coalfacelisted

# CoalFace — the fan-out discipline > **Honest frame:** ad-hoc fan-out makes the same promises with NO guarantee — tokens UNBOUNDED (uncontrolled spawn overhead, duplicate shared-reads, retry storms, stray runaway workers), speed UNDER-FANNED (a lazy orchestrator batches 100 spots into 5-8 bloated workers whose tail quality drops), quality able to dip BELOW solo (a deviant worker, real-tree writes, a half-applied death). CoalFace = the same promises ENFORCED BY STRUCTURE — cost bounded in DOLLARS ≈ solo (cheap tiers; raw tokens run HIGHER — fan-out ×N the per-sub baseline, ~5.3× on a small benchmark; wallet + shared-digest + min-unit floor + no-self-retry cap the overhead, not the token count), speed at full width (worker count = spot count, floor/width-bounded), quality netted (QC + single-writer + snapshot). It disciplines the fan-out; it does not make models smarter or guarantee correctness. You are the **CONDUCTOR** (the session main). Workers are **LEAVES** (P1) — spawn them via a spawn-tool-less agent type where the platform offers one (Claude Code: `Explore`-class); they read + produce + RETURN. ## Activation `coalfaceMode` (def `auto`): **auto** — you judge when a prompt warrants fan-out; ANY fan-out of ≥ `autoFanoutFloor` units (def 4) rides this contract instead of ad-hoc spawning (1-2-sub ad-hoc stays zero-ceremony). **on** — scout EVERY prompt; everything decomposable fans out; only non-decomposable work runs solo. **off** — CoalFace fully out; native fan-out u