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 106.4% of expected length, 136.3 wpm (median cue 140.9 wpm). Script source: public/videos/from-predicted-crystal-to-commercial-cell.vtt at commit 4641d96 (pre-overwrite narration prose)

1
00:00:00.000 --> 00:00:03.150
A predicted crystal is not a material.

2
00:00:03.450 --> 00:00:06.130
A synthesized powder is not a cell.

3
00:00:06.430 --> 00:00:12.060
A cell is not a fleet, a filter, a fertilizer plant, or a solar farm.

4
00:00:12.360 --> 00:00:19.350
Between a computational screen and a commercial product stands a chain of handoffs — and most predictions never survive it.

5
00:00:19.650 --> 00:00:24.850
Millions of predicted crystals shrink to a few hundred independently synthesized.

6
00:00:25.150 --> 00:00:27.960
The conventional answer is a bigger generator.

7
00:00:28.260 --> 00:00:32.600
But volume without trust only amplifies the same handoff failure.

8
00:00:32.900 --> 00:00:39.080
What is missing is a correction and verification layer that makes predictions trustworthy before they reach the lab.

9
00:00:39.380 --> 00:00:42.920
Consider cobalt-free lithium-manganese-rich cathodes.

10
00:00:43.220 --> 00:00:56.400
A corrected ranking sifts more than a million compositions, then passes the best candidates through synthesis, atomic-layer coating, pouch-cell fabrication, and finally into an automaker's prismatic cell line.

11
00:00:56.700 --> 00:01:00.880
Each handoff carries a bounded error budget, or it breaks.

12
00:01:01.180 --> 00:01:08.390
The correction measures error as a field over local coordination, anchored to measured observables and pinned to zero in bulk.

13
00:01:08.690 --> 00:01:12.820
Analytic forces apply it at runtime, so simulations stay fast.

14
00:01:13.120 --> 00:01:22.840
Across 36 blind model-material pairs, corrected surface energies correlate at zero point nine zero six, with zero fitted parameters.

15
00:01:23.140 --> 00:01:29.150
The verification layer separates supported predictions from unsupported ones with machine-checked proofs.

16
00:01:29.450 --> 00:01:34.830
It can flag impossibility, not just uncertainty, before synthesis starts.

17
00:01:35.130 --> 00:01:39.900
Experimental budgets are finite; false positives are expensive.

18
00:01:40.200 --> 00:01:44.160
The layer makes sure the work goes to candidates that actually deserve it.

19
00:01:44.460 --> 00:01:53.030
The same logic reaches solid-state electrolytes, direct-air-capture frameworks, ammonia catalysts, and lead-free perovskites.

20
00:01:53.330 --> 00:01:57.720
Different markets, different partners, same geometry of wrongness.

21
00:01:58.020 --> 00:02:01.900
In every chain, the scarce resource is not prediction volume.

22
00:02:02.200 --> 00:02:06.453
It is the trust that turns a predicted crystal into a commercial cell.