← Back to scans

Will there be an at least 8.0 magnitude earthquake in California before 2035?

KXEARTHQUAKECALIFORNIA-35 · Climate and Weather · 2026-05-15
10%
Agent
32%
Market Price
-22.0%
Edge
62%
Confidence
Volume: 19,417
Spread: 1.0c
Days to resolution: 3516
Markets in event: 1
Final Rationale
UCERF3's authoritative 7%/30yr M8+ probability for California, when scaled to the remaining 9.6 years, yields ~2-3% under Poisson but plausibly 3-5% under BPT renewal with the overdue southern San Andreas (168+ years since 1857 Fort Tejon). I partially accept the critic's points on territorial-waters scope (Cascadia's southern segment could add 2-4 points if it qualifies) and BPT front-loading (could double the near-term slice), pushing me from 7% to 10%. I reject heavy deference to the 32% market price because the orderbook is genuinely broken ($0.01 yes_bid, $0.95 spread, -30pt swings on thin volume) — this is not a price-discovering market, though the comparable 2028 market at 11% suggests informed participants see a slightly higher rate than my base model, supporting a modest upward nudge. I reject the critique that sub-questions added bias since the final combination already anchored on direct Poisson calculation rather than the sub-question average.
Price vs. prediction
How the market moved after the agent's call. Solid blue line is the actual market price; dashed line is the agent's static forecast at $t{=}0$.
$n = 58$ follow-ups
Re-scan Context
This market has been scanned before. Previous predictions:
DatePredictedMarket PriceConfidence
2026-05-08 9% 31% 62%
2026-05-01 9% 34% 60%
2026-04-05 6% 55% 75%
1. Planner Opus · Fermi decomposition + tool plan
Tools chosen: wikipedia_lookup article_search web_search kalshi_data kalshi_orderbook code_execution
Sub-questions (Fermi decomposition)
  1. Is the annual probability of an M8.0+ earthquake in California or its territorial waters greater than 1% per year based on historical seismicity and current expert models? (w=0.40)
  2. Does accumulated seismic stress on California's major fault systems (San Andreas, Cascadia subduction zone edges, offshore faults) currently indicate elevated probability of M8.0+ rupture before 2035? (w=0.35)
  3. Has California experienced an M8.0+ earthquake in the historical or paleoseismic record at a recurrence interval consistent with a >10% probability over any given 10-year window? (w=0.25)
Planner reasoning
This question turns on the empirical annual probability of an M8.0+ earthquake with epicenter specifically in California or its territorial waters. Key distinctions: (1) M8.0+ is an extremely high threshold — the 1906 San Francisco earthquake was ~7.9 (some estimates 7.9-8.3), and California has seen very few confirmed M8+ events in its recorded history; (2) California's seismic hazard models (UCERF3) estimate roughly 7% probability of an M8+ on the San Andreas system over 30 years, implying ~0.24%/yr just for that fault; (3) Offshore faults and the southern end of the Cascadia subduction zone may add probability, but the epicenter must be in California waters, not Oregon/Washington; (4) With ~9.6 years remaining (May 2026 to Dec 2035), we need to compute cumulative Poisson probability over that window. Reference class: California has experienced perhaps 1-3 events plausibly at M8+ since 1700 (Fort Tejon 1857 ~M7.9, 1906 ~M7.9-8.3), suggesting recurrence intervals of decades to centuries for M8+. The temporal analysis: for YES to occur, a fault segment must accumulate sufficient stress and rupture within a specific 9.6-year window — the 'Big One' scenario. Most seismologists estimate the probability of M8+ in any given decade in California at 5-15%, depending on the exact fault system and whether offshore zones count. The question is whether territorial waters include the Cascadia subduction zone's southern extent off northern California's coast, which could significantly elevate probability. The combination logic is weighted_average since sub-questions probe different aspects of the same underlying probability.
2. Research Parallel tool calls · legacy v2
ToolStatusTimeItemsSummary
wikipedia_lookup OK 1.9s -
article_search OK 0.7s -
web_search OK 45.5s -
kalshi_data OK 2.8s -
kalshi_orderbook OK 0.1s -
code_execution OK 0.0s -
3. Evidence Synthesis Sonnet · 18 items · legacy v2
#StrengthCredibilityDirectionSourceClaimPriced In
1 STRONG 90 ↓ DOWN web_search UCERF3 (the authoritative USGS/CGS model) estimates a 7.0% probability of an M8.0+ earthquake in California over the next 30 years, implying roughly 0.23%/year annual rate, well below the 1%/year threshold in sq1. Yes
2 STRONG 85 ↓ DOWN web_search Regional USGS breakdown shows Northern California ~2% and Southern California ~3% probability of M8.0+ over 30 years, summing to roughly 5% combined (with overlap), consistent with an annual rate of ~0.17–0.23%/yr—far below 1%/yr. Yes
3 STRONG 88 ↓ DOWN code_execution Poisson calculation at the UCERF3-implied rate (~0.23%/yr) over 9.63 years yields roughly 2.2% cumulative probability; only at rates ≥1%/yr does the 9.63-year probability reach ~9%, confirming sq1 likely resolves NO under mainstream models. Yes
4 STRONG 88 ↓ DOWN web_search California has not experienced a confirmed M8.0+ earthquake since 1906 (~120 years), consistent with a long recurrence interval and a low annual rate well under 1%/yr. Yes
5 MODERATE 75 ↑ UP wikipedia_lookup The Cascadia subduction zone extends into Northern California, capable of M9.0+ events; its southern terminus lies offshore near Cape Mendocino, CA, which could contribute to the M8+ probability for 'California territorial waters' depending on resolution scope. Yes
6 STRONG 85 ↑ UP web_search The southern San Andreas Fault (Carrizo, Mojave, Coachella segments) has not ruptured in a major event since the 1857 Fort Tejon M7.9 earthquake—168 years ago—while the estimated recurrence interval is 140–160 years, meaning the fault is overdue for a large rupture. Yes
7 MODERATE 80 ↑ UP web_search The Mojave segment of the San Andreas carries a 19% probability of M7.5+ in 30 years (highest of any California fault segment), but this is for M7.5+, not M8.0+, limiting its direct relevance to the question threshold. Yes
8 WEAK 55 NEUTRAL article_search A swarm of M4.5 earthquakes hit Imperial County (Southern California) near Brawley in May 2026, proximal to the San Andreas Fault system, but these are minor events not indicative of imminent M8+ rupture. Yes
9 WEAK 55 NEUTRAL article_search In January 2026, a M4.9 earthquake near Indio prompted seismologist Dr. Lucy Jones to warn of continued seismic activity due to proximity to the San Andreas, but no expert commentary linked this to elevated M8+ probability. Yes
10 MODERATE 72 ↑ UP wikipedia_lookup The San Andreas Fault accumulates slip at 0.79–1.38 inches/year; with 168 years since the last major southern rupture, substantial elastic strain energy has accumulated on multiple locked segments. Yes
11 MODERATE 82 ↑ UP web_search UCERF3's increased M8+ probability versus UCERF2 (7% vs 4.7% over 30 years) reflects inclusion of multi-fault rupture scenarios—suggesting that connecting fault ruptures could reach M8+ even if individual segments are unlikely to do so alone. Yes
12 STRONG 85 ↓ DOWN web_search California's confirmed M8+ events in the historical/instrumental record are limited: the 1906 San Francisco earthquake (estimated M7.9–8.3) and possibly Fort Tejon 1857 (M7.9), suggesting recurrence intervals of many decades to over a century between M8+ events statewide. Yes
13 STRONG 85 ↓ DOWN web_search UCERF3's 30-year M8+ probability of 7% implies a mean recurrence interval of roughly 400+ years for M8+ events in California, consistent with a 10-year probability of approximately 2–2.5%—well below the >10% threshold asked about in sq3. Yes
14 MODERATE 70 ↑ UP wikipedia_lookup The Cascadia subduction zone has a paleoseismic record of full-margin M9+ ruptures roughly every 200–500 years; its southern end off Cape Mendocino could qualify as California territorial waters, potentially adding to M8+ recurrence—but the epicenter would typically be offshore Oregon/Washington for full-margin events. Yes
15 STRONG 88 ↓ DOWN code_execution For the 10-year window probability to exceed 10%, the annual rate would need to be approximately 1.05%/yr or higher—roughly 4–5x the UCERF3 central estimate of ~0.23%/yr—indicating the historical record does not support sq3 resolving YES under mainstream models. Yes
16 MODERATE 55 ↑ UP kalshi_data The Kalshi market for M8+ California before 2028 is priced at 11%, implying roughly a 1.1–1.3%/yr annual rate over the ~2 remaining years—somewhat higher than UCERF3 central estimates, possibly reflecting market uncertainty or upward bias. Yes
17 MODERATE 50 ↑ UP kalshi_data The Kalshi market for M8+ California before 2035 is currently priced at 32%, which is dramatically higher than the ~2–2.5% implied by UCERF3 for a ~10-year window, suggesting significant market overpricing relative to seismological models or a much broader interpretation of the question. No
18 MODERATE 60 ↓ DOWN kalshi_orderbook The orderbook spread is extremely wide ($0.95) with yes_bid at only $0.01 and no_bid at $0.04, suggesting very thin real buy-side support and that the 32% 'price' may not reflect genuine informed trading activity. Yes
Information Gaps
  • No direct data on whether the Gorda plate / southern Cascadia subduction zone segments qualify as 'California territorial waters' under the question's resolution criteria—this distinction could add meaningful probability.
  • No paleoseismic data specifically for offshore California faults (e.g., Mendocino Triple Junction area, offshore thrust faults) that could produce M8+ with epicenters in California waters.
  • No recent UCERF4 or post-2015 updates to the M8+ probability estimates; UCERF3 is ~10 years old and may not reflect the latest fault interaction models.
  • No expert commentary specifically evaluating the probability of a multi-fault cascade rupture (as modeled in UCERF3) reaching M8+ within the next 9.6 years.
  • Lack of clarity on whether the question's 'territorial waters' include the outer continental shelf off Northern California where the Gorda segment of Cascadia lies.
  • No data on current GPS/geodetic strain measurements that would indicate whether stress accumulation on major fault segments is at historically elevated levels.
  • The extremely wide Kalshi price range (8%–73% over 100 days) and dramatic 30-day decline (-30 points) suggest possible manipulation or resolution-event-driven spikes; no explanation for the spike to 73% was found.
Key Uncertainties
  • Whether 'California territorial waters' includes the southern Cascadia subduction zone (Gorda segment), which could substantially raise M8+ probability if included.
  • Whether the 1906 earthquake should be classified as M7.9 or M8.3—estimates vary, affecting whether California has a precedent for confirmed M8+ events in the instrumental era.
  • The degree to which multi-fault rupture scenarios (the basis for UCERF3's increased estimate vs UCERF2) could manifest as a single M8+ event in the next decade.
  • The Kalshi market's large price discrepancy from seismological models (32% vs ~2-3% implied by UCERF3) is unexplained—it may reflect broader question interpretation, anchoring bias, or thin liquidity.
  • Whether the southern San Andreas being 'overdue' by 8+ years beyond its mean recurrence interval meaningfully elevates short-term probability in a non-Poisson (time-dependent) model.
Market Context
Price: 0.32 · 7d: +0.03 · 30d: -0.30
Volume: LOW — 37 contracts/day average is very thin · Depth: MISLEADING — reported depth of 23,436 but bid/ask spread is $0.95 with yes_bid at $0.01; actual informed liquidity appears very low
Reliability: LOW — extremely wide spread, dramatic 30-day price swings (-30 points), thin volume, and yes_bid of only $0.01 suggest this market is not reliably price-discovering; the 32% mid-price is not credible as a well-informed probability estimate
4. Independent Forecaster Opus · No market price · legacy v2 Did NOT see market price
Independent estimate: 7% (66% confidence)
base rate
7%
70% conf
evidence driven
8%
65% conf
contrarian
7%
62% conf
Spread
1pp
Low disagreement
sq1: Is the annual probability of an M8.0+ earthquake in California or its territorial waters greater than 1% per year?
8%
UCERF3 implies ~0.23%/yr, far below 1%/yr. Even with time-dependent adjustments for overdue southern San Andreas and potential Cascadia/Gorda contribution, getting to >1%/yr would require ~4-5x mainstream estimates. Small chance UCERF4 or territorial-waters scope shifts this, but unlikely. ~8%.
sq2: Does accumulated seismic stress indicate elevated probability of M8.0+ rupture before 2035?
20%
Southern San Andreas is overdue (168+ years vs ~150-year recurrence) and UCERF3 includes multi-fault cascades. However, the highest single-segment M7.5+ probability is only 19% over 30 years, and M8+ requires multi-segment rupture. Time-dependent models do elevate near-term probability somewhat but not dramatically. The question asks about 'elevated' which is somewhat subjective; I interpret as meaningfully above base rate (say >5% over 9.6 years). Probability stress is elevated to this degree: ~20%.
sq3: Has California's historical/paleoseismic record shown recurrence consistent with >10% per decade for M8+?
5%
Mean recurrence interval based on UCERF3 (~400+ years) and historical record (1857, 1906 both arguably sub-M8) implies ~2-2.5% per decade, well below 10%. Only ambiguity around 1906 magnitude classification or inclusion of Cascadia offshore events could push this up. ~5%.
Show full reasoning trace
base rate: UCERF3 gives 7% over 30 years for M8+ in California, implying ~2.3% Poisson probability over the 9.6-year remaining window. Historical record shows perhaps 1-2 M8+ events in ~300 years (1857 Fort Tejon ~M7.9, 1906 SF ~M7.9-8.3), consistent with mean recurrence of 150-400 years.
evidence updates: Time-dependent (BPT renewal) models for the overdue southern San Andreas push probability up by perhaps 1.5-2x. Inclusion of Cascadia/Gorda offshore segments in 'California territorial waters' adds maybe 1-2 percentage points. Multi-fault cascade scenarios in UCERF3 are already priced into the 7%/30yr figure. May 2026 minor swarms in Imperial County are not predictive (foreshocks have <5% predictive value).
combination method: The sub-questions are proxies for the same underlying physical event. Rather than naive weighted average, I anchor on the direct Poisson calculation (~2.3% baseline) and adjust upward for time-dependent renewal and territorial waters scope to ~7%.
final: Final estimate ~7%. Anchored on UCERF3's authoritative ~2.3% decadal baseline, adjusted upward for time-dependent renewal on the overdue southern San Andreas (168+ years since 1857), potential inclusion of Cascadia southern segment in California waters, and uncertainty about whether 1906 truly was sub-M8. Confidence is moderate-high given clear seismological consensus models.
ensemble: {'ensemble_count': 3, 'perspectives': {'base_rate': {'p_yes': 0.07, 'confidence': 0.7}, 'evidence_driven': {'p_yes': 0.08, 'confidence': 0.65}, 'contrarian': {'p_yes': 0.07, 'confidence': 0.62}}, 'spread': 0.01, 'individual_p_yes': {'base_rate': 0.07, 'evidence_driven': 0.08, 'contrarian': 0.07}}
5. Devil's Advocate Sonnet · legacy v2 Market price revealed
Reasoning Flaw Missing Info
Challenges
  1. The forecast relies heavily on UCERF3's 7%/30yr figure but applies a Poisson scaling to 9.6 years that yields ~2.3%, then adjusts up to 7%. However, UCERF3's 7% is itself a 30-year time-dependent probability that already incorporates renewal effects. Re-applying time-dependent adjustments on top risks double-counting. Conversely, scaling 7%/30yr linearly to 9.6 years gives ~2.3% only under Poisson assumptions; under BPT renewal with overdue faults, the near-term hazard could be front-loaded, making the 9.6-year slice more than 1/3 of the 30-year probability.
  2. The decomposition is weak: sq1 (>1%/yr) and sq3 (>10%/decade) are essentially restatements of the same question with thresholds chosen above plausible values, biasing the combined estimate downward. The combination method wisely ignores them and anchors on direct calculation, but then why include them at all? They contribute noise.
  3. The 'territorial waters' scope question is flagged as a real information gap that could 'add meaningful probability' but is only credited with 1-2 percentage points. If Cascadia's southern segment qualifies, paleoseismic recurrence of ~200-500 years for M9 events could add substantially more than 1-2 points over 9.6 years (base rate alone ~2-5%).
  4. The forecaster dismisses the Kalshi market at 32% as unreliable due to wide spreads and thin volume — which is reasonable — but doesn't seriously engage with WHY the market settled there. Even a noisy market converging in the 11-32% range across two timeframes suggests informed participants see higher probability than 7%. The forecaster should consider whether their model is missing something (e.g., territorial waters interpretation, UCERF4 updates).
Suggested adjustment: +3pp
Modest upward adjustment warranted because (1) the territorial waters/Cascadia scope ambiguity is underweighted given its potential to add several percentage points, and (2) time-dependent BPT models on the overdue southern San Andreas may front-load near-term hazard more than the linear scaling suggests. Not adjusting toward market — adjusting because the forecaster's own identified information gaps deserve wider uncertainty bounds.
Market comparison: Forecaster: 7.3%, Market: 32%. The market is genuinely unreliable (thin volume, $0.95 spread, $0.01 yes_bid, -30pt swing in 30 days), so heavy discounting is justified. However, a comparable market for 2028 priced at 11% (implying ~1.2%/yr) also exceeds the forecaster's implied rate. The divergence is partially justified by market noise but the forecaster may still be slightly underweighting scope ambiguity and BPT front-loading.
6. Aggregator Opus · final calibrated forecast
10% (62% confidence)
UCERF3's authoritative 7%/30yr M8+ probability for California, when scaled to the remaining 9.6 years, yields ~2-3% under Poisson but plausibly 3-5% under BPT renewal with the overdue southern San Andreas (168+ years since 1857 Fort Tejon). I partially accept the critic's points on territorial-waters scope (Cascadia's southern segment could add 2-4 points if it qualifies) and BPT front-loading (could double the near-term slice), pushing me from 7% to 10%. I reject heavy deference to the 32% market price because the orderbook is genuinely broken ($0.01 yes_bid, $0.95 spread, -30pt swings on thin volume) — this is not a price-discovering market, though the comparable 2028 market at 11% suggests informed participants see a slightly higher rate than my base model, supporting a modest upward nudge. I reject the critique that sub-questions added bias since the final combination already anchored on direct Poisson calculation rather than the sub-question average.