Candidate record (the source of truth)

Materials · case/candidate/record.md

# Jordan Hale: record

This is Jordan's own account of what they did, kept for accuracy. Where the CV is shorter or more flattering, this record controls.

## Education

- PhD, Organizational Behavior, Ridgeview University, 2016. Dissertation on how the order in which teams hear dissenting views affects whether the dissent changes the decision.
- BA, Psychology, Carrow College, 2010.

## Positions

- **Independent researcher, 2024 to present.** No employer. Work is self-funded and is described under Projects.
- **Research Lead, Measurement, Kestrel Media (streaming service), 2022 to 2024.** Led a team of three researchers (direct reports). Rebuilt the viewer-satisfaction survey after finding that its main item tracked recent outages more than satisfaction with the catalog. Validated the new measure against later cancellation. Owned the measurement plan for experiments run by the product teams but did not run those experiments.
- **Senior Research Scientist, Parcel (consumer delivery app), 2018 to 2022.** Designed and analyzed product experiments on delivery-time estimates and tipping prompts. Built the company's first holdout process for long-running experiments. Mentored two analysts informally. No direct reports.
- **Postdoctoral Fellow, Ridgeview University Decision Lab, 2016 to 2018.** Ran laboratory experiments on group decisions with student participants. Managed the lab's participant pool.

## Projects

- **Relay (2024 to present).** A testbed Jordan built for studying how information changes as it passes through chains and networks of language-model agents. The wiring between agents, the model at each position, and what an agent may pass on are set in code. Every message is stored with its lineage, so any final message can be traced to its source. Jordan validated the language-model judges that code message changes against a set of 300 messages that Jordan and one research assistant coded by hand, and found that the judges agreed well on whether facts were dropped but poorly on whether qualifications were dropped. Jordan also caught a configuration error before the first full run: the model-access route was adding hidden context to every agent call, which would have contaminated every condition. Relay has been used only with language-model agents. It has not been used with people.
- **Pricing-experiment rehearsal (2025).** Jordan used language-model agents as simulated shoppers to rehearse the design of a tiered-pricing experiment before a client ran it. The rehearsal planted a known effect and checked whether the design, sample size, and analysis plan would detect it. All participants were simulated agents. The rehearsal says whether the design can detect a planted effect in simulation. It says nothing about how real shoppers respond to tiered pricing, and the client's own experiment had not reported results as of this record.
- **Satisfaction measure rebuild, Kestrel Media (2023).** See Positions.

## Papers

1. Hale, J., & Moreno, L. (2019). Hearing dissent late: Order effects in team decisions. *Organization Science*. Published. First author.
2. Ibarra, K., Chen, S., Hale, J., & Novak, P. (2021). Measuring psychological safety from meeting transcripts. *Journal of Applied Psychology*. Published. Jordan is third of four authors and contributed the measurement model and the robustness analyses. Ibarra led the study design and data collection.
3. Hale, J. (2026). Message drift in chains of language-model agents. Under review at *Management Science*, first round, submitted July 2026. Not accepted. Reports Relay experiments with language-model agents only.
4. Hale, J. (2025). Rehearsing experiments with simulated participants. Working paper posted on jordanhale.example. Not submitted.

## Talks

- **Invited:** "When can a language-model judge be trusted?" Measurement in AI Workshop, Varrow University, May 2025.
- **Invited:** "Late dissent" at the Academy of Management annual meeting, 2019. Paper session, competitively selected, not an invited keynote.
- **Not a talk:** "Building an agent testbed" (2025) is a recorded walkthrough Jordan produced and posted on jordanhale.example. No one invited it and it was not presented to an audience.

## Teaching

- Teaching assistant, Organizational Behavior (MBA core), Ridgeview University, 2013 to 2015.
- Guest lecture on survey measurement, Ridgeview University undergraduate methods course, 2021.

## Skills actually used in the work above

Python, R, SQL, experimental design, survey design and validation, multilevel models, inter-rater reliability, language-model evaluation, prompt and pipeline engineering for research use. No experience building reinforcement-learning environments or training models.