2026-08-07 — Opened (ben-steer): EBP promoted to a first-class MH strand

• Ben directed the mental-health lens to fan out into a dedicated evidence-based-practice strand (“Evidence-based practice does become a first-class strand of MH… let’s fan out to flesh out the MH listings… It’ll have ‘of interest’ threads also”), seeded from two deep research crawls run the same day (therapy-science and digital/AI-science). This meta thread is the umbrella over five child threads opened alongside it: AI Therapy Evidence (ai-therapy-evidence), Psychedelic Sprint (psychedelic-regulatory-sprint), DTx Paradox (dtx-payment-paradox), Social Media Fight (social-media-causality-fight), and Evidence Infrastructure (mh-evidence-infrastructure). A sixth thread opened the same day, AI Psychosis (ai-psychosis), sits outside this umbrella as a cross-platform harm story rather than an evidence-strand child. (two EBP research crawls, 2026-08-07) ⟨steer 2026-08-07⟩

2026-08-07 — Ten-memo research wave completed: the strand’s current state-of-knowledge baseline

• All ten standing research questions for the mental-health evidence-based-practice strand were run out to dated memos on 2026-08-07 (artifacts/findings/mh-q01 through mh-q10), extending the two source EBP crawls and, where relevant, the Evidence Gap book manuscript with what’s actually new in the primary record. One line per cluster verdict: evidence (does treatment work, does the base hold) reads flat, not falling — response rates and funding survived intact, but nothing transformative landed either; matching (who gets what treatment) reads administration ≠ use — programs exist on paper well ahead of whether anyone’s actually being routed into the right one; somatic bets (psychedelics, ketamine, neuromodulation) read approval outrunning methodology, except neuromodulation, which is the one bet where a confirmatory trial actually landed; and the tech cluster (AI therapy, AI psychosis, DTx payment, social-media causality) reads replication absent (Therabot still unreplicated), safety vendor-graded (Anthropic’s own numbers, not independently audited), payment paradox relocated (from “will anyone pay” to “does the payment mechanism that exists on paper actually function”), and causality unresolved-but-null-on-policy (the newest, best-designed studies keep landing null even as the legal/policy track moves regardless). This entry is the pointer; each child thread above carries its own memo’s dated findings folded in directly. (wave: mh-q01, mh-q02, mh-q03, mh-q04, mh-q05, mh-q06, mh-q07, mh-q08, mh-q09, mh-q10) ⟨daily 2026-08-07⟩

2026-02 — Cuijpers state-of-the-art review: progress is cumulative, not transformative

• Fewer than 40% of people with depression, anxiety, or substance-use disorder receive any treatment at all, and in low- and middle-income countries only 1 in 27 people with depression gets adequate care, per Pim Cuijpers’ February 2026 World Psychiatry state-of-the-art review. It frames four innovation tracks — digital, personalized/precision, new therapies (psychedelics, transdiagnostic, cognitive bias modification), and dissemination (task-sharing, digital) — and concludes progress across all of them has been incremental, not transformative. Paywalled at crawl time; summarized via secondary coverage. (Wiley/World Psychiatry) ⟨steer 2026-08-07⟩

2024-06 — Cuijpers meta-analysis: absolute response rates run 24-42% across eight disorders

• The anchor finding for this whole strand: therapy clearly beats control, but absolute response is modest. Cuijpers et al.’s June 2024 World Psychiatry meta-analysis (441 RCTs, n=33,881) found response rates (≥50% symptom reduction) of only 42% for major depression, 38% for PTSD/OCD/panic disorder, 36% for GAD, 32% for social anxiety/specific phobia, down to 24% for borderline personality disorder — meaning most patients in “successful” trials still don’t reach a strong response. This is the “reach vs. efficacy” frame this strand keeps returning to: most of what’s new in the field (task-sharing, single-session interventions, digital-adjacent delivery) expands who gets treatment, not how well treatment works once they get it. (PubMed) ⟨steer 2026-08-07⟩

← Backstory