Every view below is generated from the same file-cited data source — 48 real services, 23 real control & data paths, traced from the code at latest main across eight repos. The first three show the whole system; the last four trace one flow at a time, because the whole system at once is too noisy to read.
The GPU in the rack is not decoration: in the 24 hours before this was drawn it ran 161 LLM jobs for the harness on a resident gpt-oss:20b (measured 2026-10-20; the breakdown is in the Ollama inspector on any map).
Start with the guided isometric tour. If you want the version without the city, open the Decision Graph.
Services as stations, flows as colored lines, opportunity_harness as the interchange everything converges on.
why it exists: maximum legibility — you can trace any payload end-to-end in seconds THE SHOWPIECEHarness City Max — 3D night cityThousands of ambient buildings, bloom, glow-sprite payloads — the real services as glowing landmark towers, click-to-inspect.
why it exists: the feel of the system — scale, activity, one bright decision layer in a dark city THE TOURHarness City — isometric, guidedInteractive isometric map with "follow the payload": an 8-step narrated walk of a signal from venue API to inbox alert.
why it exists: the only version that teaches the flow step-by-step instead of showing it all at onceOne detection cycle pulled from the rack's Tempo: 34 venue snapshots, then detect and two paper fills, drawn to scale with a playhead.
why it exists: it is not a drawing — every bar is a real span with its real duration and event count THE LINEOne Line — one journey at a timeEight real journeys, each drawn as a single straight metro line with the flow label and cadence on every segment and a payload that runs it.
why it exists: the metro map with everything removed except the path you asked about THE BOARDCircuit — one net lit at a timeEvery service a chip, every flow a copper trace on two layers. Quiet at rest; hover a chip and only its nets glow, with a probe readout.
why it exists: the whole topology stays on screen but only one thing is ever emphasized THE GRAPHDecision Graph — if the agent takes this pathFour real control flows as decision graphs: LLM steps, deterministic tool steps, branches read from the code. Pick a path and the lit route explains itself.
why it exists: the business-friendly one — what happens, and where it ends, without the city