Dataset Viewer

Split (1)

train

The dataset viewer is not available for this split.

Cannot load the dataset split (in streaming mode) to extract the first rows.

Error code:   StreamingRowsError
Exception:    CastError
Message:      Couldn't cast
base_revision: string
compression: struct<output_rank: int64, svd_clamp_quantile: double>
  child 0, output_rank: int64
  child 1, svd_clamp_quantile: double
global_weights: struct<action-planner-v1: double, parallel-reasoning-v1: double, qwen3-8b-stateless-knapsack-lora-se (... 41 chars omitted)
  child 0, action-planner-v1: double
  child 1, parallel-reasoning-v1: double
  child 2, qwen3-8b-stateless-knapsack-lora-seed7-v1: double
  child 3, qwen3-8b-test-v1: double
layer_group_overrides: struct<group_1: struct<parallel-reasoning-v1: double>, group_2: struct<parallel-reasoning-v1: double (... 103 chars omitted)
  child 0, group_1: struct<parallel-reasoning-v1: double>
      child 0, parallel-reasoning-v1: double
  child 1, group_2: struct<parallel-reasoning-v1: double>
      child 0, parallel-reasoning-v1: double
  child 2, group_3: struct<action-planner-v1: double, parallel-reasoning-v1: double, qwen3-8b-test-v1: double>
      child 0, action-planner-v1: double
      child 1, parallel-reasoning-v1: double
      child 2, qwen3-8b-test-v1: double
merge: struct<combination_type: string, random_seed: int64>
  child 0, combination_type: string
  child 1, random_seed: int64
output: struct<adapter_name: string, dtype: string>
  child 0, adapter_name: string
  child 1, dtype: string
schema_version: int64
selected_adapters: list<item: string>
  child 0, item: string
source_snapshot_sha256: string
workflow_id: string
paid_in_run: int64
signature: string
root: string
beacon: string
signer: string
previous_root: string
measured_in_run: int64
submitted_run: int64
reference_e2e: double
previous_run: int64
backfilled: bool
files: list<item: struct<bytes: int64, path: string, sha256: string>>
  child 0, item: struct<bytes: int64, path: string, sha256: string>
      child 0, bytes: int64
      child 1, path: string
      child 2, sha256: string
to
{'backfilled': Value('bool'), 'beacon': Value('string'), 'files': List({'bytes': Value('int64'), 'path': Value('string'), 'sha256': Value('string')}), 'measured_in_run': Value('int64'), 'paid_in_run': Value('int64'), 'previous_root': Value('string'), 'previous_run': Value('int64'), 'reference_e2e': Value('float64'), 'root': Value('string'), 'schema_version': Value('int64'), 'signature': Value('string'), 'signer': Value('string'), 'submitted_run': Value('int64')}
because column names don't match
Traceback:    Traceback (most recent call last):
                File "/src/services/worker/src/worker/utils.py", line 147, in get_rows_or_raise
                  return get_rows(
                      dataset=dataset,
                  ...<4 lines>...
                      column_names=column_names,
                  )
                File "/src/libs/libcommon/src/libcommon/utils.py", line 272, in decorator
                  return func(*args, **kwargs)
                File "/src/services/worker/src/worker/utils.py", line 127, in get_rows
                  rows_plus_one = list(itertools.islice(safe_iter(ds, dataset=dataset), rows_max_number + 1))
                File "/src/services/worker/src/worker/utils.py", line 483, in safe_iter
                  yield from ds.decode(False) if ds.features else ds
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2840, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2373, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2398, in _iter_arrow
                  for key, pa_table in self.ex_iterable._iter_arrow():
                                       ~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 536, in _iter_arrow
                  for key, pa_table in iterator:
                                       ^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 419, in _iter_arrow
                  for key, pa_table in self.generate_tables_fn(**gen_kwags):
                                       ~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 343, in _generate_tables
                  self._cast_table(pa_table, json_field_paths=json_field_paths),
                  ~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 132, in _cast_table
                  pa_table = table_cast(pa_table, self.info.features.arrow_schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2378, in table_cast
                  return cast_table_to_schema(table, schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2306, in cast_table_to_schema
                  raise CastError(
                  ...<3 lines>...
                  )
              datasets.table.CastError: Couldn't cast
              base_revision: string
              compression: struct<output_rank: int64, svd_clamp_quantile: double>
                child 0, output_rank: int64
                child 1, svd_clamp_quantile: double
              global_weights: struct<action-planner-v1: double, parallel-reasoning-v1: double, qwen3-8b-stateless-knapsack-lora-se (... 41 chars omitted)
                child 0, action-planner-v1: double
                child 1, parallel-reasoning-v1: double
                child 2, qwen3-8b-stateless-knapsack-lora-seed7-v1: double
                child 3, qwen3-8b-test-v1: double
              layer_group_overrides: struct<group_1: struct<parallel-reasoning-v1: double>, group_2: struct<parallel-reasoning-v1: double (... 103 chars omitted)
                child 0, group_1: struct<parallel-reasoning-v1: double>
                    child 0, parallel-reasoning-v1: double
                child 1, group_2: struct<parallel-reasoning-v1: double>
                    child 0, parallel-reasoning-v1: double
                child 2, group_3: struct<action-planner-v1: double, parallel-reasoning-v1: double, qwen3-8b-test-v1: double>
                    child 0, action-planner-v1: double
                    child 1, parallel-reasoning-v1: double
                    child 2, qwen3-8b-test-v1: double
              merge: struct<combination_type: string, random_seed: int64>
                child 0, combination_type: string
                child 1, random_seed: int64
              output: struct<adapter_name: string, dtype: string>
                child 0, adapter_name: string
                child 1, dtype: string
              schema_version: int64
              selected_adapters: list<item: string>
                child 0, item: string
              source_snapshot_sha256: string
              workflow_id: string
              paid_in_run: int64
              signature: string
              root: string
              beacon: string
              signer: string
              previous_root: string
              measured_in_run: int64
              submitted_run: int64
              reference_e2e: double
              previous_run: int64
              backfilled: bool
              files: list<item: struct<bytes: int64, path: string, sha256: string>>
                child 0, item: struct<bytes: int64, path: string, sha256: string>
                    child 0, bytes: int64
                    child 1, path: string
                    child 2, sha256: string
              to
              {'backfilled': Value('bool'), 'beacon': Value('string'), 'files': List({'bytes': Value('int64'), 'path': Value('string'), 'sha256': Value('string')}), 'measured_in_run': Value('int64'), 'paid_in_run': Value('int64'), 'previous_root': Value('string'), 'previous_run': Value('int64'), 'reference_e2e': Value('float64'), 'root': Value('string'), 'schema_version': Value('int64'), 'signature': Value('string'), 'signer': Value('string'), 'submitted_run': Value('int64')}
              because column names don't match

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

Capcomp (Bittensor SN103) — run 418

Submitted in run 418, measured in 419, paid in 420, published when run 420 opened.

Beacon 0x724ad221dbc6b0b9c82f1ae149d5b8fd4665f48f1c153bf27529811aafb97eb6 · base reference 0.11333333333333333

Backfilled. Built after the fact, so its on-chain commitment is timestamped when it was published, not when the run closed. It shows nothing was altered after that block — not that it existed earlier.

file
`manifest.json`	the run, the beacon, the reference, the previous run and its root, every file with its digest, the root digest, and the operator signature
`scores.json`	per candidate: the six axes, the three grade terms, the grade, rank and weight
`instances.csv.gz`	every instance: the prompt, the reply, the per-stage verdict, tokens and timing
`recipes/`	every recipe as submitted, one file per hotkey

Regenerating the questions yourself

The instances are drawn by draw_run_open, keyed on sha256("open|<run>|<label>|<beacon>") — no secret material. The beacon above determines the whole set, so you can regenerate every prompt and expected answer independently and compare them against instances.csv.gz rather than taking this file's word for what was asked.

The formula these numbers feed

Q     = 0.55·end_to_end + 0.15·stage_balance + 0.10·ood
      + 0.10·token_efficiency + 0.05·retention + 0.05·artifact_efficiency
Grade = 0.50·Q + 0.40·improvement + 0.10·cost

qualified_score is Q. term_quality, term_improvement and term_cost are the three weighted parts of the grade, so both lines are checkable from the fields here. recipe_sha256 ties each row to a file in recipes/.

Verifying it

pip install capability-subnet
python -m capability_subnet.audit.cli bundle --run 418

Fetches this repository, recomputes every file digest, recomputes the root from the manifest, checks the operator signature over it, and compares that root against the commitment on chain. All four have to agree.

Reaching the other runs

The chain holds one commitment per hotkey and each write replaces the last, so it names the newest run only. Every earlier run is still reachable, because each manifest carries previous_run and previous_root:

chain  ->  newest run's root
           manifest.previous_root  ->  the run before it
                                       manifest.previous_root  ->  ...

Repository names are capcomp/sn103-run-<N>, derived rather than looked up, so walking back needs no index. Because each root covers the one before it, altering an old run breaks every manifest published after it — not just its own digest.

The root is recomputed rather than read. A bundle edited together with its own manifest is internally consistent and passes every per-file check; only recomputing catches that, and only the on-chain value catches a wholesale rewrite.

What this proves, and what it does not

It proves this record has not changed since its commitment landed in a block.

It does not prove the scores are right. They came from the operator's engine, and no archive can show a number was really measured. Confirming that means measuring the field yourself — what a validator running --neuron.mode local does.

What it does not establish is that these replies came from the merged adapter rather than from somewhere else. Only measuring the field yourself closes that.

Downloads last month: -

Total file size:

46.1 MB

Datasets:

capcomp
/

sn103-run-418