Two Candidate Readouts of a Proposed Common Race: Effort-Value Attribution and Commit-Position as Substrate Signatures of Race-Architecture
Paper 6BC · Pødenphant Lund, T. (2026r) · Preprint · Live on Zenodo
A programmatic proposal, not a completed theoretical contribution. A race-architecture under bounded resources leaves a substrate state during and after each race-resolution. This paper sketches two candidate substrate signatures that can, in principle, be measured from logprobs alone on language-model substrates, and specifies the measurement programme each would require to become a falsifiable result.
| DOI (concept) | 10.5281/zenodo.20339431 |
| Status | Preprint live as v1, 2026-05-27 |
| Cite-letter | 2026r |
| Author | Tomas Pødenphant Lund [ORCID] |
TL;DR
A race-architecture under bounded resources (parallel evaluation of candidates, accumulation under resource limits, irreversible commit) leaves a substrate state during and after each race-resolution. This paper proposes that two distinct readouts of that substrate state may be measurable from logprobs alone, and lays out the empirical programme each commits to.
Readout 1: Post-encoding trace-dominance (in comparative-evaluation races). Friction invested during encoding is proposed to lay down a deeper hysteresis trace; that trace then carries a larger signal-share in subsequent comparative-evaluation races. Six classical biases in the value-attribution literature (the IKEA effect, the endowment effect, the sunk-cost fallacy, the generation effect, effort justification, and the effort heuristic) are argued to share this race-mechanic component, but the claim is carefully bounded: the race-mechanic is proposed as one component of an effort-essential subset of these effects, not as a single-mechanism reduction of all six. Endowment under passive-ownership manipulations and sunk-cost under commitment-without-effort have well-documented occurrences in conditions where the friction-trace cannot be the operative mechanism. Those occurrences are not explained away.
Readout 2: Commit-position (in the response trajectory). A race-architecture must commit at some point in its output. Where in the trajectory it commits, how that position drifts across conditions, and how its variance is shaped by training are themselves measurable substrate properties. A preliminary computational probe on Qwen2.5-7B-Base versus its instruction-tuned counterpart, fine-tuned on identical content, shows three direction-consistent patterns: the base model exhibits a 3.4× wider cross-condition spread in commit proportion than the instruct model; drifts away from the secretary-problem optimum 1/e ≈ 0.368 as task-interpretation deepens; and shows a meta-race coupling between recognition and commit markers (Pearson r = 0.528) that the instruct model lacks (r = 0.104).
Critical caveat the paper insists on: the single-cell numerical match with 1/e is recorded as a coincidence to be replicated, not as a finding. The base-vs-instruct spread asymmetry is the directional pattern; the 1/e match itself awaits multi-cell replication before it can be claimed as a substrate-property.
Three-grade taxonomy of claims:
- Grade (a) — re-expression: established phenomena re-expressed in the framework's vocabulary, parsimony without novel empirical content. Applied to the IKEA effect under restricted scope, the generation effect, the effort heuristic.
- Grade (b) — candidate derivation with surplus content: a quantitative or qualitative prediction the native literature does not state. Applied to the predicted inverted-U over effort intensity (vanishing at trivial difficulty, peak at productive friction, dissipation at overwhelming difficulty — Norton et al. 2012 documents the dissipation boundary without explaining it); and to the commit-position spread × condition signature distinguishing base from instruction-tuned substrates.
- Grade (c) — speculative extension: framework applied where the empirical case is not established. Applied to the loss-aversion-as-soft-irrevocability-reversal-cost interpretation (offered as a speculative coda in §5.3, with no new evidence); and to the "three asymmetries" structural-homology speculation.
What this paper is: a programmatic statement of two candidate substrate signatures, a stratified accounting of their empirical readiness, an explicit list of the falsifiers each commits to, and a self-contained engagement with the accumulator-models literature (DDM, LCA; Appendix B) so the proposal can be evaluated without consulting the companion papers.
What this paper is not: a confirmed unification of effort-value biases, a validated quantitative match to 1/e, a derivation of loss aversion from first principles, or a replacement for any of the native-vocabulary treatments. Readers should expect a research direction worth pursuing, not a result.
Readout 1 in detail: post-encoding trace-dominance
A race-architecture that resolves under bounded resources leaves a substrate trace whose depth depends on the friction invested in the resolution. The proposal: in a subsequent comparative-evaluation race between two items, the item carrying the deeper friction-trace contributes a larger signal-share to the comparator's accumulation, producing systematic asymmetry in the comparative outcome. The asymmetry is the operational signature; the friction-trace depth is the substrate variable.
This race-mechanic is proposed as one component of an effort-essential subset of six well-documented effort-value biases:
- The IKEA effect (Norton, Mochon & Ariely 2012) — assembled-yourself items valued higher than identical pre-assembled items. Effort-essential under productive-friction conditions; the dissipation boundary at overwhelming difficulty is predicted from flow-breakdown.
- The endowment effect (Kahneman, Knetsch & Thaler 1990) — owned items valued higher than non-owned. The race-mechanic component applies to ownership accompanied by effort/engagement; passive-ownership manipulations (mere-instructed-ownership paradigms; Reb & Connolly 2007) produce endowment without the friction-trace and are therefore not explained by this account.
- The sunk-cost fallacy (Arkes & Blumer 1985) — prior investment biases future decisions. Race-mechanic applies under commitment-with-effort; commitment-without-effort variants (Cunha & Caldieraro 2009) show sunk-cost effects the friction-trace cannot mediate.
- The generation effect (Slamecka & Graf 1978) — self-generated answers retained better than passively-read ones. Effort-essential by construction; one of the cleanest grade-(a) applications.
- Effort justification (Aronson & Mills 1959) — outcomes following effortful processes valued higher. Race-mechanic applies; cognitive-dissonance accounts remain compatible at higher levels of description.
- The effort heuristic (Kruger et al. 2004) — perceived effort used as a value proxy in the absence of direct quality information. Effort-essential by construction; the substrate trace is what makes "perceived effort" a substrate-bound quantity rather than a free belief variable.
The paper's central grade-(b) prediction is an inverted-U over effort intensity: each effort-value effect should vanish at trivial difficulty (no friction-trace to lay down), peak at productive friction (maximal trace), and dissipate at overwhelming difficulty (flow-breakdown). Norton et al. 2012's IKEA-effect dissipation boundary is consistent with this without being predicted by the native effort-justification or signalling accounts. The prediction extends to all six biases under the effort-essential subset.
Readout 2 in detail: commit-position
A race-architecture under bounded resources must commit. The position in the response trajectory at which commit occurs, and how that position shifts across conditions, is a measurable substrate property, not a hidden internal state. The signature: spread of commit-position across conditions, drift away from the secretary-problem 1/e optimum, and the coupling structure between recognition markers and commit markers distinguish race-architectures with different substrate-discipline.
The preliminary computational probe (preregistered, single-cell): Qwen2.5-7B-Base versus the instruction-tuned counterpart, both fine-tuned on identical content, on a recognition-then-commit task structured to allow the substrate to commit at any of several positions in the trajectory.
- Cross-condition commit-proportion spread: base model 3.4× wider than the instruct model. The instruct model's narrower spread is interpreted as RLHF-imposed commit-discipline; the base model retains the unconstrained race topology.
- Drift away from 1/e ≈ 0.368 as task-interpretation deepens: the base model approaches the secretary-problem optimum at low interpretation-depth, drifts away as interpretation-depth rises. The single-cell numerical match with 1/e is recorded as a coincidence to be replicated; the directional drift is the testable pattern.
- Meta-race coupling between recognition markers and commit markers: base model Pearson r = 0.528; instruct model r = 0.104. The base model's coupling is interpreted as a recognition-modulated commit-trigger; the instruct's lack of coupling suggests the RLHF reward-signal has decoupled commit-timing from recognition-state in the trained-out substrate.
The three patterns are direction-consistent; each requires multi-cell replication before the substrate-property claim is independent of the specific test. The honest framing: this is a probe, not a result.
What the two readouts share and how they differ
Both readouts depend on the same race-architecture: bounded-resource parallel evaluation with irreversible commit (Friction Theory R1–R3, Paper 1). They measure the substrate at two different moments:
- Readout 1 (trace-dominance) measures the substrate after a race-resolution — the persistent hysteresis trace whose dominance is read out in a subsequent comparative-evaluation race.
- Readout 2 (commit-position) measures the substrate during a race-resolution — the position-locked commit-trigger whose timing reveals the race's accumulation dynamics.
The shared race-substrate is what justifies grouping them. They are not two empirical tests of one prediction; they are two different observational windows on the same race-architecture, asking different mechanistic questions of the same substrate.
Position relative to allied frameworks
The paper engages explicitly with the accumulator-models tradition (Appendix B is dedicated to this):
- Drift-Diffusion Models (Ratcliff 1978; Ratcliff & McKoon 2008) — the race-account is compatible with DDM at the single-race level and inherits its commit-position vocabulary. The race-architecture adds the substrate-trace persistence between races, which standard DDM does not natively represent.
- Leaky Competing Accumulator (Usher & McClelland 2001) — native LCA already accommodates the parallel-evaluation-under-resources picture; the race-architecture extends LCA to commit-position-as-substrate-property and to inter-race trace persistence.
- Race-against-time models (Mulder, Wagenmakers et al. 2014) — structural cousin; the substrate signatures we propose are intended as cross-cutting predictions that any race-against-time model with a substrate-persistence layer should also accommodate.
The paper's contribution is not a replacement for these models. It is a substrate-level vocabulary in which the two proposed readouts can be measured on language-model substrates from logprobs alone, with explicit cross-mapping to where DDM/LCA would predict the same patterns and where they would not.
Connections to other papers in the series
- Paper 1 (Friction Theory) — the substrate-universal framework whose race-axioms (R1–R3) the two readouts rely on.
- Paper 6 core (Matched Friction Under Hysteresis) — the schema paper. The two readouts in 6BC instantiate two specific signatures of Paper 6's measurement vocabulary. 6BC is the empirical sibling to 6 core's vocabulary statement.
- Paper 4B (Substrates Encode Experience) — encoding-through-loading. The trace-dominance readout depends on the same encoding mechanism that 4B characterises empirically; trace-dominance is what 4B's encoding leaves behind, read out in a subsequent race.
- Paper 13 (Operational Friction Theory) — specifies race-opening, recursive resolution, manifested behaviour, thermodynamic termination. 6BC's commit-position readout is the empirical handle on Paper 13's race-opening / commit-position structure.
- Paper 2B (ICL/FT memory) — the working-memory / long-term-memory distinction. 6BC's base-vs-instruct commit-proportion asymmetry is consistent with 2B's prediction that fine-tuning compresses the calibrated distribution; commit-discipline is one specific manifestation of that compression.
The Paper 6 family in current state
The original Paper 6 was restructured in May 2026 into a core paper plus companions. Current state:
- Paper 6 core — Matched Friction Under Hysteresis: A Programmatic Proposal for a Measurement Schema of Learning Optima. LIVE on Zenodo (concept DOI
10.5281/zenodo.20059863). The schema paper. - Paper 6BC — this paper. Consolidation of the originally-planned 6B (effort-value attribution) and 6C (commit-position) into one paper at Tomas's direction (2026-05-26). LIVE on Zenodo as of 2026-05-27.
- Paper 6D — the originally-planned mirror-friction paper was dropped as a standalone after the §3-substantiation pass showed the material was better placed as an expansion in Paper 4B §7.4d. The receiver-side mirror-friction content is integrated into Paper 4B v2 instead.
Read the paper
The full paper is on Zenodo (concept DOI 10.5281/zenodo.20339431):