The senior engineer's point of contact: eliciting truth from clients, capturing it cleanly, and pulling accumulated knowledge back into the conversation in real time.
The engineer's job is shifting from writing code to managing coding agents. The new bottleneck is no longer technical — it's human intake. This project investigates what a senior software engineer needs to learn, from intelligence tradecraft, requirements-engineering disciplines, capture law, and Karpathy-flavored LLM-Wiki practice, to handle the consultant–client point of contact professionally.
The piece is built on a tension: the rigorous systematic Discovery process Sextant has already implemented (the destination of the funnel) meets the warm, ambiguous, real-time human conversation that feeds it. Source Code is the research project on the human side of that surface.
This project is organized into five research domains. All five are Wave 1, active, and independent. Within each domain, lenses provide distinct angles of analysis.
Status: active Unlocked by: —
Intelligence-gathering as a defined discipline. The literature, doctrine, and field practice of getting information from humans, drawn from the institutions that have studied this longest and most adversarially.
| # | Lens | Description |
|---|---|---|
| 1 | Elicitation | The named skill of conversational information-gathering that doesn't feel like questioning |
| 2 | HUMINT & Source Handling | How intelligence services run a relationship with a human source over time |
| 3 | OSINT & Pre-Engagement Dossier | Open-source intelligence as preparation; the discipline of pre-meeting research |
| 4 | Structured Analytic Techniques | Heuer, ACH, and the intelligence-analyst toolkit for reasoning about ambiguous information |
See tradecraft/README.md for full lens descriptions and source types.
Status: active Unlocked by: —
The same underlying skill — eliciting truth from humans — as it lives in respectable professions with mature, public literatures. Where Tradecraft is shadow-discipline, this is the daylight craft.
| # | Lens | Description |
|---|---|---|
| 1 | The Business Analyst | BABOK elicitation taxonomy and the requirements-engineering tradition |
| 2 | The Investigative Journalist | The interview as a profession |
| 3 | The Ethnographer & User Researcher | Eliciting unstated and tacit knowledge |
| 4 | The Clinician's Interview | Motivational interviewing, reflective listening, and the therapeutic stance |
See practitioners-craft/README.md for full lens descriptions and source types.
Status: active Unlocked by: —
Converting fluid live conversation into typed artifacts that can feed a planning funnel — legally, ethically, and cleanly. The capture problem as discipline.
| # | Lens | Description |
|---|---|---|
| 1 | Recording Law | Consent jurisdictions, GDPR, fiduciary obligations |
| 2 | Transcription & Tooling | State of the art in 2026; latency / accuracy / security trade-offs |
| 3 | The Notebook Tradition | Contemporaneous notes as professional craft |
| 4 | The Two-Phase Bridge | Reason-freely-then-structure; the source literature behind Sextant's extraction step |
See recorders-posture/README.md for full lens descriptions and source types.
Status: active Unlocked by: —
Karpathy's territory. The consultant's accumulated knowledge corpus wired into the conversational framework as a personal LLM Wiki — a queryable, agent-augmented memory that retrieves and communicates relevant understanding in real time, across both written and spoken surfaces.
| # | Lens | Description |
|---|---|---|
| 1 | The LLM Wiki Thesis | Karpathy's specific framing; what makes the LLM Wiki different |
| 2 | Corpus Curation | What goes in; provenance, recency, decay, trust |
| 3 | Real-Time Retrieval | The latency-vs-relevance problem under conversation-speed constraints |
| 4 | The Conversation Surface | Meeting copilots, sales-engineering tooling, written and spoken surfaces |
| 5 | The Look-Up Stigma | The cultural cost of visibly retrieving mid-conversation |
See knowledge-front/README.md for full lens descriptions and source types.
Status: active Unlocked by: —
Interpersonal craft as a learned skill, not a personality trait. The literature, training traditions, and lived practice of people who built this skill out of necessity rather than inheriting it. Framed throughout as the author's own learning project, not as methodology dispensed from competence.
| # | Lens | Description |
|---|---|---|
| 1 | Learned Extraversion | The empirical literature on whether interpersonal skill is trainable |
| 2 | Energy Management | The introvert's resource problem |
| 3 | Performance Ethics | Manufactured warmth — integrity or inauthenticity? |
| 4 | The Sales Training Traditions | Sandler, Challenger, SPIN, MEDDIC — what each actually teaches |
See apprenticeship/README.md for full lens descriptions and source types.
Each domain produces a Domain Synthesis rolling up its lens findings. The Cross-Domain Synthesis integrates all domain syntheses into a unified analysis — identifying convergences, tensions, and implications that only become visible when domains are laid side by side.
- Research — Documents land in each lens folder as markdown files, cite-at-creation, confidence-tagged
- Consistency check — Per-domain pass catches contradictions, then project-wide
- Three-layer review — Gap analysis first (with corrections), then source quality, then fact verification
- Synthesize — Lens summaries → domain summaries → cross-domain synthesis
Domains can be researched and reviewed independently. Cross-domain synthesis requires all domain summaries to be complete.
See CLAUDE.md for full methodology.
The design spec for this research project lives in the parent Legacy Code repo at docs/superpowers/specs/2026-05-22-source-code-research-design.md.