, 1906) are pinned at Mw 7.9, not rounded to 8.0 — implies resolution will likely hinge on clear-cut instrumental Mw readings, with little rounding ambiguity risk given current data quality standards. 6. **Kalshi YES price / comparable markets?** Current market 26% YES (2035 horizon); comparable KXEARTHQUAKECALIFORNIA-28 (2028 horizon) prices only 4% YES, consistent with a much shorter window; KXEARTHQUAKEJAPAN-30 (8.0 threshold) prices 40% YES, reflecting Japan's higher subduction-zone seismicity. No Polymarket equivalent found (0 matches) [kalshi_related, polymarket_related]. # Key facts (high-confidence, factual) 1. [Wikipedia] No CA earthquake has ever reached M8.0 instrumentally; max recorded is M7.9 (1857, 1906). 2. [conservation.ca.gov via claude_news] UCERF3 (2015) is the authoritative model: 30-yr P(M≥8.0)≈7%. 3. [USGS] San Andreas physical geometry caps standalone rupture magnitude near 8.3. 4. [Kalshi] 2035 contract at 26% YES; 2028 contract (shorter horizon) at only 4% YES — directionally consistent with time-scaling. 5. [code_execution] Poisson-scaled UCERF3 rate to 10.1-yr window ≈2.4-2.5%; GR-extrapolation upper bound ≈5-6.5%. # Cross-market signals - Kalshi related: KXEARTHQUAKECALIFORNIA-28 (2028) at 4% YES — implies market believes cumulative risk rises materially 2028→2035, consistent with longer window, but the 2035 contract's 26% seems high relative to the 4% 2028 baseline (a ~6.5x jump for ~1.4x more time is inconsistent with constant-hazard Poisson scaling). - Polymarket: no matching markets found. - Sportsbook implied: N/A (not applicable to this category). # Analyst opinions and speculation - Some seismologists (Univ. of Bern study) argue stress on parts of southern San Andreas is at a multi-century high, raising subjective "Big One" concern beyond the static UCERF3 baseline [abc7news.com] — but no updated quantitative probability has been published. - UC Irvine/ASU geologists note major San Andreas quakes recur every 45-144 years, and it's been ~150+ years since 1857, suggesting "overdue" narrative common in media [foxnews.com] — this speaks to overall "Big One" risk, not specifically M≥8.0 vs M7.0-7.9 distinction. # Directional lean per outcome - **Yes**: Supported by rising "overdue" sentiment, elevated fault-stress studies, and UCERF3's multi-fault rupture scenarios enabling M8+. Opposed by: zero historical precedent of CA M8.0+, physical fault-depth constraints capping single-fault magnitude near 8.3, and quantitative models (UCERF3, Jeffreys-prior, Poisson) converging on ~2.4-2.5% baseline probability for the actual window — far below the market's 26%. - **No**: Strongly favored by base-rate/model math (~93.5-97.5% implied); the current 26% Kalshi price appears substantially overpriced relative to all quantitative estimates, though GR-extrapolation upper bound (~5-6.5%) still falls well short of 26%. # Gaps / unknowns - No explanation found for the recent -22pt weekly Kalshi price swing; may reflect thin-volume noise, a single large trade, or unreported news not captured by GDELT (rate-limited, 0 articles returned). - No updated post-2015 statewide M8+ probability from USGS NSHM 2023 — UCERF3 remains the latest formal estimate, now 10 years old. - Magnitude-rounding/revision risk for borderline events (e.g., a preliminary 7.9 later revised to 8.0) not explicitly modeled. # Calibration anchors - Kalshi current YES price: **26%** (anchor, but volume thin at 130/day and highly volatile 7-65% range). - Quantitative model consensus (UCERF3-based Poisson, empirical Jeffreys prior): **~2.4-2.5%** for the ~10.1-yr window; upper bound via Gutenberg-Richter extrapolation ~5-6.5%. - Historical precedent: 0 of ~2 centuries of CA seismic record has produced M≥8.0 (max M7.9 twice).

1. Both forecasters converge tightly around 5-6%, but neither explains why they discount the Kalshi price so heavily rather than partially anchoring on it—dismissing a live market entirely (even a thin one) as pure "noise" is itself a strong claim that deserves more justification, especially since 130 contracts/day and a 7-65% range could also reflect informed traders reacting to unmodeled tail risk (e.g., Cascadia-adjacent triggering, multi-fault cascade scenarios) rather than pure noise. 2. Neither forecast addresses the "overdue" recurrence-interval evidence in a quantitative way — the brief notes 45-144 year recurrence for major San Andreas quakes and ~150+ years since 1857, which if used to adjust a time-varying (non-Poisson, elastic-rebound-style) hazard model rather than constant-rate Poisson, could push probability upward; both default to constant-rate assumptions without testing sensitivity to this alternative model. 3. The magnitude-rounding/revision risk is mentioned in the brief as an explicit gap ("a preliminary 7.9 later revised to 8.0 not explicitly modeled") but both forecasts treat it as a minor upward nudge rather than estimating its magnitude — given historical events are pinned at 7.9 twice, even a small probability mass on "borderline event gets revised/reinterpreted as 8.0" could matter for a threshold question, and this deserves explicit quantification rather than a qualitative footnote. 4. Both forecasts nearly identically ignore the GR-extrapolation upper bound (5-6.5%) as a real possibility rather than a tail scenario—Forecast 1 lands at 6% (essentially at the upper bound) while Forecast 2 lands at 5% (near the lower-model estimate), yet neither explains why they weight the UCERF3/Jeffreys point estimate so much more than the upper-bound scenario, which given "elevated stress at 1000-year highs" might deserve more weight. 5. Neither forecast considers that the resolution criteria includes California's "territorial waters," which could include Cascadia subduction zone overlap or offshore fault systems (e.g., areas near the CA-Oregon border) with potentially higher M8+ capacity than the shallow San Andreas—this geographic broadening is not addressed by either forecaster despite being explicit in the resolution language. 6. Both forecasts treat the two independent estimation methods (UCERF3 scaling and Jeffreys-prior empirical) as fully independent confirmatory evidence, but they are not truly independent since UCERF3 itself was likely calibrated against the same ~200-year historical record—this could create false confidence from what is essentially one data source counted twice.

7% (medium confidence)

Yes: 7% No: 93%

All independent quantitative anchors converge low: UCERF3's 30-yr P(M≥8.0)≈7% scales to ~2.4% over the remaining ~10 years, the empirical record (zero M≥8.0 in ~200 years, max 7.9 twice) gives ~2.5%, and a Gutenberg-Richter upper bound reaches only ~5-6.5%. I give partial credit to the critique's valid points — time-varying elastic-rebound hazard given ~150+ years since 1857 and 1000-year-high stress, magnitude-revision risk on a borderline 7.9, and the resolution's inclusion of territorial waters (offshore Mendocino/southern Cascadia sources with genuine M8+ capacity) — which collectively justify sitting at or slightly above the GR upper bound rather than the point estimate. I do not anchor meaningfully on the 26% Kalshi price: it is internally inconsistent with the same platform's 4% 2028 contract (6.5x price for 1.4x exposure time), trades ~130 contracts/day, and swung 22 points with no identified hazard news, so it is far more plausibly a thin-market artifact than informed repricing. Final call: Yes ~7%, modestly above both forecasters to absorb model-uncertainty and non-San-Andreas offshore pathways, but far below the market.

Total pipeline time: 224.3s