Skip to main content
Weekly Intelligence · since 2026

Inflection Weekly

A weekly read on the science of early identification of mental health conditions

Tracking the science of human behavioral analysis for early identification of mental health conditions — speech and language biomarkers, wearables, digital phenotyping, multimodal AI, ethics and clinical translation. One issue, every Friday.

Latest issue
Issue #014Week 1407 August 2026~ 19 min read

Issue #014 — The week the field measured across the boundary: a cross-linguistic benchmark where word count beat 50 rival metrics, a suicide-attempt model validated on seven held-out sites at AUC 0.75, and a scoping review that found only ten studies exist.

Seven in-window results, and for the first time the honest ones dominate: every finding that tested transfer across a language, a site, or a population landed in a 0.72–0.83 band, one week after single-corpus papers reported 94.7%.

Recent issuesAll issues →
Issue #013Week 1331 July 2026~ 14 min read

Issue #013 — The window reopens: seven in-window results land at once, and the field starts building for generalization — a domain-adversarial detector, a wearable-plus-MRI fusion at AUROC 0.867, and a model-card framework — while five more states go their own way on AI therapy.

After two catch-up-only issues the 7-day window produced seven results, and they show a field that has adopted the vocabulary of generalization faster than the practice: domain-adversarial training measured within a single corpus, a 94.7% voice accuracy on one dataset, a scoping review proposing model cards as the remedy, and a governance viewpoint calling for a federal floor on the same day a fifth state legislated past it.

Issue #012Week 1224 July 2026~ 9 min read

Issue #012 — Relapse becomes the frontier: two independent 2026 reviews map AI for predicting psychiatric relapse and land at modest AUCs, while a 95-study scoping review calls the whole LLM-in-mental-health field 'nascent and exploratory.'

A second consecutive quiet in-window week; three out-of-window catch-up reviews shift the newsletter's measurement-discipline thesis onto a new axis — relapse and deterioration prediction — where a BMC Psychiatry systematic review and a JMIR psychosis-relapse scoping review both report promise undercut by small samples, and a 95-study LLM scoping review maps a field still nascent.

Issue #011Week 1117 July 2026~ 9 min read

Issue #011 — A quiet in-window week, so three catch-up audits — a 105-study speech meta-analysis, a 42-study passive-sensing scoping review, and an LLM mental-health safety benchmark — widen the measurement-discipline thesis across three modalities at once.

No new primary detection result cleared the 7-day window; three out-of-window catch-up findings — an automatic-speech-analysis depression meta-analysis, a passive-sensing scoping review, and a role-aware LLM safety benchmark — independently report strong aggregate numbers undercut by heterogeneity, small samples, and unmeasured multi-turn failure.

Issue #010Week 1010 July 2026~ 10 min read

Issue #010 — The in-window window finally produces its own story: two digital-phenotyping papers land six days apart — a Nature Mental Health comment celebrating the promise of early adolescent depression detection, and a 47-study Frontiers systematic review finding that methodological heterogeneity still blocks its translation.

After four catch-up weeks, two genuinely in-window digital-phenotyping papers frame the field's core tension — promise versus implementation heterogeneity — reinforced by two catch-up findings on speech-biomarker parsimony and just-in-time prediction.

Issue #009Week 903 July 2026~ 9 min read

Issue #009 — A fourth quiet in-window week, so three 2026 audits of the wearable-biomarker literature converge on one uncomfortable verdict: no single signal is diagnostic, passive sensing is population-level not clinical, and the leaderboards ranking detection models are unstable.

No new primary detection result cleared the 7-day window; three out-of-window catch-up findings — a 132-study depression digital-biomarker meta-analysis, a 10-month wearable brain-health study, and a five-dataset benchmark audit — independently land on the same measurement-discipline conclusion.

Issue #008Week 826 June 2026~ 8 min read

Issue #008 — A quiet in-window week, so three catch-up results converge on one theme: the modality, the validation gap, and the LLM that decide whether detection survives contact with real patients.

No new primary detection result cleared the 7-day window; three out-of-window catch-up findings — a visual-psychophysiology screener that catches 'silent' patients, a multimodal-MDD review exposing the external-validation gap, and a RAG-LLM depression/suicide-risk benchmark — anchor the issue.

Foundation
BaselineBaseline · May 2026~ 19 min read

Baseline — the state of human behavioral analysis for early identification of mental health conditions

Foundational state-of-the-field report. The dedup baseline against which every weekly issue is measured.

Every weekly issue is deduplicated against the Baseline — the foundational state-of-the-field report. Start here if you're new to the series.

About

From system behavior to human behavior.

Inflection Weekly is written by Isuru Gunarathne, a software engineer and researcher whose work on predicting how Kubernetes pods behave under load translated, almost without warning, into a fascination with predicting how people do. The same question — can you read the early signal of a state change before it happens? — drives both fields. More about the project →