Analysis of HC-4 - Bound on the exogenous terminal-catastrophe rate

What the analysis says

The cluster asks whether the Tegmark-Bostrom bound — exogenous terminal-catastrophe rate below ~1/Gyr at 99.9% confidence — holds. Two members, exhaustive by construction: H-5 - The exogenous terminal-catastrophe rate is below about 1 per Gyr at 99.9 percent confidence and its negation H-10 - The exogenous terminal-catastrophe rate is not bounded below 1 per Gyr. The prior deliberately sits near even odds, [0.504, 0.496], built as a product of three modeling hinges (distribution model 0.7 × uncorrelatedness 0.8 × derivation 0.9). A single edge, E-6 - O-10 × HC-4 — Earth’s formation date is typical, not early, does all the updating: Earth’s typical formation date (O-10 - Habitable-planet formation times span many Gyr and Earth’s formation date is unremarkably late) is likely under H-5 (0.7) and unlikely under H-10 (0.15, a rate-mixture average), at trust 0.8 (the S-23 cap). Posterior: [0.7615, 0.2385]. A-6 - Observer selection makes naive survival-based risk bounds uninformative (corrected/checked) removes the anthropic defeater rather than adding a discount, since the method compares among survivors.

What the model may not capture

1. The whole bound stands on one paper’s semi-empirical planet-formation-time distribution (Lineweaver-type), and the 99.9% claim lives in its weakly constrained tails; the prior prices this (p_dist_model 0.7) but that number is a judgment with no real reference class. The cluster note itself names this as the dominant failure route into H-10.
2. H-10 is a wide disjunction (any hinge fails, any rate above 1/Gyr), so its 0.15 likelihood is a mixture average over a range nothing in the model resolves — a marginally-failing bound (rate ~1-3/Gyr) is far more compatible with the datum than the two-member posterior split suggests, and the carve hides that.
3. Being ¬H-5, the answer is trivially “on the list”; the real out-of-model risk is that the typicality-test framing itself is wrong (e.g. observer-weighting depends on habitability evolution, not just survival), which would land in H-10 but for reasons no block prices.
4. Exogeneity restriction: the bound says nothing about human-modulated rates — correctly firewalled in the cluster’s independence note, but a reader could over-read the backstop.

What would help

1. An independent planet-formation-history model’s tail behaviour compared against Lineweaver’s — exists, unread; would most directly move p_dist_model.
2. The bound recomputed at lower confidence (e.g. 99%) to see how tail-sensitive the headline is — exists, unread (in S-23’s derivation).
3. A rate-resolved carve of H-10 (marginal vs. gross failure) — does not exist in this run; step-4 rework.

Confusions and contradictions

E-6’s “Why this is evidence” and A-7 - Earth’s unremarkably late formation time yields a selection-bias-free catastrophe-rate bound near 1 per Gyr state opposite tilt directions: E-6 says a high rate pushes surviving observers toward early-formed planets (later formation = larger past-light-cone exposure); A-7’s step-6-checked derivation says the observer-weighted distribution tilts late (late formers have less elapsed time at risk). These are different hazard models (exposure since Big Bang vs. since formation) and cannot both be the paper’s argument; the step-6 “checked” trace endorsed one while the edge asserts the other. The update direction survives either way — a typical date is the modal outcome under H-5 and surprising under a high rate whichever way the tilt runs — so the numbers stand, but the contradiction is real and shipped unfixed; the fix is a step-6/step-8 re-trace against S-23, not an edit here.