v0.1.2 — runnable examples, architecture section, version fix

Inference-Pareto-Router 0.1.2

Adds runnable examples and an architecture section, and fixes a version mismatch in the package metadata.

Added

• examples/ — three scripts that run with no API key:
  • 01_route_and_inspect.py — one prompt, full decision breakdown (axis scores, tier, effort, mode, thinking budget, dispatch model).
  • 02_compare_strategies.py — the same prompts routed under all three presets, printed as a comparison table.
  • 03_dispatch_payload.py — builds the provider payload and prints the exact request; --live dispatches it with httpx.
• Architecture section in the README: the module map for core/, evaluator/, gateway/, telemetry/, benchmarks/ and cli.py, plus the one-way pipeline diagram showing the single impure stage.

Fixed

• pareto_router.__version__ was pinned at 0.1.0 while the distribution was already 0.1.1. Both now read 0.1.2.
• README quickstart printed 25488 thinking tokens; the code returns 25464. Corrected to the real value.
• Latency claim qualified: p50 is ~0.1 ms and the 100-scenario benchmark reports mean 0.099 ms / p99 0.181 ms, with a one-off warmup of a few milliseconds on the first call in a process.

Packaging

• sdist now ships examples/.

v0.1.1 — Apache-2.0 licence, packaging metadata and PyPI-grade README

Inference-Pareto-Router 0.1.1

First release with a declared licence, complete packaging metadata and a PyPI-grade README.

Added

• Apache-2.0 LICENSE (Copyright 2026 Sandeep Hipparagi), with an express patent grant. pyproject.toml now declares license, a licence classifier, authors, keywords and project URLs.
• pareto-router console script via [project.scripts] — the CLI shipped in cli.py was previously unreachable from an install.
• README rewritten (12.5k → 13.3k chars): PyPI/Python/licence/CI badges, a prominent installation section, a 15-line quickstart with cost/latency/quality thresholds, the full decision maths (overthinking clamp, effort rescaling, tier rule, token budgets), a strategy threshold table, the 100-scenario benchmark table, the provider routing table, and an explicit implemented/not-implemented matrix.
• [tool.hatch.build.targets.sdist] include list so the sdist carries tests and the benchmark fixture.

Changed

• Version 0.1.0 → 0.1.1.
• Description tightened to what the package actually does: sub-millisecond 2D model-tier and reasoning-effort routing.

Removed

• 23 committed .pyc files that were already covered by .gitignore and would otherwise be packaged into the wheel.

Known limitation

• ParetoRouter.create() still returns a deterministic stub InferenceResponse; the live provider transport is not wired in 0.1.x. Dispatch the ProviderPayload produced by ProviderNormalizer.to_provider_payload() with your own client for real inference. This is documented in the README rather than implied away.