Tools chosen: kalshi_direct kalshi_related polymarket_related wikipedia claude_news code_execution

Sub-questions (Fermi decomposition)

Planner reasoning

This is essentially a pure base-rate question: the probability of an M8.0+ epicentered in California over a ~3-year window. Historical maximum magnitudes on California faults (1857 Fort Tejon ~7.9, 1906 San Francisco ~7.9) suggest M8.0 is at or above the physical ceiling for California fault segments, so the base rate is extremely low. The main research tasks are anchoring on the Kalshi price, confirming USGS/UCERF3 forecast probabilities, and computing a Poisson base rate.

# Event Will a California-epicentered earthquake reach at least magnitude 8.0 before December 31, 2028 (resolves via USGS epicenter records)? # Outcomes to forecast - Yes - No # Kalshi market anchor **Current YES price: 9%** (KXEARTHQUAKECALIFORNIA-28). 7-day change: +1pt; 30-day change: -9pt (down sharply from a recent high). Price range over 36 days: 4%–18%. Average daily volume: 138 contracts (thin but active). Market has been volatile and trending down over the past month, currently sitting near the low end of its range. # Sub-question answers 1. **Kalshi YES price/trading** — Currently 9%, down from as high as 18% a month ago; low-volume, thinly traded market with meaningful noise (kalshi_direct). 2. **Historical M8.0+ CA epicenters** — Zero confirmed instrumental/historical M8.0+ earthquakes with a California epicenter since records began (~1769). The largest known events (1857 Fort Tejon, 1906 San Francisco) were both ~M7.9 (Wikipedia). No M8.0+ event has ever occurred in the ~250-year California record. 3. **UCERF3 30-year forecast** — USGS UCERF3 gives ~7% probability of an M8.0+ CA earthquake in 30 years (up from 4.7% in UCERF2), with southern San Andreas most likely source (claude_news/SCEC). Scaled via Poisson to the ~2-3 year window remaining before this market closes, implied probability ≈ 0.5%–1.0% (code_execution). 4. **Max credible magnitude** — USGS/experts state San Andreas maxes out around M8.0–8.3; true M9+ requires subduction-zone-scale rupture area not present on California's strike-slip faults. An M8.0+ event is physically plausible but would exceed any recorded historical CA quake (SF Chronicle, earthquakenearme.com). 5. **Recent 2024-2026 activity** — Mid-2026 study found San Andreas/San Jacinto stress at a 1,000-year high, noting it's been >100 years since a major rupture (Fox Weather). May 2026 Brawley Seismic Zone swarm (hundreds of quakes, max M4.7) and other 2026 swarms (Geysers, offshore Oregon border M5.7) occurred but did NOT trigger any elevated statewide M8+ advisory (watchers.news, edhat, patch.com). 6. **Related market signals** — KXEARTHQUAKECALIFORNIA-35 (before 2035, ~7 more years) prices at 25% YES, down sharply from 30-day high of 65% (kalshi_related). KXEARTHQUAKEJAPAN-30 prices at 37% YES. No Polymarket equivalents exist (0 matches). # Key facts (high-confidence, factual) 1. [Wikipedia] No M8.0+ CA-epicentered earthquake has ever been recorded; largest are ~M7.9 (1857, 1906). 2. [SCEC/USGS UCERF3] 30-year M8+ probability ≈ 7%, driven mainly by southern San Andreas multi-fault rupture scenarios. 3. [USGS via SF Chronicle] San Andreas maximum credible magnitude ≈ 8.0-8.3. 4. [Fox Weather, 2026] San Andreas/San Jacinto stress at highest level in 1,000 years; >100 years since last major rupture. 5. [watchers.news, 2026] Brawley swarm (May 2026) was largest recent seismic event but max magnitude only M4.7; no M8+ advisory issued. # Cross-market signals - Kalshi related: KXEARTHQUAKECALIFORNIA-35 (longer horizon, +7yr) at 25% YES — roughly consistent with a longer window pricing higher cumulative risk than the 2028 market's 9%. - Kalshi related: KXEARTHQUAKEJAPAN-30 at 37% YES, reflecting Japan's much higher subduction-zone M8+ base rate vs. California's strike-slip regime. - Polymarket: no matching markets found (0/100 scanned). - Sportsbook implied: N/A. # Analyst opinions and speculation - Poisson/base-rate modeling (code_execution) converges on ~0.7%–1.0% central probability for the 2028 window (range 0.2%–1.5%), using both zero/one-historical-event counts and UCERF3-derived rates. - Fox Weather/researchers flag elevated stress buildup as a qualitative "overdue" narrative, but this hasn't been quantified into a raised statewide M8+ probability by USGS. - claude_news synthesis explicitly concludes the Kalshi 9% price appears elevated relative to base-rate models (~0.5-1.5%), suggesting a "long-shot/tail-risk premium" is being priced in, possibly due to illiquidity or event salience (Fort Tejon overdue narrative, stress-study headlines). # Directional lean per outcome - **Yes**: Supported by real long-term tectonic loading (San Andreas overdue), UCERF3's non-trivial 7%/30yr rate, and physical plausibility of M8.0-8.3 ruptures. Opposed by zero historical precedent in ~250 years, short remaining window (~2-3 yrs), and no current elevated seismic advisory. - **No**: Strongly favored — historical base rate, UCERF3 scaling, and lack of any 2024-2026 precursor advisory all imply <1.5% probability; Kalshi's 9% appears to overprice tail risk. # Gaps / unknowns - No official USGS short-term (multi-year) M8+ probability statement exists to directly validate the linear Poisson scaling assumption. - Unclear why Kalshi 30-day price dropped from 18% to 9% — could reflect a market correction toward fundamentals, or a specific news-driven repricing not captured in research. - No data on order book depth beyond average volume (138/day), so price could be somewhat noise-driven given low liquidity. # Calibration anchors - Kalshi current YES price (anchor): **9%** - Historical/UCERF3-derived model estimate: **~0.5%–1.5%** for the 2028 window - Precedent: zero M8.0+ CA earthquakes recorded in ~250 years of documented history

1. Both forecasters land at exactly 2.5% with nearly identical reasoning and near-identical wording, essentially ignoring the Kalshi 9% anchor almost entirely despite claiming to "respect market information" — a 9% price with active (if thin) trading and a directionally consistent longer-horizon market (CALIFORNIA-35 at 25%) is meaningful information they discount too heavily, landing much closer to the pure model estimate (~0.5-1.5%) than to any market-blended value. 2. Neither forecast explains why they chose 2.5% specifically rather than, say, 1.5% (top of model range) or 4-5% (a genuine blend toward market price); the convergence to identical values from two "independent" forecasters suggests anchoring on each other or the brief's framing rather than independently justified reasoning. 3. Both underweight the cross-market consistency check: CALIFORNIA-35 (7-year window) at 25% implies an annualized rate far higher than UCERF3 scaling suggests (25% over 7yr vs. ~2-3% over 2yr expected under Poisson consistency), which is a meaningful discrepancy neither forecaster reconciles — either both markets are overpriced by similar tail-premium logic, or the model scaling itself is wrong. 4. Neither addresses the "gap/unknown" flagged in the brief: no official USGS short-term multi-year M8+ statement exists to validate the linear Poisson scaling assumption, and UCERF3's 30-year rate may not scale linearly given time-dependent (not memoryless) rupture models — since San Andreas is "overdue" (>100 years since last major rupture, stress at 1,000-year high), a time-dependent hazard model could imply a higher near-term probability than simple Poisson scaling, which both forecasts apply uncritically. 5. Both forecasts treat the 9%→18%→9% price swing purely as noise/illiquidity without considering alternative explanations (e.g., a specific news event or informed trader activity behind the 18% spike) that the brief itself flags as an unresolved unknown ("could reflect a market correction... or a specific news-driven repricing not captured in research"). 6. Given the physical implausibility argument (max credible magnitude ~8.0-8.3, essentially at the threshold), the resolution is highly sensitive to whether any large rupture just barely reaches 8.0 vs 7.9 — neither forecast discusses this knife-edge magnitude-determination risk, which could argue for either a lower or higher probability depending on measurement/rounding conventions USGS uses near the threshold.

3% (78% confidence)

Yes: 3% No: 97%

No M8.0+ earthquake with a California epicenter has ever been recorded in ~250 years, and USGS puts the San Andreas maximum credible magnitude at only ~8.0–8.3, so a qualifying event would be unprecedented. UCERF3's 7%/30-year figure scales to roughly 0.5–1.0% over the ~2-year remaining window, and even allowing for time-dependent (non-Poisson) hazard given the 'overdue'/high-stress narrative, the upward adjustment is at most a factor of two or three. The Kalshi anchor of 9% (thin volume, 4–18% range, and a CALIFORNIA-35 market at 25% that is itself internally inconsistent with any credible hazard model) most plausibly reflects a systematic long-shot/tail-risk premium common to low-probability catastrophe contracts rather than private information. I take the red team's point that both forecasters clustered mechanically at 2.5% and shade slightly upward toward the market and toward model uncertainty (including knife-edge M7.9-vs-8.0 determination risk), settling at 3% — well below the market but above the pure base-rate estimate.

Total pipeline time: 96.9s