← Evidence index · 中文 · Main post
Each section: Claim · Why · Evidence · Analogue · Would update if · Conf (H/M/L).
Parent: correlation matrix · main forecast §Branch model
Purpose: Load-bearing X → Y links we treat as causal (not mere correlation). When X moves in evidence, update Y before re-stitching branch weights.
Status: Working list 2026-07-04 — integrate into Layer B2 of public methodology.
How this differs from Clusters A–H
| Clusters A–H | Causal edges (this doc) | |
|---|---|---|
| Question | Which P’s move together? | Which P’s cause which, in what direction? |
| Stitching | Branch weights encode joint mass | Conditional update rules on linked P’s |
| Wrong move | Multiply marginals as independent | Ignore downstream when upstream flips |
Rule: Edges inform P updates and branch weight nudges. They do not replace the three-branch mixture in my pdoom.
Top 10 edges
| # | Cause → Effect | Mechanism (one line) | Update rule (modeling) | Cluster | If edge is stronger than modeled |
|---|---|---|---|---|---|
| E1 | Tier 2 bio live (Ci ~C5) → P(Tier 3 before screening | Tier 2) | Capability gate: misuse path only opens once assist-tier tools exist | Treat T2 as prerequisite; do not draw Tier-3-before-screening mass if T2 timeline slips | C | Bio branch weight ↓; extinction ↓ ~2–4pp if T2 stalls 2+ yr |
| E2 | Hidden near-miss stays hidden (N10, P~0.55) → P(BMIA / mandatory screening by 2027-12) | Successful internal containment → no Kingdon window (“inverse Y2K”) | BMIA P ↓ ~0.05–0.15 when hidden-ratio evidence ↑; apply with S2 modifier | H, F | Bio branch weight ↑; extinction ↑ ~1–2pp |
| E3 | Hidden near-miss (N10) → P(Tier 3 before screening | Tier 2) | More real near-misses than public Trigger E → screening lags real risk | Tier-3-before-screening ↑ ~0.02–0.06 per hidden-ratio tranche | C, H | Bio P(ext|branch) ↑; headline extinction ↑ ~1–2pp |
| E4 | US governance paralysis S2 (P~0.55) → P(BMIA) and N6 gridlock | Federal capacity collapse: bills stall, EO-only framework | ×0.7 on BMIA (and related N2 tails) when S2 elevated — already in crosscut table | F | Bio branch ↑ ~2–3pp if S2 persists through 2028 |
| E5 | Weight theft / natsec shock (N3) → P(race accelerates post-theft) → P(no federal pause) | Crisis framing → faster release, weaker halt coalitions | Post-theft race P ↑ → no-pause crux ↑ ~0.03–0.08 (Cluster A block) | A | Misalign branch weight ↑; extinction ↑ ~1–3pp |
| E6 | Whistleblower modal (N4: oversight not halt, P~0.58 given E) → P(no training halt through 2028) | Scare → audits/hearings, training continues | Whistleblower modal ↑ → no-pause ↑ ~0.04–0.10; not independent of N11 | A, D | Whimper ↑; misalign conditional ↑ |
| E7 | Corp governance hollowing (N11, P~0.65 modal) → P(halt | internal crisis ~0.12) → P(no shutdown at ASI threshold) | Board/exec capture; safety veto hollowed | Hollowing ↑ → halt|crisis ↓ → no-shutdown ↑ ~0.05–0.10 | H | Misalign P(ext|branch) ↑ ~2–4pp |
| E8 | GAAIA preemption wins (N6) → SB53 / state transparency duty weakened → hidden near-miss sustained | Federal weak standard overrides state incident reporting | Preemption ↑ → N10 hidden-ratio ↑ ~0.05; feeds E2/E3 | A, F | Bio + disclosure joint tail ↑ |
| E9 | Open-weight + tamper tail (N5 T5-B) → P(Tier 3 before screening | Tier 2) | Public weights lower barrier for skilled misuse | Open-weight tail ↑ → Tier-3-before-screening ↑ ~0.03–0.08 | C | Bio branch weight ↑ ~1–2pp |
| E10 | C7–C9 internal RSI + cloud/software locus (N12, P~0.55) → misalign branch weight and P(extinction | misalign tail) | Superhuman lab R&D + opaque stack → harder shutdown, faster takeoff geometry | N12 cloud-first ↑ → misalign branch +3–6pp weight (see Node 12); ties to N7 race speed | H, E | Extinction ↑ ~2–5pp if cloud RSI + no halt joint |
Worked propagation (example: E2 → E3 → branch)
Observe: voluntary Tier-2 near-miss disclosure (falsifier for N10)
→ E2: hidden-ratio ↓
→ E3: P(Tier 3 before screening | Tier 2) ↓ ~0.03–0.06
→ Bio branch weight 20% → ~17–18%
→ Headline extinction ~19% → ~17–18% (before other edges)Full bio path in blog §Worked example; this table generalizes the update order.
Edges we deliberately do not treat as one-way
| Pair | Why |
|---|---|
| BMIA pass ↔ race continues | Cluster B partial ↓ with A: screening can strengthen while race continues (modal storyline) |
| EU friction ↔ US pause | N9 can bind deployment without US training halt |
| Interp prod halt ↔ deception crux | Partial ↓ on misalign conditional only — not headline alone |
Modeling checklist (when evidence moves)
- Identify which edge(s) fired (table above).
- Update downstream P using stated ranges (not full independence).
- Nudge branch weights (modal / bio / misalign) if joint story shifts.
- Re-run mixture; apply correlation discount to headline if raw sum moved.
- Log in my pdoom update table + future
forecast_tracker.md.
Mermaid (Layer B2 sketch)
flowchart TD
T2[Tier 2 live E1] --> T3[P Tier 3 before screening]
N10[Hidden near-miss E2 E3] --> BMIA[P BMIA]
N10 --> T3
S2[US paralysis E4] --> BMIA
N3[Weight theft E5] --> NP[P no pause]
N4[Whistle modal E6] --> NP
N11[Corp hollowing E7] --> NS[P no shutdown]
N6[GAAIA E8] --> N10
N5[Open-weight tail E9] --> T3
N12[RSI locus E10] --> MIS[misalign branch]
T3 --> BIO[bio branch weight]
BMIA --> BIO
NP --> MOD[modal vs misalign weights]
NS --> MIS
BIO --> MIX[Layer C mixture]
MIS --> MIX
MOD --> MIX
Update log
| Date | Change |
|---|---|
| 2026-07-04 | Initial top-10 list for Layer B2 integration |
| 2026-07-04 | Phase 2c lock — linked forecast_tracker.md; headline 19/24/30 unchanged |