WEBVTT

NOTE Narration transcript. Text is the script synthesized by minimax (speech-2.8-hd, voice English_expressive_narrator); each cue was synthesized as its own audio file and is timed by that file's measured duration. Verified at 113.7% of expected length, 127.6 wpm (median cue 135.7 wpm). Script source: public/videos/cement-concrete-and-the-weight-of-the-built-world.vtt at commit 4641d96 (pre-overwrite narration prose)

1
00:00:00.000 --> 00:00:04.340
Concrete is the most widely used manufactured material on Earth.

2
00:00:04.640 --> 00:00:13.560
The cement that binds it emits about 2.8 gigatonnes of CO₂ every year, roughly 8% of global anthropogenic emissions.

3
00:00:13.860 --> 00:00:18.680
And about 60% of that comes from limestone calcination, not fuel.

4
00:00:18.980 --> 00:00:22.910
Clean kilns help, but they cannot change the chemistry.

5
00:00:23.210 --> 00:00:29.990
As long as limestone is the main calcium feedstock, calcination releases CO₂ regardless of heat source.

6
00:00:30.290 --> 00:00:39.500
Real decarbonization needs new chemistries: alternative binders, alternative clinkers, and CO₂-cured concrete.

7
00:00:39.800 --> 00:00:45.900
All three routes depend on amorphous gels, metastable hydrates, and multi-component oxides.

8
00:00:46.200 --> 00:00:53.410
These are the exact environments where universal machine-learning potentials soften the energy surface by 15 to 60 percent.

9
00:00:53.710 --> 00:00:59.890
A dissolution energy off by 15 percent can invert the ranking of two binder formulations.

10
00:01:00.190 --> 00:01:06.080
Alternative binders like slag and geopolymers are glasses and gels, not unit cells.

11
00:01:06.380 --> 00:01:12.520
Alternative clinkers span multi-component oxide spaces with metastable essential phases.

12
00:01:12.820 --> 00:01:19.770
CO₂-cured concrete is limited by carbonate diffusion barriers at under-coordinated surface sites.

13
00:01:20.070 --> 00:01:22.880
Each front needs trustworthy rankings.

14
00:01:23.180 --> 00:01:31.320
The correction measures error as a field over local coordination, anchored to three observables and pinned to zero in bulk.

15
00:01:31.620 --> 00:01:37.800
It applies analytic corrections at runtime, keeping molecular dynamics nearly as fast as the base model.

16
00:01:38.100 --> 00:01:46.970
Across 36 blind model-material pairs, corrected predictions correlate at 0.906 with zero fitted parameters.

17
00:01:47.270 --> 00:01:55.370
The verification layer uses build-locked Lean 4 theorems to separate supported predictions from synthesis-dependent ones.

18
00:01:55.670 --> 00:02:02.960
For cement, that matters because the most useful phases are metastable and cannot be judged by convex-hull stability alone.

19
00:02:03.260 --> 00:02:06.120
Cement is not a special case.

20
00:02:06.420 --> 00:02:14.010
It is one instance of a geometry of wrongness that extends across batteries, catalysts, sorbents, and refrigerants.

21
00:02:14.310 --> 00:02:17.290
The bottleneck is not a shortage of candidates.

22
00:02:17.590 --> 00:02:20.815
It is a shortage of predictions laboratories can trust.