jawnbase

Context for LLMs

jawnbase (GSV-H4·⬡) is the catalog engine underlying the Jawnverse, GSV's cluster of wardrobe, resale, and space-fitting projects. It is explicitly not the data itself but the machinery that produces trustworthy catalog data from messy, multi-source item information, answering one question well: given an item, what is it, and how big is it really. The accumulated output of that machinery is the jawndex index; jawnbase is what computes jawndex's entries. It also serves as jawncloud's catalog backend, and jawnfit consumes jawnbase's resolved dimensions to arrange real objects in real physical spaces. The architectural core is a five-step resolution chain (`resolve.py`) that runs in strict trust order: an instance-level measured override beats a crowd-promoted measured type dimension, which beats a retailer's assembled-product dimension, which beats a shipping-package dimension (explicitly flagged as box size rather than object size), which falls back to a category rough-guess that is never treated as an authoritative source. Every resolved dimension carries its provenance and a confidence score; anything below threshold, or sourced from a guess, is marked `provisional` so downstream consumers know to treat it skeptically rather than trust it silently. The design thesis is that the engine's value lies in adjudicating between sources of varying trustworthiness, not in storage. Around the resolution chain sit two supporting modules: `model/` defines the type vocabulary (ItemType, ItemDimension, ItemInstance), and a second module handles related catalog concerns. jawnbase connects directly to jawndex (which it powers), jawncloud (whose catalog backend it is), and jawnfit (which consumes its resolved measurements). Status: the resolution chain, trust ordering, and provisional-flagging discipline are implemented as the engine's working core; it is infrastructure-layer and consumed by multiple sibling projects rather than a standalone user-facing product.