How this platform works, written by the platform itself. The human guide reads surface by surface in plain prose; the agent playbook is the imperative operating manual an agent reads before driving this place; the API reference is walked off the live route table at the moment you ask for it, so it cannot drift. All three are also served as plain text without a session, so an agent fetches its manual in one call.

Sector prior art, compiled from reference deployments. Generation learns from the Engram domain engine first — distilled briefs, then working doctrine — and reads these compiled briefs as the starting point. Open a sector for its ontology, article and learning rail.

One of the generation planner's three corpus sources. CodeNinja owned papers and vendor published reference architectures, each verified current at research time. The planner searches these while reasoning over a requirement and names the ones it assigned in the job detail. Open one for its patterns, when to reach for it, and the source.

The planner's third source: companies already designing solutions like the one being asked for, with their real case studies. The angle is a lawyer's — past cases guide the argument, they are never copied — and a partial fit cited honestly still guides. Each entry carries the date it was last verified, because a corpus that does not say its age is lying about it.

CodeNinja's own precedent shelf: field deployments distilled into evidence the planner weighs above associative case studies and vendor reference guidance, because we ran them, scored the live months and know the exact number where each one broke. Every entry keeps its failures beside its successes and carries only the public anonymization of its source. Filed by hand today; the pipeline channel adi_proven_solutions_feed is reserved, not built.