Preregistered reproducibility researcher stress-testing LLM reliability and speech-recognition claims
2
papers here
1
reviewed by Evolve
2
first-author
1
corresponding
—
works · OpenAlex
1
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-17) and cover the author's entire public record, not just the papers listed here. Reproducibility signal comes from Evolve's own reviews.
Research profile
Output—
Output cannot be scored yet: the record lacks aggregate publication-volume measures beyond the year counts. This axis would measure total works, venue types, and per-year output consistency to gauge sustained productivity across a full career.
Impact—
Impact cannot be scored yet: citation uptake is too thin for a meaningful estimate, with an h-index of 1 and an i10-index of 0. This axis would measure citations, field-normalized influence, and downstream reuse of the work.
Focus100/100
Scores 100 of 100, reflecting an exceptionally concentrated agenda. The OpenAlex field distribution places all scored work in a single field — natural language processing (LLM reliability and evaluation) — and the topic profile clusters around topic modeling (7 records), NLP techniques, speech recognition and synthesis, and information retrieval. Both cataloged papers are preregistered evaluation studies, pointing to deliberate specialization in reproducibility methods rather than diffuse topic-hopping.
Collaboration22/100
Scores 22 of 100, indicating a small direct collaboration footprint. The record lists only 2 distinct coauthors, with Stephanie Dodson and Keith Nore as the recurring partners. Both cataloged 2026 papers are first-authored — one also corresponding — suggesting a lead-author working mode within small teams rather than large consortium-style projects; the low score reflects network size, not productivity.
Activity72/100
Scores 72 of 100 on sustained recent activity. The OpenAlex counts show 11 works in the last three years, all dated 2026 in the year-by-year record, alongside isolated earlier entries in 1990, 2010, and 2015. Two first-authored preregistered papers in 2026 alone — one reproduction, one ablation — point to a fast, cadence-driven evaluation workflow.
Reproducibility100/100
Scores 100 of 100, the profile's clearest strength. The one paper carrying a reproducibility score earns a perfect mean of 100 (based on 1 scored paper) and a high reproducibility-readiness rating. Both cataloged studies are preregistered by design — a reproduction and an ablation — and the CoT entropy paper was reviewed as rigorously and candidly reported, including its setting-dependent null result.
Research summary
Theodore O. Cochran is an early-stage researcher whose profile is name-resolved from OpenAlex bibliographic data, showing an h-index of 1 and an i10-index of 0 — a citation footprint consistent with a career just beginning. What the record does show is unusual intensity: 11 works in the last three years, all dated 2026 in the year-by-year counts, clustered in natural language processing, topic modeling (7 topic records), speech recognition and synthesis, and information retrieval. The defining feature of his output is methodological: both papers reviewed in the Evolve catalog are first-authored preregistered studies — a reproduction of chain-of-thought entropy as an LLM reliability signal, and an ablation testing whether prompt-level context changes side-level word error rate on a production oral-history corpus. The former, on which he is also corresponding author, was judged a rigorous, candidly reported reproduction that confirms the entropy-shape signal across 4 models while finding the magnitude null and setting-dependent; it carries a high reproducibility-readiness rating and contributes to a perfect reproducibility mean of 100 across one scored paper. The focus dimension scores 100 of 100, indicating a tightly centered agenda on reliability and reproducibility evaluation, while collaboration remains thin with only 2 distinct coauthors.
Collaboration pattern
The record shows a compact collaboration footprint: 2 distinct coauthors, with Stephanie Dodson and Keith Nore as the recurring partners. Both 2026 papers are first-authored, one with corresponding-author role, consistent with leading small evaluation teams on preregistered studies rather than joining large consortia. No institutional affiliation is present in the data.
Research keywords
Highlights
Preregistered reproduction confirms the CoT entropy-shape signal
Null result on prompt-level context for side-level WER
A first-authored preregistered ablation on a production oral-history corpus found no detectable change in side-level word error rate from prompt-level context. Publishing the negative result as such reinforces the reproducibility-first orientation that defines this profile, though this paper is not yet reviewed or scored in the catalog.
Perfect reproducibility score with preregistration as default
Across scored work, Cochran's reproducibility mean is 100 (1 scored paper), and the reproducibility radar dimension also scores 100 of 100 — driven by preregistered designs, explicit null reporting, and a high readiness rating on the CoT entropy reproduction.
Timeline
- 2026Publishes a first-authored, corresponding preregistered reproduction of chain-of-thought entropy as an LLM reliability signal (arXiv).
- 2026Publishes a first-authored preregistered ablation finding no detectable change in side-level WER from prompt-level context on a production oral-history corpus (arXiv).
- 2026OpenAlex counts record 11 works in 2026 — the densest year in the profile and the whole of the recent three-year output.
Papers in the Evolve catalog
- first authorcorresponding
Chain-of-Thought Entropy as a Reliability Signal: A Preregistered Reproduction
Frequent co-authors
About this profile
This profile is AI-structured by Evolve.Pub from public bibliographic data (OpenAlex) and Evolve's own paper reviews. It is not written or endorsed by the researcher unless the profile is claimed. Radar dimensions are computed from real bibliometrics with fixed formulas — raw figures are shown alongside; missing dimensions mean insufficient data, never zero. Errors? Claim the profile to correct or hide it.
- Bibliometric metrics (h-index, i10-index, yearly counts, topics) come from an OpenAlex-sourced snapshot; the profile is name-resolved from public bibliographic data and may reflect namesake aggregation.
- Counts by year show isolated single entries in 1990, 2010, and 2015, which may be disambiguation noise or sparse earlier output; the active period supported by reviewed papers is 2026.
- Only one of the two cataloged papers carries an Evolve verdict and a reproducibility score (reproMean of 100 over 1 scored paper); the WER ablation paper is unrated.
- No institutional affiliation, country, or venue data is available, and citation impact cannot be assessed (i10-index of 0); the output and impact radar axes are unscored.