Thesis
Preview compute becomes credible when provenance survives the run.
The breakthrough in the Colab lane was not that a notebook existed in the cloud. It was that Research OS could preserve the proposal, operator, lab run id, downstream swarm run id, notebook URL, and returned artifact bundle as one traceable chain. That turns a prototype into evidence instead of spectacle.
01 · Argument
The proof that mattered was downstream reality, not a prettier contract.
Research OS rejected synthetic completion and demanded a real run. Orbital responded with a non-mock proof: the execute path persisted a real downstream swarm id, status reconciled against the worker, and the completion payload surfaced notebook and artifact outputs without hand-patching Firestore. That moved the lane from architectural promise into operator-usable substrate.
Evidence
A CSV went in. A notebook URL and chart came back.
The acceptance run uploaded market_analysis.csv, dispatched a real preview workload, and eventually surfaced revenue_trajectory.png plus a valid Colab notebook URL. The screenshots matter because they prove artifact ingress was real: the notebook could see the uploaded file inside /content, execute plotting code, and return a named chart rather than a fabricated receipt.
Diagram · Research OS -> Orbital -> Swarm -> Colab -> artifact bundle
The lane now has a legible execution path: Raziel frames and authorizes the question, Research OS opens a lab run, Orbital dispatches into the local Swarm-backed Colab proxy, the notebook performs the analysis, and the completion payload rehydrates notebook and artifact references back into the run record. The case study value is the closed provenance loop, not only the notebook output.
“Preview compute is acceptable when provenance is explicit.”
“The notebook is not the product. The returned artifact bundle is.”
Tension
The lane is real now and insufficient later.
The preview lane is strong enough for operator-approved prototyping, artifact-heavy analysis, and early Concept Genome work. It is not yet the final institutional substrate because notebook ownership can still remain personal and artifact ingress is still lightweight. The case study matters precisely because it names both truths at once.
Implication
Concept Genome can begin in preview mode.
Raziel no longer has to wait for perfect institutional compute before exploring larger questions. Imported OpenAI and Claude histories, code archives, case studies, and demos can now be analyzed in operator-approved notebook runs, then returned to the library as provisional but citeable evidence. That creates a serious research tempo advantage while the governed substrate catches up.