← Back to Observatory ENGINEERING LAB · MULTI-AGENT DAG 8-Year Folio →
04 // LIVE ORCHESTRATION SANDBOX

Run the Autonomous Agent Pipeline

Pick one of 4 real shipped missions and execute the 5-stage DAG right in your browser — watching how Curator, Architect, Parallel Builders, and Independent Playwright QA ship production code.

4m 18s simulated · 1,152 tests · 12 refs · 0 ambiguity

Mobbin Research & Scoped Blueprint

Zero designing from memory or generic AI templates. Every product starts by locking 10–20 real shipped references into a Mobbin collection and writing `docs/blueprint.md` with explicit non-negotiables.

[01 · CURATOR] Locked 12 Mobbin B2B fintech & workspace references -> docs/blueprint.md approved. 12 refs · 0 ambiguity
$ mobbin search_sections "developer platform architecture diagram" -> 20 keepers locked
$ cat docs/AWWWARDS_V2_BLUEPRINT.md
> NON-NEGOTIABLES: 13 pre-rendered routes | 0 dead ends | YekanBakh + IRANYekanX VF | 0px overflow
DETERMINISTIC QUALITY GATES

Why My Agent Pipeline Ships Production Software, Not Fragile Demos

GATE 01

Builder Never Self-Certifies

Every UI or backend slice built by a worker agent is inspected by a separate read-only critic agent + Playwright bounding-box rect assertions.

GATE 02

Real Mobbin References First

Zero generic AI layouts from memory. Every screen anchors to documented shipped product patterns saved in Mobbin collections.

GATE 03

Desktop 1440px + Mobile 390px Zero Overflow

Automated headless Chromium sweeps enforce `scrollWidth <= innerWidth`, zero console errors, and genuine variable Persian fonts in RTL.

GATE 04

Zero Secrets & Zero Fabricated Claims

Machine-enforced validators (`FACTS.md`, `80-owner-resume.md`, `0600` out-of-repo credentials) guarantee factual & security hygiene.