skills/lq-reflect/references/technique-ladder.md

# The technique ladder

Three tiers, matched to the levels. Teach ONE rung at a time, demonstrated on the
user's own work, and only after an aha — the why comes after the magic, not before.
Debrief lessons and office-hours focus areas draw from the user's current tier;
a technique observed from a higher tier is evidence for the inferred level.

## Day One → toward Early Builder
- **Meta-prompting** — ask the model to improve the prompt before running it; the
  role/inputs/checks/output structure.
- **Model switching** — different models for drafting vs checking vs long documents;
  a second model as reviewer.
- **Rendered deliverables** — turn analysis into a clean client-ready page instead
  of a wall of text.
- **Verification habit** — never let an authority into work product unchecked; ask
  for answer + method + confidence in one prompt.

## Early Builder → toward Legal Quant
- **Second brains** — a personal wiki of the knowledge worked out on matters;
  record method, never the matter.
- **Building skills** — turn the workflow you repeat into a reusable, testable
  instruction set with its own checks.
- **Context management** — what to load, what to summarise, what to leave on disk;
  cost and context-window awareness.
- **Precise failure naming** — hallucination vs drift vs truncation vs bad
  retrieval; you cannot fix what you cannot name.

## Legal Quant — frontier maintenance
- **Loop and graph engineering** — multi-step, multi-agent workflows with
  receipts; parallel review lanes; evals that prove the loop.
- **Your own data layer** — structured stores and connectors over your practice's
  sources; retrieval you control.
- **Encoding legal logic in structure** — the reasoning lives in the tool's shape,
  not in a prompt you hope it follows.
- **Responsible deployment** — confidentiality boundaries, approval design, and
  knowing exactly where the tool would run.