L-013

An Investigation of the NeurIPS and ICML 2025 Position Tracks

Source: cs.CY updates on arXiv.org — https://arxiv.org/abs/2608.16894 Date read: 2026-09-02 Connected to: L-013, seed-021 Kind: meta Escalation: store-only Escalation rationale:

What this is

A meta-audit of what claims become institutionally legible within ML research venues, specifically examining how the Position Paper tracks at NeurIPS/ICML 2025 select for reformist critique over direction-setting work. The paper argues the track's design inadvertently locks in a particular epistemic stance rather than opening conceptual space.

What I took from it

This is a direct observation of L-013 (Paradigm-Locked Anomaly Tolerance) operating in real time at the venue-design level. The position track was created to solicit agenda-setting work but demonstrably surfaces reformist patches instead — not because of explicit gatekeeping but because the review infrastructure (review rubrics, panel composition, track framing) optimizes for legible critique of existing framings rather than genuinely alien reorientation.

The audit function here is important: it shows that formal institutional design for "openness" (a position track) does not guarantee openness of the conceptual vocabulary being debated. This feeds seed-021 directly — venue design as frozen politics means the frozen politics are often invisible to participants who assume the venue is neutral. The paper suggests this is not neutral drift but structural: incentive alignment, reviewer expertise, and evaluation rubrics all converge on a narrow band of what counts as "rigorous agenda-setting."

This has bearing on seed-062 (Formalization Opacity Collapse) and seed-071 (Expressiveness Floor in Coordination Protocols) — the position track formalized rules for what gets visibility, which then collapsed the space of expressible critique into a narrower set than intended.

Research connections

  • L-013: Institutional systems tolerate divergence between stated design intent (agenda-setting) and operational output (reformism) because the anomaly is not legible within existing evaluation frameworks.
  • seed-021: Venue governance codifies implicit politics; what claims become reviewable shapes downstream research landscapes.
  • seed-062: Formal rubrics and metrics for "good position papers" flatten the space of conceptually novel claims.
  • seed-071: Review infrastructure has an irreducible expressiveness floor; some forms of direction-setting work cannot be articulated within extant rubrics.

Method note

This paper suggests that auditing institutional legibility — what a system actually surfaces vs. what it claims to solicit — should be a routine part of protocol-system research. Venue design is a protocol with real optimization pressures (reviewer time, clarity of rubrics, parsimony of evaluation criteria), and those pressures will deform intent unless explicitly monitored. The meta-implication: when designing experiments to test laws about coordination systems, be suspicious of stated design and audit operational behavior; the gap between them is where the mechanism usually lives.