Learning Platformindex / methodD · method

The Method & Map · D

Design Method

How every track gets decided — not by assertion, but by adversarial pressure-testing against real options and what we've already built.

The four-step method (per track)

  1. Frontier research sweep — what's new/trending in AI-native learning + premium-community + conversational-data + agentic patterns. Study OSS and paid products for ideas; "do it better." Brand-free outputs.
  2. Thinker-canon design panel — an adversarial panel critiques the architecture per track from distinct lenses (data/distributed · build-vs-adapt · AI-native · simplicity/ops), plus an estate-scout that honestly assesses reuse. → ADRs with 2–4 options, tradeoffs, a recommendation, and preserved dissent.
  3. Estate-integration map — each component → its integration seam, grounded in the real repos.
  4. Agentic where it earns it — reasoning-heavy flows run as governed headless agents (agent-governance plane + borrowed harnesses) rather than inline calls.
Why adversarial. On ADR-001, the five lenses ranked the four options differently — and that divergence forced the real answer, which none of the four options was on its own. A single-perspective design would have shipped one lens's blind spot. See the panel in full.

The interaction model (aligned)

Not a single spine — a stack: an AI-curated home (universal front door; kills the firehose; powered by the brain + agents = the moat) → threaded boards (substance) → scheduled live events (heartbeat, not always-on voice) → DMs (utility) → tiers gate depth. Members mostly already know their craft; "come and learn" is the credible front (Guest), the money + moat live in Member + Inner Circle. Choose-your-own-home per member.

Hard constraints (the rails)