# Askable Labs catalogue — Keys

Generated: 2026-07-21 · Reference tables for what a Labs delivery contains and how it is produced: capture moments (M01–M16), deliverable types (A/B/C tiers), and the T01–T06 capture stack. These apply across all four data types; the specific mix is set per brief.
The four data types: https://labs.askable.com/agents/catalogue.md

## Capture moments key

- **M01** — Flow-state stretches
- **M02** — Reasoning narration
- **M03** — Expert-led correction
- **M04** — Genuine confusion
- **M05** — Disagreement and resolution
- **M06** — Decision-point hesitation
- **M07** — Backtracking and recovery
- **M08** — Tacit knowledge surfacing
- **M09** — AI-acceptance moments
- **M10** — AI-rejection moments
- **M11** — Tool-switching reasoning
- **M12** — Edge-case encounter
- **M13** — Pedagogical pacing
- **M14** — Identity reveal
- **M15** — Negotiation under constraint
- **M16** — Performance of judgment

## Deliverable types key

- **A1** (tier 1) — Composite recording (single MP4)
- **A2** (tier 1) — Per-participant audio tracks (Opus/OGG, 48 kHz)
- **A3** (tier 1) — Per-participant screen capture (VP8/WebM)
- **A4** (tier 1) — Per-track wall-clock JSON (Unix nanoseconds)
- **A5** (tier 1) — Per-session track manifest
- **A6** (tier 1) — Recording infrastructure provenance
- **A7** (tier 1) — Cross-correlation alignment provenance
- **A8** (tier 1) — Plain-text timestamped transcript
- **A9** (tier 1) — Facilitator track flagged in transcript
- **A10** (tier 1) — Track segmentation honesty (gaps + restarts)
- **A11** (tier 1) — Participant demographic record
- **B1** (tier 1) — Submission record (single JSON per submission)
- **B2** (tier 1) — Verbatim AI prompts with turn index
- **B3** (tier 1) — Attachment descriptions (type, subject, intent)
- **B4** (tier 1) — AI tool context (tools used, frequency, fluency)
- **B5** (tier 1) — Task goal statement (verbatim)
- **B6** (tier 1) — Outcome self-assessment scores
- **B7** (tier 1) — Quality-score reasoning (verbatim)
- **B8** (tier 1) — Where-it-worked / where-it-struggled breakdown
- **B9** (tier 1) — Time-to-result comparison (with-AI vs without)
- **C1** (tier 2) — Aligned word-level transcripts (diarised JSON)
- **C2** (tier 2) — Decision-point annotations (human-tagged)
- **C3** (tier 2) — Reasoning-segment annotations (human-tagged)
- **C4** (tier 2) — Application-switch events (derived from screen)
- **C5** (tier 2) — Mouse-path stream
- **C6** (tier 2) — Keystroke timing (rhythm-only by default)
- **C7** (tier 2) — Tool-state events (form opens, modals, pages)
- **C8** (tier 2) — Voice-activity detection stream
- **C9** (tier 2) — Per-track quality summary
- **C10** (tier 2) — Attachment binaries (with consent)
- **C11** (tier 2) — AI model outputs paired to prompts
- **C12** (tier 2) — Inter-prompt timing
- **D1** (tier 3) — Expert-side screen capture
- **D2** (tier 3) — Webcam video per participant
- **D3** (tier 3) — Eye-tracking
- **D4** (tier 3) — External-instrument capture
- **D5** (tier 3) — 3+ participant configuration
- **D6** (tier 3) — Live narration alongside Mode B submission
- **D7** (tier 3) — Aligned target-language transcript
- **D8** (tier 3) — Code-switching annotation layer
- **D9** (tier 3) — Multi-expert divergence markers
- **D10** (tier 3) — Prediction-then-reveal markers
- **D11** (tier 3) — Explicit rubric and ranking artifact
- **D12** (tier 3) — Asynchronous review artifact
- **D13** (tier 3) — Paired-session metadata (matched A/B tasks)
- **D14** (tier 3) — Final artifact produced in session

## Proprietary capture stack (T01–T06)

Every Labs capture is produced through one audited platform — one identity system, one consent ledger, one audit log, one delivery pipeline. Operating since 2017; 230,000+ research sessions run to date.

- **T01 · Sourcing — Participant panel.** ~900,000 identity-verified humans across 75 countries. Triple-verified, paid, consenting. Any demographic, industry, or geography, including the niche audiences others can't reach.
- **T02 · Operations — Participant management engine.** Scheduling, calendar sync, SMS & email reminders, identity verification, brief-specific consent, incentive payments, and automated replacements. The entire participant lifecycle, productised.
- **T03 · Session — Video call platform.** LiveKit-powered real-time A/V, screen-share, and multi-party rooms. Sessions recorded against the consented brief, encrypted in transit and at rest, multi-region routing for low-latency capture.
- **T04 · Capture — Multi-method capture suite.** Moderated interviews, AI-moderated interviews, remote and live mobile/website testing, prototype tasks, surveys, diary studies. Every modality is a code path on the same platform, mixable per brief.
- **T05 · Unmoderated — Human-behavior testing suite.** Card sorting, tree tests, first-click tests, five-second tests, stack ranking, prioritization. Structured, unmoderated tasks run at scale, producing clean, quantifiable behavioral signal alongside qualitative capture.
- **T06 · Processing — Data processing engine.** Real-time transcription, speaker diarisation, semantic embedding, AI annotation, and evidence-linked output structuring. Every finding traces back to a timestamp, a quote, and a participant.

## Provenance

- **Participant verification.** Identity triple-verified at recruitment through a multi-layered verification stack, and re-verified at session.
- **Consent.** Per-session, brief-versioned, signed before capture. Specifies which downstream uses the lab has rights to: training, evaluation, research publication, model-card attribution.
- **Audit trail.** Recruitment, consent, capture, transcription, annotation, and delivery all run as system calls against the Askable production platform. Every step logged with timestamps and operator identity. No manual chain of custody.