card
Browse files
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-
|
| 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 |
|
| 34 |
## 2026-11-03: stricter novelty rule
|
|
@@ -46,11 +46,11 @@ METHODS.md §3 and §7, and in the `prior_reports` column of `dft_results.csv`.
|
|
| 46 |
|
| 47 |
| | count |
|
| 48 |
|---|---|
|
| 49 |
-
| Structures generated |
|
| 50 |
-
| Kept as "discoveries" (stable by ML, and the composition is not reported in any database or paper we check) |
|
| 51 |
-
| ML hull check finished |
|
| 52 |
| ML hull check at or below 0 eV/atom | 1 |
|
| 53 |
-
| DFT jobs (passes / fails / still running or queued) |
|
| 54 |
|
| 55 |
### How the prompts were chosen, and how each way does
|
| 56 |
On 2026-11-01 the autopilot switched from random formulas to mostly substitution-based prompts (see METHODS.md).
|
|
@@ -61,16 +61,16 @@ New rows need a few nights of hull checks before their numbers mean much.
|
|
| 61 |
|
| 62 |
| prompt mode | skipped as already known | structures | new | ML hull checked | median e_hull (eV/atom) | within 0.05 | DFT passes |
|
| 63 |
|---|---|---|---|---|---|---|---|
|
| 64 |
-
| random formula (before 11-01) | 0 | 3713 | 696 |
|
| 65 |
-
| surprise (before 11-01) | 0 | 1681 | 101 |
|
| 66 |
-
| substitution |
|
| 67 |
-
| random |
|
| 68 |
-
| surprise |
|
| 69 |
-
| control | 0 |
|
| 70 |
| human | 0 | 10 | 0 | 0 | - | 0 | 0 |
|
| 71 |
|
| 72 |
-
ML verdicts: unstable
|
| 73 |
-
Novelty at generation time: skipped
|
| 74 |
|
| 75 |
### Files
|
| 76 |
- `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 2026-11-09. 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 |
|
| 34 |
## 2026-11-03: stricter novelty rule
|
|
|
|
| 46 |
|
| 47 |
| | count |
|
| 48 |
|---|---|
|
| 49 |
+
| Structures generated | 5832 |
|
| 50 |
+
| Kept as "discoveries" (stable by ML, and the composition is not reported in any database or paper we check) | 894 |
|
| 51 |
+
| ML hull check finished | 620 |
|
| 52 |
| ML hull check at or below 0 eV/atom | 1 |
|
| 53 |
+
| DFT jobs (passes / fails / still running or queued) | 12 (3 / 2 / 7) |
|
| 54 |
|
| 55 |
### How the prompts were chosen, and how each way does
|
| 56 |
On 2026-11-01 the autopilot switched from random formulas to mostly substitution-based prompts (see METHODS.md).
|
|
|
|
| 61 |
|
| 62 |
| prompt mode | skipped as already known | structures | new | ML hull checked | median e_hull (eV/atom) | within 0.05 | DFT passes |
|
| 63 |
|---|---|---|---|---|---|---|---|
|
| 64 |
+
| random formula (before 11-01) | 0 | 3713 | 696 | 486 | 0.39 | 0 | 0 |
|
| 65 |
+
| surprise (before 11-01) | 0 | 1681 | 101 | 66 | 0.117 | 11 | 3 |
|
| 66 |
+
| substitution | 195 | 261 | 68 | 54 | 0.216 | 1 | 0 |
|
| 67 |
+
| random | 93 | 78 | 21 | 12 | 0.407 | 0 | 0 |
|
| 68 |
+
| surprise | 148 | 62 | 8 | 2 | 0.158 | 0 | 0 |
|
| 69 |
+
| control | 0 | 27 | 0 | 0 | - | 0 | 0 |
|
| 70 |
| human | 0 | 10 | 0 | 0 | - | 0 | 0 |
|
| 71 |
|
| 72 |
+
ML verdicts: unstable 2104, rock solid 1462, plausible 1307, borderline 776, falls apart 183.
|
| 73 |
+
Novelty at generation time: skipped 3063, known 977, new composition 894, known composition 893, rediscovered 5.
|
| 74 |
|
| 75 |
### Files
|
| 76 |
- `structures.jsonl.gz`: one JSON object per structure: formula, space group, CIF, how the prompt was chosen
|