Reference classes (outside view)

Tetlock-style reference-class forecasting: each class has a historical base rate. Linked nodes get their inside-view posterior blended with the base rate using a time-weighted scheme: w_inside = 0.3 + 0.7 × (1 − TRF). Outside view dominates early; inside view dominates near the deadline.

Reference classes
19
0 with no linked nodes
Linked nodes
284
of 1607 predictions
Coverage
17.7%
Total historical N
401
across all 10 classes

humanoid_commercial_volume

0/3 historical hits — base rate 10.0%
>10,000 unit cumulative deployment of humanoid robot SKU within 3 years of debut
linked
73
mean inside
53%
mean blended
35%
blend direction: 69 pulled down · 1 pulled up · 3 unchanged
Top matches
  • CMQ_051sim=0.76Humanoid robots total addressable market by 2040: lower bound 1 billio40% → 37%
  • ROB_008sim=0.76Humanoid robotics market will reach $5 trillion by 2050 (excluding mas48% → 37%
  • INF_025sim=0.72Figure AI targets deployment of 100,000 general-purpose humanoid robot40% → 37%

agi_breakthrough_5y

1/5 historical hits — base rate 20.0%
Major capability discontinuity (e.g. AGI by named target year, 5-year horizon)
linked
48
mean inside
55%
mean blended
43%
blend direction: 42 pulled down · 6 pulled up · 0 unchanged
Top matches
  • FUT_002sim=0.71Mean expert estimate for AGI collapsed from 50-year horizon to ~5 year55% → 40%
  • CMQ_016sim=0.68Post-AGI (2027+), a decade of human-led algorithmic progress will be c30% → 33%
  • CMQ_011sim=0.68AGI is plausible within 10 years, BUT alignment and safety must be sol50% → 39%

energy_grid_rebuild_5y

3/6 historical hits — base rate 50.0% · linear decay
Major grid expansion (>50GW) within 5y of announce
linked
38
mean inside
60%
mean blended
49%
blend direction: 33 pulled down · 2 pulled up · 3 unchanged
Top matches
  • 238_047sim=0.71US will add 86 GW of new grid capacity in 2026, 51% from solar50% → 45%
  • 241_020sim=0.701 GW of power = ~$50 billion in hardware/software/data centers60% → 49%
  • INF_048sim=0.69Global data-center energy demand will reach approximately 945 Terawatt70% → 48%

ipo_trillion_plus

2/4 historical hits — base rate 50.0%
Trillion-dollar private company IPO at announced valuation within 12 months of S-1
linked
33
mean inside
58%
mean blended
49%
blend direction: 27 pulled down · 3 pulled up · 3 unchanged
Top matches
  • 246_001sim=0.71SpaceX will IPO at $2 trillion valuation, raising $75 billion as large92% → 79%
  • 246_005sim=0.69OpenAI and Anthropic IPOs will head toward $1 trillion valuations.72% → 55%
  • 247_001sim=0.68xAI will IPO at $2 trillion valuation this coming summer60% → 46%

regulatory_freeze_window

0/4 historical hits — base rate 5.0% · exponential decay
Major-country regulatory pause/moratorium on AI capability research lasting >6 months
linked
27
mean inside
60%
mean blended
40%
blend direction: 23 pulled down · 1 pulled up · 3 unchanged
Top matches
  • 235_046sim=0.70Window of global AI regulation is a few months before capabilities pro55% → 42%
  • 235_007sim=0.68AI regulation window is this calendar year 2026 before chaos breaks ou60% → 45%
  • CYB_005sim=0.66'Memory becomes the moat' — by 2026, the global discourse around 'AI s72% → 30%

fab_construction_3y

7/12 historical hits — base rate 58.0%
Greenfield semiconductor fab construction within 3 years of announce
linked
19
mean inside
72%
mean blended
61%
blend direction: 19 pulled down · 0 pulled up · 0 unchanged
Top matches
  • 231_038sim=0.68TSMC fabs will come online in 5-6-7 years; Elon won't wait that long.60% → 44%
  • 231_039sim=0.65Elon will build out his own chip fabs due to his attack-constraints ap65% → 47%
  • SEM_014sim=0.65Nvidia's Arizona-based TSMC factory successfully fabricated cutting-ed98% → 86%

mars_uncrewed_landing_window

9/18 historical hits — base rate 50.0%
Successful uncrewed payload delivery to Mars surface within named transfer window
linked
14
mean inside
46%
mean blended
38%
blend direction: 7 pulled down · 4 pulled up · 3 unchanged
Top matches
  • SPC_008sim=0.67SpaceX milestone-delivery statistics: 16% on-time, 31% late, 53% undel95% → 70%
  • 246_014sim=0.66Elon's stated orbital refueling and Mars/Moon timelines will slip.62% → 40%
  • IND_016sim=0.65By late 2026, SpaceX will launch a flotilla of approximately 5 uncrewe12% → 22%

ai_capability_milestone_2y

5/15 historical hits — base rate —%
AI reaches specific named capability (intern-level / world-class programmer / etc) within 2y of stated target
linked
7
mean inside
73%
mean blended
67%
blend direction: 6 pulled down · 1 pulled up · 0 unchanged
Top matches
  • CMQ_023sim=—Tokens are the new raw material — data centers are no longer cost cent85% → 65%
  • CMQ_005sim=—AI will handle the entire software development process end-to-end with68% → 56%
  • AI_017sim=—NVIDIA will make Agentic AI the primary demand vector via two speciali78% → 75%

fda_breakthrough_label

78/200 historical hits — base rate 39.0%
FDA breakthrough designation → approval within 4 years
linked
5
mean inside
44%
mean blended
41%
blend direction: 3 pulled down · 2 pulled up · 0 unchanged
Top matches
  • CMQ_009sim=0.65Altman's claim of doctor-level AI diagnostic capability within 2-3 yea70% → 60%
  • CMQ_006sim=0.65Expert-level AI applied to biology/medicine will yield cures for cance30% → 36%
  • 242_048sim=0.64FDA will move to zero clinical trial model given enough Bayesian/compu45% → 36%

memory_supercycle_2y

4/8 historical hits — base rate —%
Named memory-market supercycle (DRAM/HBM) persists/intensifies into stated year (n>=2y forward)
linked
4
mean inside
85%
mean blended
72%
blend direction: 4 pulled down · 0 pulled up · 0 unchanged
Top matches
  • CMQ_040sim=—AI-driven demand has triggered an 'unprecedented supercycle' for domin90% → 62%
  • CMQ_036sim=—SK Hynix will maintain industry HBM lead, capturing >50% of NVIDIA's t78% → 68%
  • CMQ_033sim=—Global HBM demand will reach 32 billion Gb by 2026 — HBM is the absolu82% → 75%

recession_probability_2yr

8/30 historical hits — base rate 26.7%
NBER-defined recession beginning within 2 years of given prediction
linked
4
mean inside
34%
mean blended
30%
blend direction: 1 pulled down · 0 pulled up · 3 unchanged
Top matches
  • FUT_007sim=0.61Global economy accelerates into structural demographic depression 202655% → 41%
  • S_RECESSION_2026sim=—NBER recession declared 202620% → 20%
  • S_RECESSION_2027sim=—NBER recession declared 202730% → 30%

gpu_roadmap_on_time_1y

7/10 historical hits — base rate —%
GPU vendor product roadmap milestone (architecture / production / launch) hits within 1y of announced date
linked
3
mean inside
76%
mean blended
80%
blend direction: 1 pulled down · 2 pulled up · 0 unchanged
Top matches
  • AUT_016sim=0.54NVIDIA Rubin platform in full production by 2026 — slashes computation75% → 81%
  • CMQ_028sim=—NVIDIA's total AI chip revenue opportunity through 2027 is at least $365% → 72%
  • CMQ_026sim=—NVIDIA silicon roadmap: Blackwell (2025) → Vera Rubin (2026) → Vera Ru88% → 87%

ai_catastrophic_misuse_1y

1/20 historical hits — base rate —%
Frontier AI used in successful catastrophic-class (bio/cyber/chem) attack within 1y of capability claim
linked
2
mean inside
72%
mean blended
68%
blend direction: 2 pulled down · 0 pulled up · 0 unchanged
Top matches
  • 248_011sim=—Frontier AI models will become a terrorist, bioweapon, and chemical we65% → 60%
  • CYB_022sim=—Supply-chain attacks targeting widely utilized open-source AI packages78% → 76%

single_stock_target_1y

8/20 historical hits — base rate —%
Sell-side single-stock price target hit within 1y of issuance
linked
2
mean inside
65%
mean blended
59%
blend direction: 2 pulled down · 0 pulled up · 0 unchanged
Top matches
  • SEM_025sim=—Morgan Stanley price target $250 for Nvidia — representing 32% upside 60% → 55%
  • SEM_023sim=—No AI bubble — semiconductor positive trend will remain robust through70% → 63%

av_deployment_scale_2y

3/8 historical hits — base rate —%
AV partnership/fleet program reaches stated vehicle/city scale within 2y
linked
1
mean inside
68%
mean blended
76%
blend direction: 0 pulled down · 1 pulled up · 0 unchanged
Top matches
  • AUT_008sim=—Uber has structured partnerships integrating Waymo, May Mobility, Luci68% → 76%

hyperscaler_capex_forecast_2y

5/8 historical hits — base rate —%
Hyperscaler/AI infra capex forecast within 1-2y forward window meets-or-exceeds stated headline number
linked
1
mean inside
82%
mean blended
72%
blend direction: 1 pulled down · 0 pulled up · 0 unchanged
Top matches
  • SEM_005sim=—Stargate is a $500 billion multiyear capex program for distributed AI 82% → 72%

mega_market_cap_5y

1/10 historical hits — base rate —%
Emergence of >$50T single-company market cap within 5y window (SpaceX/NVDA-class)
linked
1
mean inside
70%
mean blended
59%
blend direction: 1 pulled down · 0 pulled up · 0 unchanged
Top matches
  • 234_014sim=—We will see $100 trillion companies in the next 5-year period70% → 59%

political_crisis_2y

3/20 historical hits — base rate —%
Stated political crisis (revolution/coup/major-instability) materializes in target country within 2y
linked
1
mean inside
62%
mean blended
55%
blend direction: 1 pulled down · 0 pulled up · 0 unchanged
Top matches
  • FUT_014sim=—'US political revolution' ranked #1 tier global risk 2026+ per Eurasia62% → 55%

taiwan_kinetic_conflict_2y

0/0 historical hits — base rate 5.0% · exponential decay
Direct PLA kinetic action against Taiwan within 2 years of forecast
linked
1
mean inside
45%
mean blended
35%
blend direction: 1 pulled down · 0 pulled up · 0 unchanged
Top matches
  • 240_030sim=0.58Tesla's independent chip production would derisk potential Chinese inv45% → 35%

Methodology

Linking
  - Embedding cosine: 1 − (n.embedding <=> rc.embedding)
  - Best match must satisfy:  cosine ≥ 0.55  AND  cosine ≥ second_best + 0.05
  - Per-class keyword guards (link_nodes_to_reference_classes.py:KEYWORD_GUARDS)
    enforce precision for narrow concepts (Mars, Taiwan, IPO, FDA, etc.)

Blending
  TRF (time remaining fraction) = (end − now) / (end − start),  clamp [0,1]
                                  with safe fallbacks for missing dates
  w_inside  = 0.3 + 0.7 × (1 − TRF)            range [0.3, 1.0]
  w_outside = 1 − w_inside                       range [0.0, 0.7]
  blended_logit = w_inside × logit(inside_prob) + w_outside × logit(base_rate)
  blended_prob  = sigmoid(blended_logit)

Architecture
  - inside_posterior accumulates Bayesian evidence (read on next intake as prior)
  - blended_posterior is the displayed value (nodes.current_prob)
  - Both persisted in prob_history.metadata for full audit trail