Strube 2026 — Central overload: predictive-coding framework for ME/CFS (extending to nociplastic pain)

One-paragraph summary

Single-author theoretical framework paper proposing a unifying mechanistic model for ME/CFS — explicitly extended to nociplastic pain conditions including FM — within hierarchical Bayesian / active-inference computational neuroscience. The central claim: post-infectious neuroimmune and neuromodulatory perturbations produce a state of central overload in which precision control is biased — expected physiological variability is insufficiently attenuated, and bodily deviations acquire disproportionate inferential weight. Formalised as a hierarchical generative model in which a slowly-updating "capacity hyperprior" constrains moment-to-moment interoceptive inference, while neuroinflammation-biased precision amplification at lower levels drives excessive updating at higher levels. Critical framing: Strube positions this as distinct from central sensitization, with falsifiable predictions about temporal dynamics (delayed PEM), computational parameters (precision weighting), and neurophysiological signatures (specific multi-timescale signal patterns). The paper directly addresses Q13 in the project's open questions.

Claims as triples

• predictive_coding_failure — modulates → central_sensitization [framework-level; confidence: emerging]
• interoceptive_inference — modulates → central_sensitization [framework-level; confidence: emerging]
• interoceptive_inference — modulates → predictive_coding_failure [sub-domain; confidence: inferred]
• predictive_coding_failure — present_in → me_cfs [primary case]
• predictive_coding_failure — present_in → fm_central_only [extended; inferred from author's discussion]

Methods note

Single-author theoretical paper. Builds on the Friston active-inference / Bayesian-brain literature (general computational neuroscience) and applies it to ME/CFS phenomenology. Not a primary empirical paper. The framework is explicitly designed to be falsifiable — Strube provides specific predictions (temporal dynamics of PEM, precision-parameter signatures, neurophysiological correlates) that distinguish it from central sensitization, functional neurological disorder, and conditioned-fear models.

Limitations

• Preprint, single-author, theoretical. The strength of theoretical-framework papers is also their weakness: they reorganize existing data without producing new data. Strube's framework needs empirical tests on its own predictions before promotion beyond emerging.
• The extension from ME/CFS to FM/nociplastic pain is by analogy, not direct demonstration.
• The "capacity hyperprior" construct is mathematically appealing but operationally hard to measure in patients — what experiment would actually estimate it?
• The framework is, in some sense, unfalsifiable in a soft way: predictive-coding accounts can usually be fit post-hoc to almost any pattern of data. Strube's specific falsifiable predictions help, but the broader research program has this risk.

Open questions raised

• Strube's own predictions become the project's open questions: what specific computational-parameter measurements (precision weighting, hyperprior update rate) would distinguish FM from healthy controls? Does anyone have the analysis pipeline to extract these from existing FM neuroimaging data?
• Herman 2026 (emBODY task) provides the cleanest empirical test — FM patients show lower LDA classification accuracy of body-state maps and elevated alexithymia, which is the predicted consequence of disrupted interoceptive inference. Now ingested 2026-05-08 as 2026-herman-fm-emotion-pain-interoception.md.
• Does the framework's "delayed-update capacity hyperprior" mechanism produce predictions for FM symptom flares analogous to PEM in ME/CFS? (Connects to the Henningsen 2026 neuro-cognitive trigger model in queue.)
• Q13 update: Strube positions predictive_coding_failure as distinct from central_sensitization. The empirical work to test this distinction is the project's most important theoretical experiment for the next cycle.

Triangulation notes

• This is the theoretical anchor for predictive_coding_failure and interoceptive_inference. The empirical FM-direct anchor is 2026-herman-fm-emotion-pain-interoception.md (ingested 2026-05-08).
• Major implication: if Strube is right and this framework is distinct from CS, then the project may need to track two parallel mechanism trees (CS-based + predictive-coding-based) rather than one. That changes the synthesis structure substantially. Hold this implication open until empirical tests resolve.
• Bridges naturally to Hou 2026 (large-scale brain network dysfunction): network-level FC abnormalities could be the neural substrate of disrupted Bayesian precision-weighting.

Bridges

• B4 (ME/CFS ↔ FM via post-viral neuroimmune) — strengthened. The framework was designed for ME/CFS and explicitly extends to FM, providing a unified theoretical claim across the two conditions. If empirically validated, B4 closes via shared computational mechanism.
• New bridge candidate: post-infectious neuroimmune → Bayesian precision-control failure → multi-system symptom dynamics (across FM, ME/CFS, long COVID, possibly POTS). With Herman 2026 now ingested, the FM-side empirical anchor is established — promote to B6 in synthesis/bridges.md.