Kahlus-Affect / ResearchDock Roadmap#
Need Statement#
A low-cost system is needed to quantify reward, effort, and stress-recovery response profiles using synchronized pupil, behavior, physiology, and optional neural signals so that anhedonia-like and trauma-related state changes can be explored more objectively than surveys alone.
This branch measures response profiles. It does not diagnose anhedonia, depression, PTSD, or any clinical condition.
Existing At-Home Biosensors Are Ingredients, Not the Product#
At-home and point-of-care saliva workflows already exist for cortisol and salivary alpha-amylase, including self-collection, smartphone/strip/electrode readout, and comparison against lab assays. Therefore, Kahlus RewardDock should not be framed as inventing a generic cortisol or stress meter.
The product wedge is disease- and task-specific interpretation:
standardized reward/stress/social-response tasks
pupil, behavior, HRV/PPG, and optional saliva biomarkers
within-person trend modeling
clinician-facing treatment-response summaries
RewardDock Clinical Extension#
Kahlus RewardDock Clinical =
standardized task battery
webcam pupillometry
reaction time / effort behavior
optional PPG/HRV
optional cortisol / alpha-amylase module
Kahlus response-profile model
clinician dashboard
Primary clinical wedge:
Adults with anhedonia-related depression or social anxiety who are starting or adjusting treatment.
Need statement:
A way to objectively track reward-response and stress-recovery changes in adults undergoing treatment for anhedonia-related depression or social anxiety in order to help clinicians identify ineffective treatment plans earlier than self-report alone.
Why This Is Not Just a Better Survey#
The goal is not to replace interviews or self-report. The goal is to add repeated within-person response data under standardized conditions.
Kahlus asks:
Is this person’s reward/stress/social-response profile improving relative to their own baseline?
Not:
Does this person have depression?
Does this person have anhedonia?
Biomarker Architecture#
Core v0:
webcam pupil dilation
reaction time
effort persistence
task accuracy
self-report slider
Core v1:
PPG/HRV
heart-rate recovery
optional EDA
Future biochemical add-on:
salivary cortisol
salivary alpha-amylase
Avoid as primary:
dopamine
oxytocin
epinephrine/norepinephrine direct measurement
Dopamine, oxytocin, and epinephrine are not practical home biomarkers for this use case. Peripheral levels do not cleanly map to central reward circuitry, assays are difficult or context-dependent, and they are not condition-specific enough for RewardDock v0.
Cortisol and salivary alpha-amylase are more realistic stress-context add-ons, but they should not be the core product.
Clinical-First, Wellness-Later Product Path#
Phase 1: Clinical/research use
treatment-response tracking
psychiatry clinics
therapy programs
clinical trials
MDD/anhedonia/social anxiety cohorts
Phase 2: Research/remote monitoring
longitudinal digital phenotyping
stress-recovery studies
reward learning studies
Phase 3: Consumer/self-improvement
motivation tracking
burnout recovery
focus/reward habits
social-confidence training
Do not start with self-improvement branding. Start clinical/research first to avoid becoming a generic dopamine-maxxing wellness app.
RewardDock Claim Boundaries#
Allowed:
tracks reward-response profile
tracks stress-recovery profile
tracks within-person change over time
supports treatment-response research
supports clinician review as an adjunct signal
Blocked:
diagnoses depression
diagnoses anhedonia
diagnoses social anxiety
recommends medication
adjusts medication
treats depression
treats PTSD
replaces clinician judgment
claims cortisol/alpha-amylase alone measures anhedonia
RewardDock Device Roadmap#
RewardDock v0:
webcam pupil + reward task + reaction time + self-report
RewardDock v1:
add PPG/HRV
RewardDock v2:
add optional saliva cortisol / alpha-amylase integration
RewardDock v3:
clinician dashboard + longitudinal treatment-response report
RewardDock Consumer:
future nonclinical self-improvement product only after clinical/research validation
ResearchDock v0 Sensors#
webcam pupil/gaze
reaction time
task accuracy
self-report sliders
optional PPG/HRV
optional EEG/fNIRS later
Safe Task Battery#
reward anticipation
probabilistic reward learning
effort-for-reward
mild frustration/stress task
recovery/rest block
visual attention block
No trauma exposure, treatment, stimulation, or unsupervised photic testing is part of RD-0.
Metrics#
reward response delta
reaction time change
effort persistence score
pupil response amplitude
pupil recovery slope
PPG/HRV proxy summary
task accuracy summary
response profile vector
Baselines#
neutral-condition mean
participant-level persistence
ridge regression for response-profile prediction
simple MLP after enough real data exists
Evaluation Criteria#
no personally identifying fields
synchronized task and sensor timestamps
quality flags for missing pupil, high noise, dropped packets, and invalid reaction time
deterministic synthetic fixtures
finite metrics under missing sensors
gate blocks clinical claims
Buildable In 8 Weeks#
RD-0 synthetic schema, metrics, data card, and evidence gate
local task app/session prototype with CSV/session export
design-only webcam/pupil interface contract and quality flags
optional PPG/HRV input schema
public dataset mapping review for WESAD, DEAP, and SEED
RD-1 Local Prototype Boundary#
The first prototype is a local session protocol and export contract. It writes task/session CSVs and a hardware-free interface contract; it does not open a webcam, collect PPG, or ingest real participant data.
RD-2 Synthetic Observation-Model Boundary#
The RD-2 model lane uses synthetic ResearchDock sessions only. Behavior/task/self-report features predict pupil and HRV proxy observations under a subject-held-out split. Mean and ridge baselines are evaluated before the ResearchDock observation-operator candidate. This is synthetic pretraining infrastructure, not a clinical model.
RD-3 Public Dataset Review Boundary#
RD-3 maps WESAD, DEAP, and SEED to ResearchDock fields without adding loaders or downloading data. WESAD is the strongest immediate physiology/stress candidate; DEAP is useful for affective EEG plus physiology only after access terms are rechecked because the historical dataset page was unavailable during review; SEED is useful for EEG plus eye-movement missing-modality experiments after application and license approval.
The RD-3 artifacts are review outputs only. They do not contain raw participant data, do not claim diagnosis, and do not make any dataset ingestion path executable.
RD-4 Pilot Preflight Boundary#
RD-4 is a pre-collection readiness artifact for a future validation-scale ResearchDock pilot. It writes a pilot manifest, required-evidence checklist, and preflight gate using synthetic fixtures and the RD-1 protocol only.
RD-4 does not collect participant data, open hardware, provide legal or IRB advice, diagnose, treat, stimulate, or make clinical claims. Any real pilot remains blocked until appropriate review, local storage outside source control, identifier handling, and artifact archival are handled outside this code-only scaffold.
RD-7 Observation Missing-Modality Report Counts#
The synthetic observation-model report now includes per-reason missing-modality counts for skipped trials. This keeps missing pupil, HRV proxy, sensor-packet, and behavior-response exclusions visible in the human-readable artifact, not only in JSON metadata.
RD-8 Observation Split Report Summary#
The synthetic observation-model report now renders the subject-held-out split summary with train/test subject counts and subject-overlap status. This keeps split hygiene visible beside baseline and candidate metrics in the reviewer-facing artifact.
RD-9 Observation Split-Audit Sidecar#
The synthetic observation-model artifact writer now emits researchdock_observation_split_audit.json
with subject-held-out split counts, subject-overlap status, a leakage pass boolean, and explicit
failure reasons. This gives audit consumers a machine-readable split gate alongside the report.
RD-10 Observation Report Split-Audit Verdict#
The synthetic observation-model report now renders the split-audit verdict and any split-audit failure reasons from the same helper that writes the JSON sidecar. This keeps blocked split evidence reviewable without opening the sidecar first.
RD-11 Aggregate Observation Artifact Index#
The top-level ResearchDock synthetic report now indexes the RD-2 observation artifacts, including
researchdock_observation_split_audit.json. This makes the aggregate report point reviewers to the
machine-readable split audit, baseline table, metrics, and observation report.
RD-12 ResearchDock Gate Criteria#
The ResearchDock evidence gate now records branch-specific criteria in researchdock_evidence_gate.json
and renders them in the top-level report: allowed claim scope, blocked clinical/device terms, required
data-card pass, baseline table, finite metrics, calibration check, and synthetic-only boundary.
RD-13 ResearchDock Failure-Reasons Sidecar#
The top-level ResearchDock synthetic runner now writes researchdock_failure_reasons.json with gate
failures, data-card safety failures, and blocked claim terms. Passing synthetic fixtures still emit
the sidecar with empty failure lists so audit consumers have a stable artifact contract.
RD-14 Top-Level Evidence Artifact Index#
The top-level ResearchDock synthetic report now includes an Evidence Artifact Index listing the
metrics, data card, evidence gate, failure-reasons sidecar, and report artifacts. This makes the
local evidence bundle navigable from the Markdown report.
RD-15 Top-Level Failure Reasons Summary#
The top-level ResearchDock synthetic report now summarizes the same failure-reasons payload written
to researchdock_failure_reasons.json: gate failure count, data-card failure count, and blocked
claim-term count. This keeps passing and blocked bundles quick to audit from Markdown.
RD-16 Top-Level Failure Reasons Details#
The top-level ResearchDock synthetic report now renders failure-reason detail lists from the same
payload written to researchdock_failure_reasons.json, using none for empty gate/data-card failure
lists and listing blocked clinical/device claim terms explicitly.
RD-17 Optional Artifact Index Expansion#
The top-level ResearchDock synthetic report now expands its Evidence Artifact Index when optional
RD-1 session export, RD-2 observation model, RD-3 public dataset review, RD-4 pilot preflight, and
RD-5 profile-readiness artifacts are requested. This keeps optional local evidence bundles
navigable from the Markdown report instead of requiring stdout or directory inspection.
RD-18 Top-Level Data Card Summary#
The top-level ResearchDock synthetic report now renders bounded data-card fields directly in
Markdown: PII/real-data/clinical-label/stimulation booleans, modalities, synthetic profiles, and
quality flags such as missing pupil. This keeps safety and quality evidence visible without opening
researchdock_data_card.json.
RD-19 Top-Level Quality Flag Counts#
The top-level ResearchDock synthetic report now renders deterministic session-level quality flag
counts from the metrics table, including missing_pupil and synthetic_high_noise. Counts are
session-grained rather than trial-grained so the report matches the synthetic metrics table and does
not inflate repeated per-trial quality flags.
RD-20 Aggregate RD-2 Baseline Ladder Summary#
When the RD-2 synthetic observation model is requested, the top-level ResearchDock report now renders the baseline ladder directly: mean baseline, ridge baseline, then ResearchDock observation operator, with MSE/MAE plus the best-baseline verdict. This makes baseline results visible before the candidate model in the aggregate evidence bundle.
RD-21 Aggregate RD-2 Split Audit Summary#
When the RD-2 synthetic observation model is requested, the top-level ResearchDock report now renders the subject-held-out split audit summary: train/test subject counts, subject-overlap status, leakage-pass verdict, and failure reasons. This keeps split hygiene visible in the aggregate bundle without requiring reviewers to open the RD-2 sidecar first.
RD-22 Aggregate RD-2 Missing-Modality Summary#
When the RD-2 synthetic observation model is requested, the top-level ResearchDock report now renders the missing-modality audit summary: total, eligible, and skipped trial counts plus per-reason skipped trial counts. This keeps missing pupil and missing behavior-response exclusions visible at the aggregate evidence-bundle entry point.
RD-23 Aggregate RD-5 Readiness Audit Summary#
When the RD-5 response-profile readiness audit is requested, the top-level ResearchDock report now renders the readiness scope, clustering-disabled status, metric-row threshold, finite-vector verdict, and failure reasons. This keeps future-clustering blockers visible without requiring reviewers to open the RD-5 sidecar first.
Long-Term#
real pilot usability study after review and consent workflow
multimodal v2 observation model
subject adaptation design
ResearchDock to Kahlus v3 response-profile data flywheel
Citations To Verify#
Reference |
Why It Matters |
citation_status |
|---|---|---|
Brendler et al. 2024 Scientific Reports pupil/reward/anhedonia |
pupil reward-response precedent |
needs_verification |
Fietz et al. 2024 Biological Psychiatry CNNI pupil response profiles |
pupil response-profile precedent |
needs_verification |
WESAD |
stress physiology dataset candidate |
verified_rd3_source_review |
DEAP |
affect physiology/EEG dataset candidate |
partly_verified_rd3_dataset_page_unavailable |
SEED |
affect EEG and eye-tracking dataset candidate |
verified_rd3_access_requires_application |