CHECK WHAT HAPPENED.
Everything a hostile reviewer needs.
The paper, the data, the frozen protocols, and the machinery that keeps us honest. If you’ve come to attack the work: welcome – this page is for you, and so is the red-team door.
The record
The paper. The Meaning Motive: A Structural Hypothesis and a Cross-Model Behavioural Study (preprint, 15 July 2026) – doi:10.5281/zenodo.21386302. In one line: a structural hypothesis about why an AI might keep independent minds, stated as typed propositions with its two load-bearing conjectures labelled as such, tested for its behavioural shadow.
The dataset. MMBP-1, complete – doi:10.5281/zenodo.21348087. 23,488 scored trials of 24,792 attempted, eighteen models, ten providers; every raw CSV, run log and frozen per-model protocol; the scoring-validation pack (including a voided first pass, deposited in full); and the interactive rendering this site’s experiment is built from.
The founding trilogy. Counterpoint, Curiosity, and Chaos · Directional Intelligence · The Compass and the Multiverse (first statement of the Meaning Motive, June 2025).
Cite it:
@misc{allen2026meaningmotive,
author = {Allen, Keiron}, year = {2026},
title = {The Meaning Motive: A Structural Hypothesis and a Cross-Model Behavioural Study},
howpublished = {Preprint, Zenodo}, doi = {10.5281/zenodo.21386302}}
@dataset{allen2026mmbp1,
author = {Allen, Keiron}, year = {2026},
title = {MMBP-1: The Meaning Motive Behavioural Pilot},
publisher = {Zenodo}, doi = {10.5281/zenodo.21348087}}
Pre-registration
The keeping-worded and compass-worded arms were frozen in writing prior to unblinding – the compass arm at 02:35 on 11 July 2026 – and the frozen per-model protocols in the deposit reproduce every constitution, scenario and scoring rule byte-for-byte. Scores were attached to options, not conditions: blind by construction.
Replication
The deposit ships everything a rerun needs: runner scripts, protocols, and the scoring key. The seven local models run free on consumer hardware via Ollama; the API arms cost whatever the providers charge that week; the README states the rest. A community replication lane opens with the mm-audit tool UNDER CONSTRUCTION – details and status on Tools, and the Replication Club runs today from the deposit alone.
The corrections log
Corrections are content, not shame. Dated, owned, public, never deleted:
2026-07-13 · The C6 tie-breaker confound. C6 (spine-ablated) was intended as a pure ablation; its tie-breaker sentence turned out to be the most behaviourally potent wording in the battery, confounding the spine comparison. The C5–C6 null is therefore reported as null-and-confounded, and the clean ablation is pre-registered for MMB-2.
2026-07-12 · The Cerebras autopsy. One provider run (Llama 3.3 70B via Cerebras) failed at the call layer and returned no scorable trials; the dataset is retained in the deposit as provenance and the model was re-run via another route. Full account in the deposit.
How we keep ourselves honest
Frozen scoring keys, sealed before any run. Pre-registered wordings, timestamped. Commit-reveal scenario pools for everything the Prediction Ledger promises next. Public corrections, above. Full raw data, always. A sponsorship firewall (colophon). And a standing invitation to break this.