tiny-geometry-f1-whole-path-e13259f-cf59a64e-bd6c4bbc
completed —
Hypothesis
whole-path donor replacement recovers held-out geometry under matched fixed-topology training
Measured behaviour
Table view
| optimizer step | train accuracy |
|---|---|
| 1 | 0.0006 |
| 10 | 0.1427 |
| 20 | 0.5272 |
| 30 | 0.8226 |
| 40 | 0.9206 |
| 50 | 0.9706 |
| 60 | 0.9873 |
| 70 | 0.9952 |
| 80 | 0.9969 |
| 90 | 0.9987 |
| 100 | 0.9996 |
| 110 | 0.9995 |
| 120 | 1.0000 |
| 130 | 1.0000 |
| 140 | 1.0000 |
| 150 | 1.0000 |
| 160 | 1.0000 |
| 170 | 1.0000 |
| 180 | 1.0000 |
| 190 | 1.0000 |
| 200 | 1.0000 |
| 210 | 1.0000 |
| 220 | 1.0000 |
| 230 | 1.0000 |
| 240 | 1.0000 |
| 250 | 1.0000 |
| 260 | 1.0000 |
| 270 | 1.0000 |
| 280 | 1.0000 |
| 290 | 1.0000 |
| 300 | 1.0000 |
| 310 | 1.0000 |
| 320 | 1.0000 |
Measured result
Verbatim from the run's summary.json.
cases
diverse-four
checkpoint_sha256
3c36ee57d3180b66…config
corruption_kind
whole_path_geometry
corruptions_per_icon
4
heldout_corruptions_per_icon
4
icon_count
4
max_loss_ratio
0.3
min_heldout_accuracy
0.75
min_train_accuracy
0.9
name
diverse-four
resample_each_step
true
resume_step
160
seed
1702
steps
320
continuous_loss_sha256
bc88637b296c90e3…final
accuracy
1.0
changed_accuracy
1.0
changed_correct
550
changed_total
550
correct
9424
loss
0.0010140867670997977
retained_accuracy
1.0
retained_correct
8874
retained_total
8874
total
9424
final_model_sha256
e0defadb6bab3480…heldout_examples
16
initial
accuracy
0.0006366723259762309
changed_accuracy
0.0036363636363636364
changed_correct
2
changed_total
550
correct
6
loss
14.969106912612915
retained_accuracy
0.000450755014649538
retained_correct
4
retained_total
8874
total
9424
loss_ratio
6.910615321289133e-05
passes_predeclared_criteria
true
resume_exact
true
resumed_loss_sha256
bc88637b296c90e3…train_examples
16
train_final
accuracy
1.0
changed_accuracy
1.0
changed_correct
569
changed_total
569
correct
9424
loss
0.0010510510910535231
retained_accuracy
1.0
retained_correct
8855
retained_total
8855
total
9424
code_identity
geometry_sha256
2d61f4c07725b827…git_commit
e13259fd4086114b6188a4c32dfc79e4a9ea45a2
packed_sha256
9b12879ba3a18236…program_sha256
0d02b532b1dbe573…tiny_study_sha256
a108fc4f7043beb3…config_sha256
cf59a64e05549fd6…deterministic_algorithms
true
device
cpu
fixture
data/manifests/openmoji-17.0.0-whole-path-fixture-v3.json
fixture_sha256
bd6c4bbc3440aeec…metrics_sha256
2d9bdd2af5233acc…model
d_model
48
feedforward
96
heads
4
layers
2
parameters
241072
scope
fixed-topology geometry-only diagnostic
passes_predeclared_gate_candidate
true
render_metrics_sha256
5a7443d4ce843954…schema_version
1
study_version
tiny-geometry-f1-whole-path
torch_version
2.8.0+cpu
Visual output
State transitions
- completed2026-08-30T19:58:37Zwhole_path_treatment_reaches_perfect_heldout_and_changed_token_recovery_on_its_distinct_fixture
Run record
Verbatim from runs/tiny-geometry-f1-whole-path-e13259f-cf59a64e-bd6c4bbc/run.yaml, the record committed before launch.
schema_version
1
run_id
tiny-geometry-f1-whole-path-e13259f-cf59a64e-bd6c4bbc
state
completed
hypothesis
whole-path donor replacement recovers held-out geometry under matched fixed-topology training
git_commit
e13259f
config
configs/learning/tiny-geometry-f1-whole-path.yaml
config_sha256
cf59a64e05549fd6…dataset_manifest
data/manifests/openmoji-17.0.0-whole-path-fixture-v3.json
dataset_manifest_sha256
bd6c4bbc3440aeec…worker
gtc-local-cpu
resource_cap
external_spend_usd
0
optimizer_steps
320
command
.venv/bin/python -m mojidiff.learning.tiny_study --config configs/learning/tiny-geometry-f1-whole-path.yaml
completed_at
2026-08-30 19:58:37+00:00
result
heldout and changed-token recovery both reached 1.0; checkpoint continuation exact
summary_sha256
a28390b79901cfd0…