feat(behaviors): tf.behaviors.discover() — emergent behavior discovery

[rrader26]

The final slice of behaviors.discover — the public SDK method. Completes the path: engine core (#39) → DiscoverBehaviors RPC (#40) → HTTP proxies (memory-thinkfleet #35, #36) → SDK.

What

tf.behaviors.discover(params?) → POST /lattice/discover. Returns clusters of like-behaving subjects the engine found in the data (not a fixed menu):

const { behaviors } = await tf.behaviors.discover()
for (const b of behaviors) {
  console.log(`${b.label} — ${(b.prevalence*100).toFixed(0)}% of subjects, stability ${b.stability.toFixed(2)}`)
}

DiscoveredBehavior: label, prevalence, stability (cohesion), size, memberSubjects (medoid first), exemplarEvidence. An empty result means the engine abstained — not enough signal — never "no behaviors exist".

Notes

• New BehaviorsResource wired into the client as tf.behaviors; types exported (DiscoverParams, DiscoveredBehavior, DiscoverResult).
• Depends on memory-thinkfleet #36 (the /lattice/discover route) being deployed for runtime; compiles independently.

Verification

tsc clean; tsup build emits ESM/CJS/DTS; tf.behaviors.discover + types public in dist/index.d.ts.

Remaining (engine follow-ups, separate)

• LLM naming of discovered clusters (neuro-symbolic)
• promotion to behavior_pattern + feed into prediction features

🤖 Generated with Claude Code