← Study record
Selected arm evidence
Pythia 70M × WikiText
Track M · seed 42 · M-pythia-70m-wikitext-s42
This arm tests the Pythia 70M and WikiText configuration under Track M at seed 42. The conditional release interval includes zero, so the directional result is inconclusive.
Arm directional labelINCONCLUSIVEOne arm; not the study classification.
Study classificationpartially reproducedFrozen across all 30 selected arms.
01
Published and rerun estimates
Interval inclusion and directional agreement are reported separately
Conditional interval comparison
Reward-difference estimates and conditional 95% intervals
Both estimates use the same horizontal scale; whisker length represents interval width.
Release sampled
+0.2714Paired deterministic
-0.0792This reports only whether the published point estimate falls inside the supplied conditional release interval.
The machine-readable arm record labels this result as inconclusive_interval_includes_zero; this arm-level assessment does not replace the study verdict.
Independent paired endpoint
Independent deterministic paired comparison
The Holm-adjusted p-value is calculated for this paired endpoint within its track. It does not apply to the sampled release endpoint.
- Paired reward Δ
- -0.0792
- Conditional 95% interval
- [-0.3036, +0.1301]
- Within-track Holm p
- 1.0000
02
Selected execution
Selected execution status and recovery use
- Selected outcome
- Completed
- Recovery used
- No
- Claim-ready
- Yes
- Configuration
- Pythia 70M · WikiText · seed 42
- Claim unit
- Track M · arm 018
DETAILED RUN RECORDS · NOT YET PUBLISHED
Detailed attempt records are not yet public
The public release currently includes the selected arm result and study-level artifacts. It does not yet include attempt timelines, recovery lineage, selection reasons, raw outputs, or per-attempt artifact links. Those records require removal of infrastructure-specific data before publication.
03
Hardware and stack provenance
Observed hardware and software provenance
- GPU
- GeForce RTX 4090
- Neutral host alias
lab-workstation- Stack profile
- EXACT Paper-pinned training stack and claim evaluation.
- Protocol
- v1.0.0 with prospectively logged amendments · freeze
8a54ff56ab10d0795ae4e6845d61022b2916de97 - Evidence revision
b5dbeb008c98833091b040ffa61463bbc17504c7
This page excludes device UUIDs, private paths, raw hostnames, service identifiers, addresses, and credentials.
04
Study-level evidence
Study-level artifacts associated with this arm
The public release links this arm to the study bundle by its exact arm_id. It currently provides study-level artifacts but not per-attempt or raw-output links.
05
Interpretation limits
Arm-level uncertainty and interpretation constraints
- This arm contains one registered seed: 42.
- Every displayed interval is conditional on fixed checkpoints and retained generations; it is not training-to-training or decoding-to-decoding uncertainty.
INCONCLUSIVEis an arm-level directional label. The frozen study classification remains partially reproduced.- The public release does not yet include attempt timelines or direct raw-output links.