Six Bets
Each bet below is a claim where BMC and a named rival theory predict different outcomes, stated together with what would refute it. These are commitments, not results. None has been decided yet. One (P-BM28) was partly run in BMC’s own simulation, but its result is also predicted by a simpler account (Rogers 1988), so it does not favour BMC; the separating test is still open.
BMC’s simulation research is in maintenance mode, so no new simulation experiments are being started. BMC’s code is closed. The five other bets can be tested on human or public data without BMC’s code; P-BM28 needs an independent implementation or human data (see its card).
Formal results within the framework and retrodictions show consistency, not differential support, and are not counted as tests.
The bets
P-D2
A word speeds recognition of another word that is close in meaning, even when people never link the two directly in free association.
Rival: spreading activation over an associative network (Collins & Loftus 1975; Nelson et al. free-association norms): priming is carried by associative paths, including indirect ones; without an associative path there should be little or no priming.
Refuted if (pre-registered): the similarity effect is indistinguishable from zero (equivalence test, TOST, $|\beta| < 0.02$); or the associative-path effect is at least twice the similarity effect; or the PPMI “blind twin” reproduces at least 80% of the effect.
Status: open. Pre-registered on 2026-08-29; the decision rules and thresholds are published on this site. The data are public (Semantic Priming Project). Power is marginal: 0.80 at $|\beta| \approx 0.045$, while effects reported in the literature are 0.03–0.04. A published result leans against BMC: in an item-level analysis across 300 associated prime–target pairs, priming grew with associative strength but did not vary with LSA similarity (Hutchison, Balota, Cortese & Watson 2008, Q J Exp Psychol 61(7):1036–1066, doi:10.1080/17470210701438111). Open
Rival: spreading activation over an associative network (Collins & Loftus 1975; Nelson et al. free-association norms): priming is carried by associative paths, including indirect ones; without an associative path there should be little or no priming.
Refuted if (pre-registered): the similarity effect is indistinguishable from zero (equivalence test, TOST, $|\beta| < 0.02$); or the associative-path effect is at least twice the similarity effect; or the PPMI “blind twin” reproduces at least 80% of the effect.
Status: open. Pre-registered on 2026-08-29; the decision rules and thresholds are published on this site. The data are public (Semantic Priming Project). Power is marginal: 0.80 at $|\beta| \approx 0.045$, while effects reported in the literature are 0.03–0.04. A published result leans against BMC: in an item-level analysis across 300 associated prime–target pairs, priming grew with associative strength but did not vary with LSA similarity (Hutchison, Balota, Cortese & Watson 2008, Q J Exp Psychol 61(7):1036–1066, doi:10.1080/17470210701438111). Open
P-WM1
Fear (FEAR) takes up about one working-memory slot; play (PLAY) takes none, at matched arousal.
Rival: attentional-control theory, or “only arousal matters”: at equal arousal, FEAR and PLAY cost the same.
Refuted if: FEAR and PLAY, matched on arousal, reduce working-memory capacity equally, or FEAR $\Delta k < 0.5$. Thresholds: FEAR $\Delta k \geq 0.5$ slot; PLAY $|\Delta k| < 0.25$ (TOST). Inconclusive if the equivalence test lacks power. Here $k_{eff}$ means working-memory capacity (K-score), not effective connectivity $K_{eff}$.
Status: open. Open
Rival: attentional-control theory, or “only arousal matters”: at equal arousal, FEAR and PLAY cost the same.
Refuted if: FEAR and PLAY, matched on arousal, reduce working-memory capacity equally, or FEAR $\Delta k < 0.5$. Thresholds: FEAR $\Delta k \geq 0.5$ slot; PLAY $|\Delta k| < 0.25$ (TOST). Inconclusive if the equivalence test lacks power. Here $k_{eff}$ means working-memory capacity (K-score), not effective connectivity $K_{eff}$.
Status: open. Open
P-CB1
About 200 documented cognitive biases reduce to 6 generating mechanisms, and this shows in data.
Rival: biases as a set of independent heuristics, or independent errors.
Refuted if: a bias-to-mechanism assignment, published before factor extraction, explains the factor loadings no better than the 95th percentile of label-shuffled assignments.
Status: open. It needs a noise floor first: the test–retest reliability of the bias battery. Open
Rival: biases as a set of independent heuristics, or independent errors.
Refuted if: a bias-to-mechanism assignment, published before factor extraction, explains the factor loadings no better than the 95th percentile of label-shuffled assignments.
Status: open. It needs a noise floor first: the test–retest reliability of the bias battery. Open
P-CB2
Cognitive load strengthens biases that come from working memory and drives (W, G), but not automatization biases (A).
Rival: the default dual-process view: load strengthens all heuristic biases.
Refuted if: A-biases change under load as much as W- and G-biases, or W/G $d < 0.3$. Thresholds: W/G $d \geq 0.3$; A $|d| < 0.15$ (TOST). Inconclusive if the equivalence test lacks power.
Status: open; testable by meta-analysis of existing studies. It depends on the P-CB1 assignment, so it stands or falls with it. Open
Rival: the default dual-process view: load strengthens all heuristic biases.
Refuted if: A-biases change under load as much as W- and G-biases, or W/G $d < 0.3$. Thresholds: W/G $d \geq 0.3$; A $|d| < 0.15$ (TOST). Inconclusive if the equivalence test lacks power.
Status: open; testable by meta-analysis of existing studies. It depends on the P-CB1 assignment, so it stands or falls with it. Open
P-TH1
PTSD therapy via memory reconsolidation (reactivation, then intervention within ≤ 6 h) gives remission above 70% and beats standard exposure therapy. Research prediction, not treatment advice.
Rival: extinction / standard exposure: no advantage.
Refuted if: remission is 70% or lower, or there is no advantage over exposure. Remission here means loss of the PTSD diagnosis on a structured clinical interview (e.g. CAPS-5), intention-to-treat, at ≥ 3 months follow-up.
Status: open, high-risk. Two placebo-controlled propranolol trials disagree: Brunet et al. 2018, Am J Psychiatry 175(5):427–433 (doi:10.1176/appi.ajp.2017.17050481) found an effect; Roullet et al. 2021, Neuropsychopharmacology (doi:10.1038/s41386-021-00984-w) found no difference. The meta-analysis is heterogeneous (Astill Wright et al. 2021, Transl Psychiatry 11:453, doi:10.1038/s41398-021-01570-w): for PTSD symptom severity at 1–4 weeks across all reconsolidation interventions (11 trials), I² = 91%. There is no head-to-head trial against exposure, and a remission rate above 70% would be at the top of what trials of established PTSD therapies report. Open
Rival: extinction / standard exposure: no advantage.
Refuted if: remission is 70% or lower, or there is no advantage over exposure. Remission here means loss of the PTSD diagnosis on a structured clinical interview (e.g. CAPS-5), intention-to-treat, at ≥ 3 months follow-up.
Status: open, high-risk. Two placebo-controlled propranolol trials disagree: Brunet et al. 2018, Am J Psychiatry 175(5):427–433 (doi:10.1176/appi.ajp.2017.17050481) found an effect; Roullet et al. 2021, Neuropsychopharmacology (doi:10.1038/s41386-021-00984-w) found no difference. The meta-analysis is heterogeneous (Astill Wright et al. 2021, Transl Psychiatry 11:453, doi:10.1038/s41398-021-01570-w): for PTSD symptom severity at 1–4 weeks across all reconsolidation interventions (11 trials), I² = 91%. There is no head-to-head trial against exposure, and a remission rate above 70% would be at the top of what trials of established PTSD therapies report. Open
P-BM28
The drive to communicate is driven by pressure on memes to spread, not by a survival benefit.
Rival: the adaptationist view of language (Pinker & Bloom 1990) and reward-trained agents (REINFORCE / PPO): communication pays off in survival.
Refuted if: in a cooperative world with predators, receiving messages raises survival but the drive to express falls with it — $R_{expr}(\text{ReceiveOnly}) / R_{expr}(\text{Blocked})$ below 0.9, at power ≥ 0.8. BMC predicts the ratio stays ≥ 0.9 (CI excluding 0.5) while $\Delta_{alive} > 0$; drive-based accounts predict that reception discharges the need. A null on survival alone does not separate the two: Rogers (1988) predicts the same null.
Caveat: in BMC's own simulation this condition cannot fire. The $R_{expr}$ ordering is a consequence of the model specification, as the $R_{expr}$ paper states; the pilot ratio was 48.5 / 45.5 ≈ 1.07. A real test needs an independent implementation or human data.
Status: partly run, in BMC's own simulation. The run part — 10 survival experiments, including predator avoidance and cooperative foraging, with no measurable survival benefit (DOI: 10.5281/zenodo.19181798) — does not favour BMC: Rogers (1988) predicts the same survival null. A small cooperative pilot without predators (20 seeds; DOI: 10.5281/zenodo.19309824) shows only a non-significant difference in energy (a survival proxy): 80.5 for ReceiveOnly vs 75.0 for Blocked ($p = 0.10$, 14/20 seeds). The separating test has not been run. The simulation code is closed, so the run is not externally reproducible. Partly run
Rival: the adaptationist view of language (Pinker & Bloom 1990) and reward-trained agents (REINFORCE / PPO): communication pays off in survival.
Refuted if: in a cooperative world with predators, receiving messages raises survival but the drive to express falls with it — $R_{expr}(\text{ReceiveOnly}) / R_{expr}(\text{Blocked})$ below 0.9, at power ≥ 0.8. BMC predicts the ratio stays ≥ 0.9 (CI excluding 0.5) while $\Delta_{alive} > 0$; drive-based accounts predict that reception discharges the need. A null on survival alone does not separate the two: Rogers (1988) predicts the same null.
Caveat: in BMC's own simulation this condition cannot fire. The $R_{expr}$ ordering is a consequence of the model specification, as the $R_{expr}$ paper states; the pilot ratio was 48.5 / 45.5 ≈ 1.07. A real test needs an independent implementation or human data.
Status: partly run, in BMC's own simulation. The run part — 10 survival experiments, including predator avoidance and cooperative foraging, with no measurable survival benefit (DOI: 10.5281/zenodo.19181798) — does not favour BMC: Rogers (1988) predicts the same survival null. A small cooperative pilot without predators (20 seeds; DOI: 10.5281/zenodo.19309824) shows only a non-significant difference in energy (a survival proxy): 80.5 for ReceiveOnly vs 75.0 for Blocked ($p = 0.10$, 14/20 seeds). The separating test has not been run. The simulation code is closed, so the run is not externally reproducible. Partly run
Two further checks (P-NM3, P-BM4) were run inside BMC’s engine; they test implementation fidelity, not BMC against a rival, and are not evidence.
The P-D2 pre-registration’s decision rules, thresholds, blind twins and power analysis are published on a separate page: P-D2 pre-registration: thresholds.
What would falsify BMC?
Each of the following would refute a named BMC mechanism:
- Zero-tension consciousness. A system with no G–M tension (pure M, no drives) that nonetheless sustains a self-model.
- Valence-specific working-memory capture. See P-WM1.
- Bias factorisation against a permutation null. See P-CB1.
- Bidirectional-capacity slope. BMC states $N_{bid}(L) \approx \text{bandwidth}(L) - 1$ . Saturation at $N_{bid} \approx 3$ under experimentally expanded working memory would refute it, but only once the slope claim is committed as robust over a stated range of the threshold parameter $\theta_{bid}$ . Until then this is a direction, not a test.
- Factor independence of CL. If no clinical state exists with low $\sigma_{SW}$ (low PCI) and intact $A_{SMC}$ (normal DMN), the CL factors are not independent and the product form is overdetermined.
- Sign inversion with magnitude preserved (P-1). Under revision: survey data already depart from ≈ 1 without moderators (Beider 2021, J Sci Study Relig 60(1):5–26, doi:10.1111/jssr.12698).
Findings that were settled before BMC was written (for example, age-related rigidity in belief change, or a primacy advantage for early beliefs) are consistency checks, not falsifiers: they could not have gone the other way.
The earlier, broader catalogue is archived as a coverage atlas — a map of what BMC speaks about, not a list of tests.
For the full formal theory, see the five pillars. For a plain-language introduction, see the Guide.