Nishi — Design→Market Engine dmkt lane

Idea → CAD → live sourcing → print → finance → demand → GO/NO-GO, generated by recombining the atlas. RACI: R=researcher · A=pm · C=engineer · I=fin. Maturity (ecomat, honest): design-market TOY → EXCEED — no parity claim. Live 2026-07-19.

We don't out-kernel Siemens NX geometry — we out-loop the category: generate novel viable products by recombining our own 16,575-node capability atlas, source + price them with live data, roll cost to a finance ledger, score demand, return a GO/NO-GO, and bank every result back to the atlas so the next cycle is smarter. Generative + closed-loop + compounding.

Pipeline — the coordinated backlog (frontier CPM)

Rung	Owner (R)	Status	What
F310	researcher	SHIPPED	atlas recombination ideator (nx_dmkt_ideator) — built, MCP-registered, ran
F310b	researcher	BLOCKED	nx_cap_harvest broad catalog (blocked on build-diagnostics debt seq117)
F311	engineer	READY	CAD parametric + STEP export (composes the PRODUCTION cad domain)
F312	engineer	READY	live electronics sourcing connector (BOM → price/stock/lifecycle)
F313	conductor	READY	print + quote connector (home printer + Xometry/PCBWay/JLCPCB)
F315	fin	BLOCKED F312	cost → margin → viability rollup on nx_fin_serve
F316	pm	READY	propensity + VHECT demand scorer (needs operator's VHECT model)
F317	pm	BLOCKED	channel viability GO/NO-GO gate (Amazon + eBay + own-site) — milestone M9
F318	librarian	BLOCKED	expose board tool + page + raci/ws rows

Hard evidence — the recombination engine RAN (not asserted)

F310 (nx_dmkt_ideator) executed via the workflow layer (nx_plan_run dmkti = steps=1 ok) and emitted 5 real need→capability composition hypotheses to the dmktideas- plane — and correctly left nx_render3d unmatched (no force-fit):

Product need	→	Composed capability (proposed)
geometry	→	nx_cad
sourcing	→	nx_supply_research_fetch
battery	→	nx_battery_research_fetch
demand	→	nx_market_research_fetch
cost	→	nx_fin_research_fetch

These are HYPOTHESES (status=proposed) — the engine's honesty gate promotes a combo to a capability only after a plan run PROVES it. And the engine generalized: nx_recombine is the same elf, arg-driven, usable by any lane (proven writing to a different out-plane).

Pilot & channels

Pilot = electric sonic massage hairbrush (scalp massager) — exercises the full loop: live electronics (haptic driver, Li-ion, PCB, USB-C) AND molding/print (housing + dual-durometer silicone). Channels modeled: Amazon + eBay + own nishifamily.com store.

On MCP / API / workflow / agent / RACI / atlas (all live)

MCP: nx_dmkt_ideator + nx_recombine registered · API: built/promoted/registered/cap-minted via the mgmt API · Workflow: ran via nx_plan_run · RACI: dmkt + recombine rows on the canonical raci- store · Agent: product crew role filled · Atlas: F224 eaten (graph rebuilt 16,575 nodes) · Hubbed: all planes NAS-side.

Honest gaps: F312/F313 sourcing+print need live external APIs (Octopart/Xometry) · F316 needs the operator's VHECT (visibility-hype-engagement-conversion-trust) model · F310b scale-harvester blocked on the build-diagnostics debt (seq117). Source: designmarket- + dmktideas- sovereign planes + frontier CPM; served by nishi-substrate-v2. WAN reachability gated on D008.