Research coverage
September 20, 2026 ยท View on GitHub
The initial design review read 109 pages indexed by the official TypeSafe Jev documentation on 19 September 2026. This covered concepts, primitives, patterns, JavaScript and Python SDK references, demos, model notes, and cookbooks. Linked external datasets, media, repositories, and full legal agreements were outside that review. Third-party snapshots are not redistributed; use the original sources below.
Primary documentation
- Introduction and index
- State and context
- Building with System One
- HTTP API
- JavaScript SDK
- Choice, Noul, and Score
- Patterns and intent routing
- Pre-parsed extraction
- Parallel questions
- Model capabilities and Jev 1.13 limitations
Host packaging and hook references appear in the source table in the design. The design review records boundary concerns and their resolutions.
Decisions informed by the review
Use narrow, named context and direct array paths. Deduplicate shared evidence and ask atomic questions about it. Results of independent questions in one request are not inputs to one another; the architecture pass therefore follows local materialization of the intake result.
Choice confidence differs from winning-option probability. Cookbook timing, cost, and accuracy examples do not establish Graphlin performance. Source-level classification does not prove runtime connectivity. The implementation's synthetic live findings record measured outcomes and limitations separately from the design.
Model and platform documentation changes independently of this repository. Recheck the relevant official source before changing wire contracts, supported host versions, model limits, or release packaging. Do not infer zero retention for every TypeSafe account from an enterprise offering.