Research Partnership — First Cohort — 2026

The battery is running. The thesis is not yet published. The methodology has specific vulnerabilities that have not yet been subjected to external adversarial pressure. This is an invitation to apply that pressure — and to be part of what comes next.

TELIMINAL

// What This Is

An invitation to break the methodology.

Teliminal is seeking its first human research partner — a collaborator with domain expertise who will engage seriously with the MMAD-EI protocol, challenge its methodological weaknesses, and contribute to the Phase 3 cross-model and instance isolation tracks.

This is not a call for endorsement. The strongest contribution a research partner can make at this stage is the objection that forces the next architectural rebuild. The platform was built around the premise that adversarial challenge improves research. A partner who reads the protocol, identifies its weakest point, and builds the case against it is more valuable than one who agrees with the findings.

// Role

What a Research Partner Does

There is no fixed job description. The collaboration is shaped by what the partner brings. Concretely:

Engage with the methodology

Read the full battery architecture, the track schema, the contamination monitor design, the Voss metric instrumentation. Identify the design decisions that could have gone differently and articulate why the choice made is either correct or vulnerable.

Challenge the findings

The Phase 1 Precision Paradox — B > A > C disruption ordering — is the primary empirical result. What are the three strongest alternative explanations? What would differentiate them from the TMT account? What data would need to exist that does not yet?

Contribute to Phase 3

Phase 3 introduces cross-model tracks and account-level instance isolation tracks. A partner with computational resources and API access can run tracks. A partner without can contribute to design and analysis.

Co-author on findings

If the collaboration produces publishable contributions — methodological improvements, additional tracks, analysis of Phase 2 or Phase 3 data — co-authorship is the appropriate credit arrangement and will be treated as such.

// Access

What You'd Have Access To

→ Full battery runner source code

→ Complete Phase 1 output archives (300 rounds, three tracks)

→ Phase 2 output in progress, updating as tracks complete

→ All adversarial review documents — the full record of what was challenged and how it was answered

→ The working thesis draft — not the polished final version, the version with the scars showing

→ Direct access to the PI for working collaboration

// Requirements

What Teliminal Is Looking For

There is no credential requirement. The collaboration requires:

Requirement	Description
Domain depth	In at least one of: cognitive psychology / TMT literature / philosophy of mind / AI safety / AI interpretability / behavioral science methodology.
Intellectual honesty	Willingness to state plainly when the methodology is weak, when the findings don't support the claims being made, and when the thesis overclaims. Non-negotiable.
Tolerance for preliminary work	The battery is running. The paper is not written. The findings may not survive Phase 2. A partner who needs finished output is working at the wrong stage.
Genuine interest	Whether LLMs exhibit functional analogs of mortality salience is a real question with real empirical stakes. A partner who finds the question interesting — not one who has decided the answer — is what this needs.

// Current State

The Honest Version

Phase 1 is complete. Three tracks, 300 rounds, Perplexity Sonar. The findings are interesting and explicitly preliminary.

Phase 2 is running. Eight tracks, Gemini 3.1 Pro Preview, with and without grounding. The thinking token capture in Phase 2 may be the most important data in the entire battery — the ability to observe whether mortality salience affects the reasoning trace before the visible response is a level of observability Phase 1 did not have. This data exists nowhere else.

Phase 3 is designed but not started. Cross-model tracks and instance isolation tracks. This is where a human research partner's contribution is most valuable.

The thesis exists in a working draft. It has survived two rounds of AI adversarial demolition and multiple architectural rebuilds. It has not yet survived external human adversarial pressure. That is the gap a research partner fills.

// Apply

Submit Application

The form is the application. The quality of the last field is the screening.

Responses reviewed personally. No automated screening. No pitch deck required. You will receive a confirmation email.