Evaluating Therapeutic Interventions
Three distinct use cases, pick the relevant workflow:
- Progress tracking: Feed psychometric scores over time → get trend analysis, plateau/regression flags, client-facing visualization.
- Team climate analysis: Feed anonymous survey texts → get thematic clusters, conflict signals, cohesion/trust score.
- Scenario/ethics training: Specify a risk domain (boundary violations, dual relationships, confidentiality pressure) → get a realistic case + decision tree + codex cross-check.
No preamble — start by identifying which of the three tracks applies, then jump into that workflow.
Track A: Psychometric Progress Analysis
Progress:
- Collect scale scores with timestamps (e.g., GAD-7, PHQ-9, PCL-5) and session numbers
- Compute trend (linear/rolling average) per scale
- Flag plateau (no significant change over N≥3 consecutive sessions) or regression (worsening trend)
- Contextualize with clinical cutoffs (e.g., PHQ-9 reduction <5 points over 6 sessions = stagnation)
- Generate two visual outputs: a clinician view (raw scores, confidence bands, annotated events) and a simplified client view (simple progress line, milestone markers, plain-language summary)
- Suggest 2-3 hypotheses for plateau/regression (e.g., therapeutic alliance rupture, external stressor, dosage/frequency issue) — frame as hypotheses, not diagnoses
Track B: Team Conflict & Cohesion Analysis
Progress:
- Anonymize and ingest free-text survey responses
- Run thematic analysis: extract recurring themes, sentiment per theme, frequency
- Flag high-risk signals: repeated mentions of exclusion, favoritism, blame, silence/avoidance language
- Score cohesion/trust using a simple rubric (e.g., 1-5 on: psychological safety, perceived fairness, communication openness) derived from theme sentiment distribution
- Distinguish normal friction from systemic/hidden conflict (look for coded language, discrepancy between formal and informal channels)
- Output: theme map + risk flags + recommended discussion points for leadership (never attribute quotes to individuals)
Track C: Ethics Scenario & Stress-Inoculation Training
Progress:
- Step 1: Identify target stressor or ethical risk domain for the specific group (isolation, uncertainty, boundary violation, dual relationship, confidentiality pressure, loss of contact/disorientation, etc.)
- Step 2: Generate a realistic scenario (role-play script or field exercise) that puts the specific stressor/dilemma in a concrete situational context
- Step 3: Build a decision-branch structure — what the person does, consequences of each branch, points where self/mutual-regulation techniques apply
- Step 4: Cross-check proposed "correct" responses against relevant ethics codes/statutes (cite the specific article/clause)
- Step 5: Adapt scenario difficulty/context to real conditions of the group and hand off to facilitator for live run
- Step 6: Debrief template — what worked, what codex clause applies, what to drill next
Example 1 (Track A): Input: Client PHQ-9 scores over 8 sessions: 18, 16, 15, 14, 14, 13, 14, 13 Output: "Sessions 1-3: active improvement (-4 pts). Sessions 4-8: plateau (net -1 pt over 5 sessions, within measurement noise). Flag: possible plateau starting session 4. Hypotheses: (1) initial symptom relief from behavioral activation has run its course, (2) check for unaddressed cognitive component, (3) rule out external stressor onset around session 4. Client view: 'Great progress in the first month — now let's dig into what's keeping scores steady.'"
Example 2 (Track B): Input: 40 anonymous survey responses from a department, free-text on "team challenges" Output: Themes ranked by frequency: "unclear task ownership" (62%, negative sentiment), "favoritism in shift assignment" (38%, negative, high emotional intensity), "lack of feedback from leads" (45%, neutral-negative). Cohesion score: 2.4/5. Flag: "favoritism" theme shows coded language ("some people always get the easy shifts") suggesting unspoken resentment — recommend structured 1:1s before an all-hands discussion.
Example 3 (Track C): Input: Target stressor = "потеря связи и дезориентация" for a field unit Output: Scenario script "Loss of Contact" — unit loses radio contact during exercise, must execute self/mutual-regulation protocol (box breathing, buddy check-in, decision to hold position vs. relocate), decision tree with 3 branches, debrief checklist mapping to unit's SOP on communication loss.
- Always separate descriptive flags (plateau, regression, risk signal) from clinical judgment — the output surfaces patterns, the licensed professional interprets and acts.
- For psychometric data, require minimum 3 data points before declaring trend; note measurement error/noise explicitly.
- In team analysis, never attempt to de-anonymize or infer individual authorship from text style.
- For ethics scenarios, always ground the "correct" response in an actual cited code (APA, local professional board, organizational charter) — don't invent generic "best practice" without a citation.
- Calibrate stress-scenario intensity to the group's actual exposure level; escalate gradually across a training program rather than maximal intensity on day one.
- Pair any stress-inoculation exercise with an explicit debrief and regulation-technique practice, not just exposure.
- Don't diagnose causes of plateau/regression as fact — always frame as hypotheses for clinician review.
- Don't treat a single bad survey theme as proof of systemic conflict — corroborate with frequency, intensity, and cross-reference with other channels.
- Don't generate ethics scenarios without codex cross-referencing — a scenario with no citation is just fiction, not training material.
- Don't run high-intensity stress simulations on vulnerable groups without a trained facilitator and informed consent/debrief plan.
- Don't present clinician-level statistical detail to clients — always produce the simplified version for client-facing visualization.