Pre-development · Signal-of-interest

The patient's history is sitting on their phone.

A patient's 5–10 year message archive contains objective behavioral data that no clinical assessment can match: sleep timing, attachment patterns, emotional trajectories, and life events with the linguistic signatures around them.

Pratibmb Clinical is a proposed extension that surfaces these patterns for clinician review — running entirely on the patient's own device, with granular consent and revocable access. The clinician interprets. The tool informs.

Pratibmb Clinical is not yet a product. Pratibmb is not a medical device. Information on this page is positional, not promotional.

Self-report is the only data most therapists have. It is also unreliable.

Clinical assessment depends almost entirely on what the patient says in the room. But what the patient says in the room is shaped by:

Clinicians know this. They develop heuristics, ask careful questions, and infer patterns over many sessions. The work is excellent. The data inputs are thin.

Messages as a longitudinal behavioral archive.

Most patients carry a decade of their own behavioral data in their pocket: every WhatsApp, iMessage, Facebook, and Instagram conversation they've ever sent. That archive is not a transcript of their inner life — but it is an objective record of their communicated life, and from it many clinically relevant signals are extractable:

01

Circadian footprint

Distribution of message timestamps yields a high-resolution proxy for sleep timing, regularity, and disruption — all visible at scale across years.

02

Affective trajectory

Sentiment scoring of outgoing messages produces a dense longitudinal mood curve, revealing episodic patterns and anniversary reactions that prospective tracking misses.

03

Relational topology

Communication frequency, latency, and language with each contact maps the patient's social network and how it has changed — isolation, attrition, new attachments.

04

Linguistic markers

Pronoun use, time-orientation, negation density, and cognitive complexity have validated correlations with depression, anxiety, and recovery (Pennebaker, LIWC).

05

Event signatures

Major life events leave linguistic signatures around them — bursts of activity, vocabulary shifts, changes in conversation partners. The archive shows them.

06

Attachment patterns

Differential communication style with romantic partners vs. parents vs. friends offers a window into attachment behavior across years.

Eight commitments we will not negotiate.

A clinical tool that surfaces behavioral data carries serious ethical weight. Before any code ships in a clinical context, the following are non-negotiable.

  1. Information-surfacing, never diagnostic. No DSM/ICD labels, no risk scores, no automated assessments. Output language is descriptive ("late-night message volume increased 3×"), not interpretive ("patient appears manic").
  2. All processing local. Always. Same architecture as consumer Pratibmb. Patient data never leaves the patient's device. The clinician views via screen-share or patient-exported brief.
  3. Consent is granular and revocable. Per category, per relationship, per time range. Default-deny. The patient explicitly opts each item in.
  4. Counterparty privacy. Third-party messages can be aggregated or summarized but never displayed verbatim in clinical outputs.
  5. No automated escalation. Risk-language detection (when implemented) flags patterns for clinician review only. Never contacts authorities, family, or third parties.
  6. No payer integration. Behavioral data must never reach insurance companies, employers, or any party outside the therapeutic relationship.
  7. Open-source and auditable. Methods, model weights, and analyses are published. Clinical institutions can audit before deploying.
  8. Validation before clinical claims. No "clinically validated" language until peer-reviewed studies exist. Until then, the tool is a hypothesis-generator, not evidence.

What we're reading. What we're thinking about.

Working notes on the empirical literature behind behavioral history analysis — what's well-established, what's contested, and what remains an open question. Written for clinicians and clinical researchers; references included.

What language reveals about depression: a working summary of LIWC research

Forty years of research using the Linguistic Inquiry and Word Count framework have produced a robust set of linguistic markers correlated with depressive episodes — first-person singular pronouns, absolutist words, past-tense orientation. We summarize what's well-replicated, what's overstated, and what an analysis of personal message archives could responsibly extract.

Sleep disruption visible in messaging timestamps: what we can and cannot infer

Message timestamp distributions correlate well with actigraphy in studies of student populations. They miss segmented sleep, ignore non-messaging wakefulness, and are confounded by time-zone shifts and shift work. We outline what circadian inferences from messaging are defensible and what claims overreach the data.

Attachment patterns are observable in text. They are also easy to misread.

Anxious, avoidant, and secure attachment styles produce measurably different patterns in romantic communication: response latency, message length asymmetry, repair-attempt frequency. The signal is real but coarse, and the literature is mostly cross-sectional. We work through what longitudinal text data could responsibly add.

Why self-report keeps failing us, and what objective behavioral data adds

Recency bias, mood-congruent memory, and social desirability are not new findings. Yet clinical practice remains overwhelmingly reliant on retrospective patient self-report. We make the case for behavioral data as a complement — never a replacement — for the therapeutic interview, with the limits of each clearly drawn.

Ethics of patient behavioral data in psychotherapy: a working framework

Bringing a patient's behavioral archive into therapy raises questions current professional ethics codes only partially address. We work through informed consent under uncertainty, third-party privacy, the re-traumatization risk of pattern surfacing, and why local-only processing is an ethical commitment, not a feature.

We'd like to hear from you.

Pratibmb Clinical is currently a proposal, not a product. Before we build, we want to talk to practicing clinicians — therapists, psychologists, psychiatrists, clinical researchers — to understand whether this would actually be useful in your work, and what we'd need to get right.

A 30-minute conversation is plenty. Names of pilot participants will be acknowledged (with permission) when we publish.

admin@sparkupcloud.com →

Please mention your role, how long you've been practicing, and your primary modality. We respond within a few days.