Matheo-b16 — What Changed Between OOv2 and OOv3#
The short version: no number changed. OOv3 is OOv2’s argument in a different shape — a lean ring of eleven sections a general reader can follow, with the rigour moved into Supporting Information Appendices. Nothing was recomputed. One thing was added that OOv2 does not have, and a little numerical detail was left behind, which is why this page exists at all.
Why bother with a changelog for a rewrite? Because “complete rewrite, same numbers” is precisely the claim a reader cannot check by reading. When the prose is rebuilt from the ground up, a reader who knows OOv2 has no cheap way to tell a re-ordering from a quiet revision — so the claim has to be stated somewhere, with a method attached, and pointed at what would falsify it. That is this page. It is deliberately short: unlike the MMv5 → OOv2 changelog, there is nothing here anyone needs to un-remember, and length would imply an instability that the check does not find.
1. The arithmetic did not move#
Method, so this is checkable rather than asserted. Every numeric token in both drafts was extracted and the two sets compared (years and file paths excluded). Result: 157 numbers appear in both. Only four appear in OOv3 and not in OOv2, and every one of them belongs to the new figures rather than to the argument:
New in OOv3 |
What it is |
|---|---|
|
image widths in two figure directives — presentation, not content. |
|
the pooled histogram (Figure 3) shows 120 runs — 40 at each of the three crisis rates already reported. Nothing new was simulated; the existing runs are shown together. |
|
an axis label read off in a caption (“note the x-axes: 60, 200 and 600 years”). |
No headline value changed. The annual probability of onset is 2.90 percent (about 1 in 34) in both. The exact median is 21.1 years and the mean 30.3 years in both. The car-crash multiple is 58x in both, the forty-year figure 73 percent in both, the crisis rate 0.1/yr and the per-crisis death probability 1/3 in both.
2. What changed is the shape#
OOv2 |
OOv3 |
|
|---|---|---|
Ring sections |
40, nested three deep (§2.0, §2.0a, §2.1 … §2.10) |
11, flat, each readable on its own |
Appendices |
none — everything inline |
19 Supporting Information Appendices (SIA-1 – SIA-14, SIA-X1 – SIA-X5) |
Words |
~24,900 |
~22,100 |
Typeset length |
55 pages |
47 pages |
The intent, in LLoL’s own framing when he commissioned it: the paper looked extremely complicated even though the core argument fits on a page or two. OOv3 is a ring structure (after Mary Douglas) — the decision sits at the centre, and every piece of rigour that was interrupting it moves outward into an SIA where a reader who wants it can find it whole.
3. What OOv3 adds that OOv2 does not have#
This is the part a reader of OOv2 would otherwise miss, and it is why OOv3 is not only a re-ordering. Added 2026m07d24, after the restructure:
A sustained answer to the strongest published objection. Tertrais (2017) is engaged at length in SIA-11: his “in all the incidents, safety mechanisms worked” argument is granted, and the answer is that the calibration already excludes the incidents procedure caught, so the forecast is built on the residue after the working system is credited in full. His own closing proposal — that the debate “should revolve around the costs and benefits equation” — now opens SIA-11 as the paper’s warrant.
Two objections answered that no earlier draft had faced: the Pascalian-mugging objection (“by this logic one must fear everything”), and the escalation-staircase objection (“a single use need not become Armageddon”), the latter recorded as a genuine open flank rather than closed.
A sharper Section 10. The decision is now stated as invariant across the whole range the record can bear: to overturn it an optimist cannot argue the rate is low, only that it is exactly zero — which no one claims. The ruin-versus-risk distinction the arithmetic leans on is now named and cited.
A robustness note on Arkhipov (Section 6), granting that the dramatic version of the B-59 story is contested and showing the forecast does not stand on it.
Six figures, replacing the two placeholder boxes OOv2 left open — including the escape route drawn on the model, and the escape as code.
None of these changed a number. They change how much weight the argument can take.
4. What OOv2 still carries that OOv3 does not#
Stated in this direction too, because a changelog that only lists gains is an advertisement. The lean ring left some derived detail behind. Where a reader wants these, OOv2 remains the fuller record:
The expected blow-up counts per forty runs — 0.35 optimistic, 1.16 base, 3.34 pessimistic. OOv3 keeps the finding in words (“roughly one run in forty”) but not the exact expectations.
The full Petrov-removed sensitivity row. Both drafts report that dropping Petrov leaves three excursions, a crisis rate of 0.075/yr and 2.19 percent per year (1 in 46); OOv2 additionally gives that scenario’s mean, median and 10-, 20- and 40-year cumulative probabilities.
The US road-death baseline (about 1 in 6,700 per person-year), which OOv2 carried alongside the global one that both drafts use for the headline comparison.
Whether that trimming is right is a judgement about audience, not about arithmetic — but it is a trim, and a reader deserves to know which direction to go for the missing digits.
5. Why OOv2 is kept#
OOv2 is superseded, not overwritten — a locked decision for this study, and one of the reasons the drafts carry version numbers at all. It stays readable for three purposes: it is the fuller numeric record noted above; it is the document the MMv5 → OOv2 changelog describes; and, in LLoL’s own words when commissioning the rewrite, its 55 pages are an artefact of how working with AI can lead to over-complication — worth preserving as evidence rather than tidied away.
Both drafts are unreviewed by LLoL and are offered not to be believed, but to be checked. The full audit trail — the restructure plan, the insert-and-figure execution pass, and the independent check of that pass — lives in the b16 study LLogs. #AuditTheMath