dsh-possibility-space

The dominant pattern in LLM tooling is generate N candidates, let the model pick one, and hope it picked right. dsh-possibility-space replaces that with a single explorable coordinate: it unfolds a finalized DSH answer into an inline two-dimensional semantic map, and the version you get is exactly the one you dragged to.

The plugin infers three independent trade-off axes from the answer itself; you choose two to form a plane. While the pointer moves nothing is requested; the moment you release, one model call produces a single variant at that coordinate. Quote locks are enforced byte-for-byte and rejected variants never enter the version trail. Custom semantic invariants stay visible for human review. Satisfying coordinates can be saved as personal preferences and restored later without any background generation. Pointer movement costs zero tokens, coordinates are quantized to 0.05 to suppress jitter, and the local cache key is deterministic from baseline, axes, and locks.

It is built for anyone who has to trade off across several semantic dimensions: architecture decisions, product strategy, tone and structure of long-form writing, or UI design alternatives. No extra API key is required the plugin reuses the provider and model of the answer you opened. It does not run tools, edit files, or submit code. All answers, variants, locks, and preferences live in browser localStorage; the original session log is never rewritten.