zer0contextlost commited on
Commit
7717729
·
verified ·
1 Parent(s): 8072169
Files changed (1) hide show
  1. README.md +12 -12
README.md CHANGED
@@ -28,7 +28,7 @@ AI generators now propose crystals much faster than anyone checks them with phys
28
  open record of what happens when you do check: most candidates that look stable to a machine-learned model turn
29
  out to be known compounds, or less stable than their real competitors, once the competitors are taken seriously.
30
 
31
- Built 2026-11-18. Rebuilt nightly while the pipeline runs. Mirrors: [Hugging Face](https://huggingface.co/datasets/zer0contextlost/crystal-lab-data);
32
  raw Quantum ESPRESSO calculations parsed and searchable on [NOMAD](https://nomad-lab.eu/prod/v1/gui/search/entries?upload_id=kNO0ljjeTyK0l_lX9522Ww) (snapshot of 2026-11-01), plus the #1856 SrCaCdP2 phonon runs ([upload of 2026-11-06](https://nomad-lab.eu/prod/v1/gui/search/entries?upload_id=VJuk3Rs9RjKGtpU7TdSF3g))
33
  and all #2571 BaCdGePd2 runs ([upload of 2026-11-10](https://nomad-lab.eu/prod/v1/gui/search/entries?upload_id=7-krAnoRQlyS_2i0FVfjIw)).
34
 
@@ -65,11 +65,11 @@ METHODS.md §3 and §7, and in the `prior_reports` column of `dft_results.csv`.
65
 
66
  | | count |
67
  |---|---|
68
- | Structures generated | 6074 |
69
- | Kept as "discoveries" (stable by ML, and the composition is not reported in any database or paper we check) | 977 |
70
- | ML hull check finished | 804 |
71
  | ML hull check at or below 0 eV/atom | 1 |
72
- | DFT jobs (passes / fails / still running or queued) | 17 (6 / 4 / 7) |
73
 
74
  ### How the prompts were chosen, and how each way does
75
  On 2026-11-01 the autopilot switched from random formulas to mostly substitution-based prompts (see METHODS.md).
@@ -80,16 +80,16 @@ New rows need a few nights of hull checks before their numbers mean much.
80
 
81
  | prompt mode | skipped as already known | structures | new | ML hull checked | median e_hull (eV/atom) | within 0.05 | DFT passes |
82
  |---|---|---|---|---|---|---|---|
83
- | random formula (before 11-01) | 0 | 3713 | 696 | 580 | 0.395 | 1 | 0 |
84
- | surprise (before 11-01) | 0 | 1681 | 101 | 84 | 0.113 | 17 | 6 |
85
- | substitution | 361 | 422 | 124 | 106 | 0.216 | 3 | 0 |
86
- | random | 177 | 124 | 37 | 24 | 0.318 | 2 | 0 |
87
- | surprise | 291 | 77 | 19 | 10 | 0.125 | 1 | 0 |
88
  | control | 0 | 47 | 0 | 0 | - | 0 | 0 |
89
  | human | 0 | 10 | 0 | 0 | - | 0 | 0 |
90
 
91
- ML verdicts: unstable 2182, rock solid 1563, plausible 1341, borderline 795, falls apart 193.
92
- Novelty at generation time: skipped 3170, known 988, new composition 977, known composition 934, rediscovered 5.
93
 
94
  ### Files
95
  - `structures.jsonl.gz`: one JSON object per structure: formula, space group, CIF, how the prompt was chosen
 
28
  open record of what happens when you do check: most candidates that look stable to a machine-learned model turn
29
  out to be known compounds, or less stable than their real competitors, once the competitors are taken seriously.
30
 
31
+ Built 2026-11-19. Rebuilt nightly while the pipeline runs. Mirrors: [Hugging Face](https://huggingface.co/datasets/zer0contextlost/crystal-lab-data);
32
  raw Quantum ESPRESSO calculations parsed and searchable on [NOMAD](https://nomad-lab.eu/prod/v1/gui/search/entries?upload_id=kNO0ljjeTyK0l_lX9522Ww) (snapshot of 2026-11-01), plus the #1856 SrCaCdP2 phonon runs ([upload of 2026-11-06](https://nomad-lab.eu/prod/v1/gui/search/entries?upload_id=VJuk3Rs9RjKGtpU7TdSF3g))
33
  and all #2571 BaCdGePd2 runs ([upload of 2026-11-10](https://nomad-lab.eu/prod/v1/gui/search/entries?upload_id=7-krAnoRQlyS_2i0FVfjIw)).
34
 
 
65
 
66
  | | count |
67
  |---|---|
68
+ | Structures generated | 6078 |
69
+ | Kept as "discoveries" (stable by ML, and the composition is not reported in any database or paper we check) | 980 |
70
+ | ML hull check finished | 834 |
71
  | ML hull check at or below 0 eV/atom | 1 |
72
+ | DFT jobs (passes / fails / still running or queued) | 18 (8 / 4 / 6) |
73
 
74
  ### How the prompts were chosen, and how each way does
75
  On 2026-11-01 the autopilot switched from random formulas to mostly substitution-based prompts (see METHODS.md).
 
80
 
81
  | prompt mode | skipped as already known | structures | new | ML hull checked | median e_hull (eV/atom) | within 0.05 | DFT passes |
82
  |---|---|---|---|---|---|---|---|
83
+ | random formula (before 11-01) | 0 | 3713 | 696 | 589 | 0.393 | 1 | 0 |
84
+ | surprise (before 11-01) | 0 | 1681 | 101 | 85 | 0.109 | 17 | 7 |
85
+ | substitution | 368 | 425 | 126 | 119 | 0.211 | 4 | 0 |
86
+ | random | 177 | 125 | 38 | 27 | 0.318 | 2 | 0 |
87
+ | surprise | 291 | 77 | 19 | 14 | 0.125 | 2 | 1 |
88
  | control | 0 | 47 | 0 | 0 | - | 0 | 0 |
89
  | human | 0 | 10 | 0 | 0 | - | 0 | 0 |
90
 
91
+ ML verdicts: unstable 2183, rock solid 1566, plausible 1341, borderline 795, falls apart 193.
92
+ Novelty at generation time: skipped 3171, known 988, new composition 980, known composition 934, rediscovered 5.
93
 
94
  ### Files
95
  - `structures.jsonl.gz`: one JSON object per structure: formula, space group, CIF, how the prompt was chosen