decision-toolkit 0.4.1__tar.gz → 0.4.2__tar.gz

{decision_toolkit-0.4.1 → decision_toolkit-0.4.2}/CHANGELOG.md

@@ -1,8 +1,15 @@
 # Changelog
 
-All notable changes to this project are documented here. Version `0.4.1` is
+All notable changes to this project are documented here. Version `0.4.2` is
 published on PyPI.
 
+## [0.4.2] — 2026-10-21
+
+Documentation cleanup: removed internal project language (`m0`, phase numbers,
+waitlist), removed the public catalog proposal and roadmap pages, and rewrote
+the model-card index for users. Every model card now includes library, CLI and
+REST snippets for every variant. No runtime changes.
+
 ## [0.4.1] — 2026-10-21
 
 Catalog presentation patch: every variant now carries its REST alias
@@ -216,6 +223,7 @@ README correction. No code or model behavior changed.
 - Typed `DecisionModel` API with lifecycle, batching, async, usage, warnings,
   normalized confidence, capabilities, structured events and atomic model store.
 
+[0.4.2]: https://pypi.org/project/decision-toolkit/0.4.2/
 [0.4.1]: https://pypi.org/project/decision-toolkit/0.4.1/
 [0.4.0]: https://pypi.org/project/decision-toolkit/0.4.0/
 [0.3.1]: https://pypi.org/project/decision-toolkit/0.3.1/

{decision_toolkit-0.4.1 → decision_toolkit-0.4.2}/CONTRIBUTING.md

@@ -76,9 +76,11 @@ uv run dtk --version
 uv run pip-audit                # dependency scan (CI runs this)
 ```
 
-CI runs these on Python 3.12 and 3.13, plus `pip-audit`. The tag-triggered
-release workflow builds and publishes the wheel with uv; see
-[RELEASE.md](docs/RELEASE.md). For rate limits, metrics, tracing and the load
+CI runs these on Python 3.12 and 3.13, plus `pip-audit`. `tests/test_help_examples.py`
+parses every example printed by `dtk --help` and checks referenced catalog
+models/variants; run the full examples against a populated cache before a
+release. The tag-triggered release workflow builds and publishes the wheel with
+uv; see [RELEASE.md](docs/RELEASE.md). For rate limits, metrics, tracing and the load
 probe see [OPERATIONS.md](docs/OPERATIONS.md).
 
 ## Commits

{decision_toolkit-0.4.1 → decision_toolkit-0.4.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: decision-toolkit
-Version: 0.4.1
+Version: 0.4.2
 Summary: Local-first toolkit for open-weights decision models: one ONNX engine, one adapter per model contract, typed choice/yes-no/score decisions with probabilities on CPU.
 Project-URL: Homepage, https://github.com/gia-uh/decision-toolkit
 Project-URL: Repository, https://github.com/gia-uh/decision-toolkit
@@ -59,7 +59,7 @@ and CLI. `dtk --help` and every subcommand help also include runnable examples.
 ## Installation
 
 ```bash
-uv pip install decision-toolkit              # library + CLI (0.4.1 on PyPI)
+uv pip install decision-toolkit              # library + CLI (0.4.2 on PyPI)
 uv pip install "decision-toolkit[onnx]"      # + onnxruntime, tokenizers (the model runtime)
 uv pip install "decision-toolkit[server]"    # + REST server (FastAPI)
 ```
@@ -67,7 +67,7 @@ uv pip install "decision-toolkit[server]"    # + REST server (FastAPI)
 The project is managed with [uv](https://docs.astral.sh/uv/) and ships a committed `uv.lock`. Published
 wheels remain compatible with `pip` for environments that do not use uv.
 
-PyPI: <https://pypi.org/project/decision-toolkit/0.4.1/>. See
+PyPI: <https://pypi.org/project/decision-toolkit/0.4.2/>. See
 [docs/RELEASE.md](https://github.com/gia-uh/decision-toolkit/blob/main/docs/RELEASE.md)
 for the Trusted Publishing flow and clean-install checks.
 
@@ -135,7 +135,7 @@ decision.truncated, decision.warnings
 validated before inference (`box`/assets are rejected by text-only models).
 Custom adapters can be registered with `register_adapter`; local ONNX models
 can be opened with `DecisionModel.from_onnx(...)`. See
-[docs/CORE.md](https://github.com/gia-uh/decision-toolkit/blob/main/docs/CORE.md) for the full phase 1 contract, error taxonomy,
+[docs/CORE.md](https://github.com/gia-uh/decision-toolkit/blob/main/docs/CORE.md) for the full API contract, error taxonomy,
 logging events and the expressiveness inventory.
 
 **2. As a CLI**
@@ -186,38 +186,26 @@ current text-only subset.
 See [docs/REST.md](https://github.com/gia-uh/decision-toolkit/blob/main/docs/REST.md) for the pinned contracts, limits, auth and
 the standard-evolution process.
 
-## The catalog is curated
+## Models and variants
 
-The catalog is **ours**, not a mirror of HuggingFace. A model enters only when it
-is understood and verified:
+The catalog lists models that are runnable with the toolkit today. Every entry
+has a real ONNX export, verified files and measured behavior; nothing is listed
+as a placeholder.
 
-1. clean commercial license;
-2. an adapter whose contract we can implement **from a real export**;
-3. a real ONNX export on the Hub;
-4. measured parity against the original on a fixed sample;
-5. CPU-viable latency.
-
-Every variant points to files that exist. No placeholders, no speculative
-adapters (there is no generic "any ONNX" adapter).
-
-| id | adapter | status | variants (default *) |
+| id | adapter | verification | variants (default *) |
 | --- | --- | --- | --- |
-| `gliner2.5-multi` | `gliner2` | **m0** — verified end-to-end (9/15, 353 ms) | `fp16`* (577 MB); `fp32` (1132 MB) |
-| `gliner2.5-decide` | `gliner2-decide` | **m0** — fused export verified (refund 0.944, 912 ms; English-only) | `q4f16`* (524 MB) |
-| `laya` | `laya` | **m0** — verified against torch (Δp ≤ 0.005) | `q8`* (528 MB) |
-| `julia-1` | `julia` | **m0** — verified: 100/100 argmax parity, 329 ms | `fp32`* (618 MB) |
-| `opendecider-nano` | `opendecider` | **m0** — official build; prompt ported from pinned code (530 ms) | `q8f16`* (469 MB); `fp16` (792 MB) |
-| `laya-multilingual` | `laya` | **m0** — fused ONNX; upstream parity (37 ms) | `fp16`* (682 MB); `fp32` (1326 MB) |
-| `open-jev-deberta-v3-large` | `open-jev` | **m0** — Open-Jev DeBERTa v3 large repro; 0.854 in-domain (241 ms) | `fp16`* (884 MB); `fp32` (1758 MB); `q4` (487 MB); `q4f16` (357 MB) |
-| `strands-decider-2b` | `strands-decider` | **m0** — reference fixture match (719 ms) | `q4f16`* (1114 MB); `q8` (1821 MB) |
+| `gliner2.5-multi` | `gliner2` | verified end-to-end (9/15, 353 ms) | `fp16`* (577 MB); `fp32` (1132 MB) |
+| `gliner2.5-decide` | `gliner2-decide` | fused export verified (refund 0.944, 912 ms; English-only) | `q4f16`* (524 MB) |
+| `laya` | `laya` | verified against torch (Δp ≤ 0.005) | `q8`* (528 MB) |
+| `julia-1` | `julia` | verified: 100/100 argmax parity, 329 ms | `fp32`* (618 MB) |
+| `opendecider-nano` | `opendecider` | official build; prompt ported from pinned code (530 ms) | `q8f16`* (469 MB); `fp16` (792 MB) |
+| `laya-multilingual` | `laya` | fused ONNX; upstream parity (37 ms) | `fp16`* (682 MB); `fp32` (1326 MB) |
+| `open-jev-deberta-v3-large` | `open-jev` | Open-Jev DeBERTa v3 large reproduction; 0.854 in-domain (241 ms) | `fp16`* (884 MB); `fp32` (1758 MB); `q4` (487 MB); `q4f16` (357 MB) |
+| `strands-decider-2b` | `strands-decider` | reference fixture match (719 ms) | `q4f16`* (1114 MB); `q8` (1821 MB) |
 
 Each variant has a REST alias in the form `id@variant`; `dtk models list` prints
 the alias next to every variant.
 
-**Only `m0` models are in the catalog.** Researched-but-unavailable models
-(planned, conversion pending, broken exports) are listed in
-[docs/ROADMAP.md](https://github.com/gia-uh/decision-toolkit/blob/main/docs/ROADMAP.md) and are not downloadable through the toolkit.
-
 Per-model cards with the contract, measured evidence and known limitations:
 [docs/models](https://github.com/gia-uh/decision-toolkit/tree/main/docs/models).
 
@@ -257,7 +245,7 @@ standard `model` field. The base id uses the server/persisted default. There is
 no `variant` field in the request body; `/v1/models` lists both the base model
 and every alias, and two aliases can be loaded concurrently.
 
-A variant becomes default only after passing the parity gate (top-1 ≥ 99%,
+A variant becomes default only after passing the verification criteria (top-1 ≥ 99%,
 mean |Δp| ≤ 0.02, accuracy delta ≤ 1 pp against the fp32 build on a fixed
 sample) and measuring latency on CPU. Unverified conversions are not catalogued
 and never become a silent fallback. See [Variants and quantization](https://gia-uh.github.io/decision-toolkit/VARIANTS/)
@@ -303,7 +291,7 @@ See [CHANGELOG.md](https://github.com/gia-uh/decision-toolkit/blob/main/CHANGELO
 
 ## Custom ONNX (at your own risk)
 
-You can run any ONNX file with an adapter you trust — outside the curated
+You can run any ONNX file with an adapter you trust — outside the
 catalog:
 
 ```python

{decision_toolkit-0.4.1 → decision_toolkit-0.4.2}/README.md

@@ -21,7 +21,7 @@ and CLI. `dtk --help` and every subcommand help also include runnable examples.
 ## Installation
 
 ```bash
-uv pip install decision-toolkit              # library + CLI (0.4.1 on PyPI)
+uv pip install decision-toolkit              # library + CLI (0.4.2 on PyPI)
 uv pip install "decision-toolkit[onnx]"      # + onnxruntime, tokenizers (the model runtime)
 uv pip install "decision-toolkit[server]"    # + REST server (FastAPI)
 ```
@@ -29,7 +29,7 @@ uv pip install "decision-toolkit[server]"    # + REST server (FastAPI)
 The project is managed with [uv](https://docs.astral.sh/uv/) and ships a committed `uv.lock`. Published
 wheels remain compatible with `pip` for environments that do not use uv.
 
-PyPI: <https://pypi.org/project/decision-toolkit/0.4.1/>. See
+PyPI: <https://pypi.org/project/decision-toolkit/0.4.2/>. See
 [docs/RELEASE.md](https://github.com/gia-uh/decision-toolkit/blob/main/docs/RELEASE.md)
 for the Trusted Publishing flow and clean-install checks.
 
@@ -97,7 +97,7 @@ decision.truncated, decision.warnings
 validated before inference (`box`/assets are rejected by text-only models).
 Custom adapters can be registered with `register_adapter`; local ONNX models
 can be opened with `DecisionModel.from_onnx(...)`. See
-[docs/CORE.md](https://github.com/gia-uh/decision-toolkit/blob/main/docs/CORE.md) for the full phase 1 contract, error taxonomy,
+[docs/CORE.md](https://github.com/gia-uh/decision-toolkit/blob/main/docs/CORE.md) for the full API contract, error taxonomy,
 logging events and the expressiveness inventory.
 
 **2. As a CLI**
@@ -148,38 +148,26 @@ current text-only subset.
 See [docs/REST.md](https://github.com/gia-uh/decision-toolkit/blob/main/docs/REST.md) for the pinned contracts, limits, auth and
 the standard-evolution process.
 
-## The catalog is curated
+## Models and variants
 
-The catalog is **ours**, not a mirror of HuggingFace. A model enters only when it
-is understood and verified:
+The catalog lists models that are runnable with the toolkit today. Every entry
+has a real ONNX export, verified files and measured behavior; nothing is listed
+as a placeholder.
 
-1. clean commercial license;
-2. an adapter whose contract we can implement **from a real export**;
-3. a real ONNX export on the Hub;
-4. measured parity against the original on a fixed sample;
-5. CPU-viable latency.
-
-Every variant points to files that exist. No placeholders, no speculative
-adapters (there is no generic "any ONNX" adapter).
-
-| id | adapter | status | variants (default *) |
+| id | adapter | verification | variants (default *) |
 | --- | --- | --- | --- |
-| `gliner2.5-multi` | `gliner2` | **m0** — verified end-to-end (9/15, 353 ms) | `fp16`* (577 MB); `fp32` (1132 MB) |
-| `gliner2.5-decide` | `gliner2-decide` | **m0** — fused export verified (refund 0.944, 912 ms; English-only) | `q4f16`* (524 MB) |
-| `laya` | `laya` | **m0** — verified against torch (Δp ≤ 0.005) | `q8`* (528 MB) |
-| `julia-1` | `julia` | **m0** — verified: 100/100 argmax parity, 329 ms | `fp32`* (618 MB) |
-| `opendecider-nano` | `opendecider` | **m0** — official build; prompt ported from pinned code (530 ms) | `q8f16`* (469 MB); `fp16` (792 MB) |
-| `laya-multilingual` | `laya` | **m0** — fused ONNX; upstream parity (37 ms) | `fp16`* (682 MB); `fp32` (1326 MB) |
-| `open-jev-deberta-v3-large` | `open-jev` | **m0** — Open-Jev DeBERTa v3 large repro; 0.854 in-domain (241 ms) | `fp16`* (884 MB); `fp32` (1758 MB); `q4` (487 MB); `q4f16` (357 MB) |
-| `strands-decider-2b` | `strands-decider` | **m0** — reference fixture match (719 ms) | `q4f16`* (1114 MB); `q8` (1821 MB) |
+| `gliner2.5-multi` | `gliner2` | verified end-to-end (9/15, 353 ms) | `fp16`* (577 MB); `fp32` (1132 MB) |
+| `gliner2.5-decide` | `gliner2-decide` | fused export verified (refund 0.944, 912 ms; English-only) | `q4f16`* (524 MB) |
+| `laya` | `laya` | verified against torch (Δp ≤ 0.005) | `q8`* (528 MB) |
+| `julia-1` | `julia` | verified: 100/100 argmax parity, 329 ms | `fp32`* (618 MB) |
+| `opendecider-nano` | `opendecider` | official build; prompt ported from pinned code (530 ms) | `q8f16`* (469 MB); `fp16` (792 MB) |
+| `laya-multilingual` | `laya` | fused ONNX; upstream parity (37 ms) | `fp16`* (682 MB); `fp32` (1326 MB) |
+| `open-jev-deberta-v3-large` | `open-jev` | Open-Jev DeBERTa v3 large reproduction; 0.854 in-domain (241 ms) | `fp16`* (884 MB); `fp32` (1758 MB); `q4` (487 MB); `q4f16` (357 MB) |
+| `strands-decider-2b` | `strands-decider` | reference fixture match (719 ms) | `q4f16`* (1114 MB); `q8` (1821 MB) |
 
 Each variant has a REST alias in the form `id@variant`; `dtk models list` prints
 the alias next to every variant.
 
-**Only `m0` models are in the catalog.** Researched-but-unavailable models
-(planned, conversion pending, broken exports) are listed in
-[docs/ROADMAP.md](https://github.com/gia-uh/decision-toolkit/blob/main/docs/ROADMAP.md) and are not downloadable through the toolkit.
-
 Per-model cards with the contract, measured evidence and known limitations:
 [docs/models](https://github.com/gia-uh/decision-toolkit/tree/main/docs/models).
 
@@ -219,7 +207,7 @@ standard `model` field. The base id uses the server/persisted default. There is
 no `variant` field in the request body; `/v1/models` lists both the base model
 and every alias, and two aliases can be loaded concurrently.
 
-A variant becomes default only after passing the parity gate (top-1 ≥ 99%,
+A variant becomes default only after passing the verification criteria (top-1 ≥ 99%,
 mean |Δp| ≤ 0.02, accuracy delta ≤ 1 pp against the fp32 build on a fixed
 sample) and measuring latency on CPU. Unverified conversions are not catalogued
 and never become a silent fallback. See [Variants and quantization](https://gia-uh.github.io/decision-toolkit/VARIANTS/)
@@ -265,7 +253,7 @@ See [CHANGELOG.md](https://github.com/gia-uh/decision-toolkit/blob/main/CHANGELO
 
 ## Custom ONNX (at your own risk)
 
-You can run any ONNX file with an adapter you trust — outside the curated
+You can run any ONNX file with an adapter you trust — outside the
 catalog:
 
 ```python

{decision_toolkit-0.4.1 → decision_toolkit-0.4.2}/docs/BENCH.md

@@ -66,7 +66,7 @@ and
 The model cards retain the historical cold-start measurements used when each
 variant was admitted. The harness reports **post-warmup steady-state** latency,
 which is lower. Compare like with like: use the harness when comparing commits
-or machines; use the card values only as historical admission evidence.
+or machines; use the card values only as historical reference measurements.
 
 ## Adding a case
 

{decision_toolkit-0.4.1 → decision_toolkit-0.4.2}/docs/CLI.md

@@ -1,4 +1,4 @@
-# CLI (phase 3)
+# CLI
 
 `dtk` is our own operational surface. Unlike REST, it has **no external
 standard** to comply with: it is consistent with itself, script-friendly, and a
@@ -9,7 +9,7 @@ CLI.
 
 | Command | Purpose |
 | --- | --- |
-| `dtk models` | Inspect the local cache and the curated catalog. |
+| `dtk models` | Inspect the local cache and the model catalog. |
 | `dtk models list` | Every catalog variant, default (`*`) and cached variants (`--license`, `--quant`, `--adapter`). |
 | `dtk models variants <id>` | Every real variant of one model, with size, role and exact install command. |
 | `dtk models download <id>` | Download the default or `--variant` quantization (`--force`, `--dry-run`). |
@@ -244,8 +244,7 @@ dtk serve --model gliner2.5-multi --warmup \
 - `--workers N` and `--reload` use an internal uvicorn factory; `--reload`
   is for development and cannot be combined with `--workers > 1`.
 - `--max-models`, `--max-body-mb`, `--timeout`, `--warmup`, `--api-key`,
-  `--cors-origin`, `--log-level` and `--log-format` are the server's phase 2
-  controls.
+  `--cors-origin`, `--log-level` and `--log-format` are server controls.
 - Custom `--adapter`/`--files` also apply to the server's default model.
 
 ## Completion
@@ -281,6 +280,6 @@ dtk doctor --json
   config), but still does not reimplement inference.
 - It does not have a compatibility obligation to System One/OpenAI; those are
   REST contracts.
-- Metrics/rate limiting are server-side phase 5 features; see
+- Metrics and rate limiting are server-side features; see
   [OPERATIONS.md](OPERATIONS.md). Wheel/PyPI releases are documented in
   [RELEASE.md](RELEASE.md).

{decision_toolkit-0.4.1 → decision_toolkit-0.4.2}/docs/CORE.md

@@ -1,4 +1,4 @@
-# Core programmatic API (phase 1)
+# Core API
 
 The programmatic API is the **superset**: the most expressive layer of the
 toolkit. The REST server is a documented, conformance-tested projection of a subset of
@@ -207,7 +207,7 @@ Events emitted by `decision_toolkit.*` loggers:
 Each event is a compact JSON object with `event`, `request_id`, model identity,
 variant, adapter, question count, token counts, latency, truncation and error
 code. State and question text are **not** logged by default. The CLI/REST
-layers configure formatting and destinations in later phases.
+layers configure formatting and destinations separately.
 
 ## Extensibility
 
@@ -254,9 +254,9 @@ Registration is in-process (no daemon, no plugin server).
 - `resolve(id, variant=...)`: local directory + role → file map + adapter for
   the core.
 
-## Phase 1 expressiveness inventory
+## Capabilities and limits
 
-This is the checklist against which the REST projection will be verified:
+The core exposes the following capabilities:
 
 | Capability | Core status |
 | --- | --- |
@@ -280,13 +280,12 @@ This is the checklist against which the REST projection will be verified:
 | Model store | atomic download, hashes, verify, import metadata |
 | Multi-label | capability flag + validation; no adapter yet |
 
-## Out of scope for phase 1
+## Not part of the core
 
-- REST contract projection and standard compliance — implemented in phase 2
-  (see [REST.md](REST.md)).
-- CLI workflows across the whole catalog — implemented in phase 3 (see
-  [CLI.md](CLI.md)).
-- Pin/hash enforcement and source CI are implemented in phase 4; wheel,
-  PyPI and distribution tagging are documented in [RELEASE.md](RELEASE.md).
-- Metrics server, rate limiting and load/fuzz testing are server-side phase 5
-  features; see [OPERATIONS.md](OPERATIONS.md).
+- REST contract projection and standard compliance are implemented in the
+  server; see [REST.md](REST.md).
+- CLI workflows across the catalog are implemented in [CLI.md](CLI.md).
+- Catalog pinning and source CI are implementation details; packaging is
+  documented in [RELEASE.md](RELEASE.md).
+- Metrics, rate limiting and load/fuzz testing are server-side features; see
+  [OPERATIONS.md](OPERATIONS.md).

{decision_toolkit-0.4.1 → decision_toolkit-0.4.2}/docs/OPERATIONS.md

@@ -1,6 +1,6 @@
-# Operations (phase 5)
+# Operations
 
-This document covers the operational behavior added in phase 5: rate limits,
+This document covers operational behavior: rate limits,
 backpressure, metrics, tracing, load testing and the PII contract. Packaging is
 still a later milestone.
 

{decision_toolkit-0.4.1 → decision_toolkit-0.4.2}/docs/RELEASE.md

@@ -39,7 +39,7 @@ Clean install test (no `PYTHONPATH`):
 ```bash
 uv venv --python 3.12 /tmp/dtk-wheel
 uv pip install --python /tmp/dtk-wheel/bin/python \
-  dist/decision_toolkit-0.4.1-py3-none-any.whl
+  dist/decision_toolkit-0.4.2-py3-none-any.whl
 uv pip check --python /tmp/dtk-wheel/bin/python
 /tmp/dtk-wheel/bin/dtk --version
 /tmp/dtk-wheel/bin/dtk models list --json
@@ -50,7 +50,7 @@ With inference and server extras:
 
 ```bash
 uv pip install --python /tmp/dtk-wheel/bin/python \
-  "dist/decision_toolkit-0.4.1-py3-none-any.whl[onnx,server,tracing]"
+  "dist/decision_toolkit-0.4.2-py3-none-any.whl[onnx,server,tracing]"
 /tmp/dtk-wheel/bin/dtk decide --model gliner2.5-decide ...
 /tmp/dtk-wheel/bin/dtk serve --model gliner2.5-decide --warmup --port 8016
 curl -s http://127.0.0.1:8016/health
@@ -84,8 +84,8 @@ No long-lived PyPI token is needed; the workflow uses GitHub OIDC
 4. Create and push an annotated tag:
 
    ```bash
-   git tag -a v0.4.1 -m "decision-toolkit 0.4.1"
-   git push origin v0.4.1
+   git tag -a v0.4.2 -m "decision-toolkit 0.4.2"
+   git push origin v0.4.2
    ```
 
 5. `.github/workflows/release.yml` builds once, publishes to **TestPyPI**, then
@@ -134,9 +134,9 @@ to TestPyPI without publishing to PyPI or creating a GitHub release.
 
 ## Published release
 
-`0.4.1` is published through Trusted Publishing on 2026-10-21:
+`0.4.2` is published through Trusted Publishing on 2026-10-21:
 
-<https://pypi.org/project/decision-toolkit/0.4.1/>
+<https://pypi.org/project/decision-toolkit/0.4.2/>
 
 `0.1.0` was the first release; `0.1.1` refreshed the PyPI description; `0.1.2`
 validated the OIDC TestPyPI → PyPI flow; `0.1.3` added post-publish install
@@ -148,7 +148,8 @@ selection across library, CLI and REST; `0.2.0` corrected the REST conformance
 claims and wire shapes; `0.2.1` changed the public documentation URLs; `0.3.0`
 added the three new ONNX decision models; `0.3.1` made the catalog list one row per variant; `0.4.0` moved REST variant
 selection to standard model aliases; `0.4.1` exposed aliases in the catalog and
-per-variant loading examples in every model card.
+per-variant loading examples in every model card; `0.4.2` removed internal
+project language and public proposal/roadmap pages.
 
 The tag-triggered workflow remains the preferred path for future releases once
 the TestPyPI/PyPI trusted publishers and protected environments are configured.

{decision_toolkit-0.4.1 → decision_toolkit-0.4.2}/docs/REST.md

@@ -1,4 +1,4 @@
-# REST projection (phase 2)
+# REST API
 
 REST is a **documented projection** of the programmatic core, not a second
 implementation. The core remains the superset; the server selects a supported
@@ -337,11 +337,11 @@ at `http://localhost:8000/v1`.
 - Transport tests cover auth, request ids, 413, CORS, readiness and per-request
   model selection.
 
-## Out of scope for phase 2
+## Not implemented by this server
 
-- Metrics/Prometheus, rate limiting and load tests are implemented in phase 5
-  (see [OPERATIONS.md](OPERATIONS.md)).
-- CLI batch/JSONL and custom-model workflows are implemented in phase 3
-  (see [CLI.md](CLI.md)).
-- Pinned catalog hashes and source CI are implemented in phase 4; wheel,
-  PyPI and distribution tagging are documented in [RELEASE.md](RELEASE.md).
+- Metrics, rate limiting and load testing are implemented as server-side
+  features; see [OPERATIONS.md](OPERATIONS.md).
+- CLI batch/JSONL and custom-model workflows are in the CLI; see [CLI.md](CLI.md).
+- Pinned catalog hashes and CI are implementation details; packaging is
+  documented in [RELEASE.md](RELEASE.md).
+- Images/assets and `box` are rejected because the current catalog is text-only.

{decision_toolkit-0.4.1 → decision_toolkit-0.4.2}/docs/VARIANTS.md

@@ -11,7 +11,7 @@ local model cache. Variants of the same model coexist, so you can compare them
 without deleting the default.
 
 Only catalogued and verified builds are selectable. Community conversions that
-have not passed the catalog gates are not exposed as variants.
+have not passed the catalog checks are not exposed as variants.
 
 ## Available variants
 
@@ -38,7 +38,7 @@ other variant was selected with `dtk models use`.
 
 Models with a single row have only one catalogued, verified build. That does
 not mean other conversions do not exist upstream; it means they have not passed
-the parity, hash and license gates for this catalog.
+the parity, hash and license checks for this catalog.
 
 ## One selection model across surfaces
 
@@ -202,7 +202,7 @@ The variant currently in use cannot be deleted.
 
 ## Why some models have only one variant
 
-The catalog is curated, not a mirror. A variant is admitted only with a pinned
+The catalog is verified, not a mirror. A variant is admitted only with a pinned
 revision, a real ONNX export, SHA256 per file, a clean license and measured
 parity/latency. An upstream repository may contain fp32, fp16, int8 or q4
 builds that are not in the toolkit because they have not been validated.

{decision_toolkit-0.4.1 → decision_toolkit-0.4.2}/docs/index.md

@@ -23,8 +23,7 @@ PyPI: <https://pypi.org/project/decision-toolkit/>
 - [Variants and quantization](VARIANTS.md) — every catalogued build and how to download/select it.
 - [Operations](OPERATIONS.md) — rate limits, metrics, tracing and runbook.
 - [Benchmarks](BENCH.md) — reproducible latency/parity/calibration.
-- [Models](models/README.md) — the curated catalog.
-- [Catalog proposal](CATALOG_PROPOSAL.md) — analysis and evidence for the latest model additions.
+- [Models](models/README.md) — the model catalog.
 - [Release](RELEASE.md) — build, TestPyPI/PyPI and Trusted Publishing.
 
 ## One example

{decision_toolkit-0.4.1 → decision_toolkit-0.4.2}/docs/models/README.md

@@ -1,8 +1,8 @@
 # Model cards
 
-Verified `m0` models. Each card documents the adapter contract, the real download,
-the measured evidence and the known limitations. Models that are not `m0` have no
-card on purpose: they are not part of the curated library yet.
+These are the models available through `decision-toolkit`. Each card explains
+how to load and use the model, its variants, the measured evidence and its
+known limitations.
 
 | Model | Adapter | Variants (default *) | Pinned | Latency (CPU) | Evidence |
 |---|---|---|---|---|---|
@@ -18,16 +18,5 @@ card on purpose: they are not part of the curated library yet.
 A `*` marks the default. To inspect sizes, hashes and exact commands, run
 `dtk models variants <id>` or read [Variants and quantization](../VARIANTS.md).
 
-Latencies above are the historical admission measurements. Reproducible
-post-warmup measurements and the exact command are in
-[BENCH.md](../BENCH.md).
-
-## What a card must contain
-
-1. Status and adapter (the model contract).
-2. Every catalogued variant: repo, files, quantization, download size and role.
-3. Measured evidence: latency on CPU, parity/accuracy against a named baseline.
-4. Known limitations (language, export layout, quantization, benchmark scope).
-5. A runnable usage example.
-6. Provenance of the export.
-7. The pinned revision and SHA256 per file, plus the verify/bench command.
+Latencies above are CPU measurements. Reproducible post-warmup measurements
+and the exact command are in [BENCH.md](../BENCH.md).

{decision_toolkit-0.4.1 → decision_toolkit-0.4.2}/docs/models/gliner2.5-decide.md

@@ -1,6 +1,5 @@
 # GLiNER2.5-Decide
 
-- **Status:** `m0` (verified end-to-end)
 - **Adapter:** `gliner2-decide` — fused graph with `[L]` label markers and `[SEP_STRUCT]`
 - **License:** Apache-2.0
 - **Parameters:** 340M · **Architecture:** DeBERTa-v3-large · **Language:** English only

{decision_toolkit-0.4.1 → decision_toolkit-0.4.2}/docs/models/gliner2.5-multi.md

@@ -1,6 +1,5 @@
 # GLiNER2.5 multi (v1 + classifier)
 
-- **Status:** `m0` (verified end-to-end)
 - **Adapter:** `gliner2` — split encoder + classifier, schema with `[P]` / `[C]` / `[DESCRIPTION]` / `[SEP_TEXT]`
 - **License:** Apache-2.0
 - **Parameters:** 287M · **Architecture:** GLiNER2.5 multi-v1 · **Languages:** multilingual

{decision_toolkit-0.4.1 → decision_toolkit-0.4.2}/docs/models/julia-1.md

@@ -1,6 +1,5 @@
 # Julia 1
 
-- **Status:** `m0` (official ONNX verified against published PyTorch logits)
 - **Adapter:** `julia` — single graph with `marker_pos` / `qtype`
 - **License:** Apache-2.0
 - **Parameters:** 141M · **Architecture:** mmBERT-small · **Languages:** 52 locales (MASSIVE)

{decision_toolkit-0.4.1 → decision_toolkit-0.4.2}/docs/models/laya-multilingual.md

@@ -1,6 +1,5 @@
 # Laya multilingual
 
-- **Status:** `m0` (fused ONNX export verified end-to-end)
 - **Adapter:** `laya` — fused encoder + decision head, `marker_pos` / `qtype`, calibrated temperatures
 - **License:** Apache-2.0
 - **Parameters:** 322M · **Architecture:** mmBERT-base · **Languages:** 100+ languages

{decision_toolkit-0.4.1 → decision_toolkit-0.4.2}/docs/models/laya.md

@@ -1,6 +1,5 @@
 # Laya (English)
 
-- **Status:** `m0` (verified against the torch runtime)
 - **Adapter:** `laya` — split encoder + head with `marker_pos` / `qtype`
 - **License:** Apache-2.0
 - **Parameters:** 421M · **Architecture:** ModernBERT-large · **Language:** English

{decision_toolkit-0.4.1 → decision_toolkit-0.4.2}/docs/models/open-jev-deberta-v3-large.md

@@ -1,6 +1,5 @@
 # Open-Jev DeBERTa v3 large
 
-- **Status:** `m0` (ONNX export verified end-to-end)
 - **Adapter:** `open-jev` — one sequence, span-pooled option scoring
 - **License:** Apache-2.0
 - **Parameters:** ~435M · **Architecture:** DeBERTa-v3-large + span head

{decision_toolkit-0.4.1 → decision_toolkit-0.4.2}/docs/models/opendecider-nano.md

@@ -1,6 +1,5 @@
 # OpenDecider nano
 
-- **Status:** `m0` (official build; prompt/readout ported from pinned code)
 - **Adapter:** `opendecider` — single graph, softmax over `[MASK]` marker logits
 - **License:** Apache-2.0
 - **Parameters:** 395M · **Architecture:** Ettin encoder 400M

{decision_toolkit-0.4.1 → decision_toolkit-0.4.2}/docs/models/strands-decider-2b.md

@@ -1,6 +1,5 @@
 # Strands Decider 2B
 
-- **Status:** `m0` (ONNX export verified end-to-end)
 - **Adapter:** `strands-decider` — Qwen3.5-2B torso + pointer head
 - **License:** Apache-2.0
 - **Parameters:** 2B · **Architecture:** Qwen3.5-2B-Base + LoRA + pointer head

{decision_toolkit-0.4.1 → decision_toolkit-0.4.2}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "decision-toolkit"
-version = "0.4.1"
+version = "0.4.2"
 description = "Local-first toolkit for open-weights decision models: one ONNX engine, one adapter per model contract, typed choice/yes-no/score decisions with probabilities on CPU."
 readme = "README.md"
 requires-python = ">=3.12"

{decision_toolkit-0.4.1 → decision_toolkit-0.4.2}/src/decision_toolkit/__init__.py

@@ -7,7 +7,7 @@ questions with a calibrated probability for every option, on CPU.
 
 from __future__ import annotations
 
-__version__ = "0.4.1"
+__version__ = "0.4.2"
 
 from .adapters import register_adapter
 from .calibration import brier_score, expected_calibration_error, threshold_sweep

{decision_toolkit-0.4.1 → decision_toolkit-0.4.2}/src/decision_toolkit/cli.py

@@ -135,7 +135,7 @@ def build_parser() -> argparse.ArgumentParser:
         ),
         "dtk models variants gliner2.5-multi",
         "dtk models download gliner2.5-multi --dry-run",
-        "dtk models download gliner2.5-multi --variant fp32 --force",
+        "dtk models download gliner2.5-multi --variant fp32 --dry-run",
     )
     models_download.add_argument("model", help="Catalog id.")
     models_download.add_argument(

decision_toolkit-0.4.2/tests/test_help_examples.py

@@ -0,0 +1,64 @@
+"""Validate every runnable example printed by ``dtk --help``.
+
+The full execution audit (real ONNX models and temporary files) is run before a
+release with a populated cache. This test keeps every example syntactically
+valid and ensures every referenced catalog model/variant still exists; it runs
+in CI without downloading models.
+"""
+
+from __future__ import annotations
+
+import argparse
+import shlex
+
+from decision_toolkit import catalog, cli
+
+
+def _walk(parser: argparse.ArgumentParser):
+    if parser.epilog and "examples:" in parser.epilog:
+        for line in parser.epilog.splitlines()[1:]:
+            if line.strip():
+                yield line.strip()
+    for action in parser._actions:
+        if isinstance(action, argparse._SubParsersAction):
+            for subparser in action.choices.values():
+                yield from _walk(subparser)
+
+
+def _examples() -> list[str]:
+    examples = list(dict.fromkeys(_walk(cli.build_parser())))
+    assert examples
+    return examples
+
+
+def _validate_model_variant(model_id: str | None, variant: str | None) -> None:
+    if model_id is None or "/" in model_id or model_id.startswith("."):
+        return
+    model = catalog.get(model_id)
+    if model is None:
+        return
+    if variant is not None:
+        catalog.select_variant(model, variant)
+
+
+def test_every_help_example_parses_and_references_catalog_entries():
+    parser = cli.build_parser()
+    for example in _examples():
+        argv = shlex.split(example)
+        assert argv[0] == "dtk", example
+        args = parser.parse_args(argv[1:])
+        command = getattr(args, "command", None)
+        if command == "models":
+            model_id = getattr(args, "model", None)
+            variant = getattr(args, "variant", None)
+            if args.models_command == "variants":
+                model_id = args.model
+                variant = None
+            elif args.models_command in {"download", "use", "delete"}:
+                model_id = args.model
+            _validate_model_variant(model_id, variant)
+        elif command in {"decide", "classify", "serve"}:
+            _validate_model_variant(getattr(args, "model", None), getattr(args, "variant", None))
+        elif command == "config":
+            if getattr(args, "config_command", None) == "set" and args.key == "current":
+                _validate_model_variant(args.value, None)

{decision_toolkit-0.4.1 → decision_toolkit-0.4.2}/uv.lock

@@ -349,7 +349,7 @@ wheels = [
 
 [[package]]
 name = "decision-toolkit"
-version = "0.4.1"
+version = "0.4.2"
 source = { editable = "." }
 dependencies = [
     { name = "huggingface-hub" },

decision_toolkit-0.4.1/docs/CATALOG_PROPOSAL.md

@@ -1,122 +0,0 @@
-# Catalog proposal: three ONNX decision models
-
-**Date:** 2026-10-21  
-**Status:** published in `0.3.0`  
-**Scope:** add Laya multilingual, Open-Jev DeBERTa v3 large and Strands Decider 2B  
-**Release:** <https://pypi.org/project/decision-toolkit/0.3.0/>
-
-## Executive summary
-
-Three public ONNX decision models meet the catalog gates for this iteration:
-
-| Model | License | Adapter | Variants | Default |
-|---|---|---|---|---:|
-| Laya multilingual | Apache-2.0 | existing `laya` + fused branch | fp16, fp32 | fp16 |
-| Open-Jev DeBERTa v3 large | Apache-2.0 | new `open-jev` | fp16, fp32, q4, q4f16 | fp16 |
-| Strands Decider 2B | Apache-2.0 | new `strands-decider` | q4f16, q8 | q4f16 |
-
-They have a real public ONNX export, complete external-data files, a pinned
-revision and a known prompt contract. The local machine verified every file
-hash and opened every ONNX session. No speculative or unavailable model was
-added.
-
-## Why these three
-
-- **Laya multilingual** closes the multilingual gap in the existing Laya
-  family. The ONNX export is fused, but the prompt and calibrated temperatures
-  are the same as the split Laya contract. It needed one adapter branch, not a
-  new model family.
-- **Open-Jev DeBERTa v3 large** is an independent, open reproduction of the
-  typed-decision shape. It has a complete fp16/fp32/q4/q4f16 family and a
-  documented span-pooling graph.
-- **Strands Decider 2B** is the first larger decision model in the catalog. It
-  has a published reference fixture and a clear pointer-head graph. It is more
-  expensive on CPU, so it is marked as a quality/agentic option rather than a
-  replacement for the small models.
-
-`gliner2.5-multi-decide` is **not** in this batch. Its ONNX export is real, but
-the prompt uses `[L]` markers instead of the existing GLiNER2 `[C]` layout; it
-needs a separate prompt adapter and its own parity run.
-
-## Verification performed
-
-- `dtk models verify --all --json`: **15/15 variants OK** (SHA256, files and
-  ONNX sessions).
-- Real inference for every new variant, using `choice`, `noul` and `score`.
-- Smoke parity suite: all added cases matched their expected labels.
-- `dtk models variants` exposes all variants and install commands.
-- Full test suite: **247 passed, 1 deselected**.
-- Ruff and MkDocs `--strict`: clean.
-- Python 3.12/3.13 CI matrix and wheel clean-install job: green.
-
-### Measured CPU latency
-
-Post-warmup p50, one simple choice question, audit machine (Python 3.13.2,
-ONNX Runtime 1.31.0, CPU):
-
-| Model | Variant | p50 |
-|---|---|---:|
-| Laya multilingual | fp16 | 37.4 ms |
-| Laya multilingual | fp32 | 34.1 ms |
-| Open-Jev DeBERTa v3 large | fp16 | 241.2 ms |
-| Open-Jev DeBERTa v3 large | fp32 | 78.2 ms |
-| Open-Jev DeBERTa v3 large | q4 | 56.0 ms |
-| Open-Jev DeBERTa v3 large | q4f16 | 219.4 ms |
-| Strands Decider 2B | q4f16 | 719.3 ms |
-| Strands Decider 2B | q8 | 1467.2 ms |
-
-The Open-Jev DeBERTa v3 large 4-bit paths are size/edge choices; they are not assumed to be
-faster or more accurate. Strands is a 2B model and is intentionally slower
-than the other catalog entries.
-
-### Evidence / parity
-
-- **Laya multilingual:** upstream conversion reports fp32 max |Δp| 1e-4 and
-  fp16 0.035 against PyTorch; local fp16/fp32 agree on the audit answers.
-- **Open-Jev DeBERTa v3 large:** upstream reports 0.854 in-domain and 0.690 OOD accuracy, plus
-  0.022 in-domain ECE. Local fp16/fp32 agree on the published example; q4
-  variants are explicitly edge builds.
-- **Strands Decider 2B:** reference fixture matches within 0.018 max
-  probability delta; q8 is closer to the reference on most fields.
-
-These are upstream parity claims plus local runtime/answer checks, not an
-independent reproduction of the full upstream training evaluation.
-
-## Quantization policy
-
-- Default variants are chosen for a balance of quality, size and CPU latency:
-  Laya fp16, Open-Jev DeBERTa v3 large fp16, Strands q4f16.
-- `fp32` remains a quality variant where published.
-- `q4`/`q4f16`/`q8` are exposed, but their thresholds must be validated on the
-  user's data; no silent fallback exists.
-- `variant` remains a **decision-toolkit extension** in REST, not a TypeSafe or
-  OpenAI field.
-
-## Risks and limitations
-
-1. **Upstream evidence vs local evidence.** Full PyTorch parity was not
-   re-run for these three models; the catalog records upstream parity and local
-   runtime checks. A larger acceptance run is still desirable.
-2. **Strands CPU cost.** q8 takes ~1.5 s per simple question; this is suitable
-   for local/agentic workflows, not high-throughput serving.
-3. **Quantization drift.** Open-Jev DeBERTa v3 large q4 and Strands q4f16 can move near-ties;
-   the model cards call this out.
-4. **Coverage.** Laya multilingual is weak on low-resource languages;
-   Open-Jev DeBERTa v3 large is English-only and weak on unseen score scales;
-   Strands is a 2B model with
-   its own upstream limitations.
-5. **No image/box support.** The catalog remains text-only.
-
-## Recommendation
-
-Accept the three models as `m0` entries with the variants listed above. Keep
-`gliner2.5-multi-decide` in the roadmap until its `[L]` adapter is implemented
-and parity is measured. The next model batch should prioritize either that
-adapter or a larger calibration/parity run for the new entries.
-
-## Related documentation
-
-- [Model cards](models/README.md)
-- [Variants and quantization](VARIANTS.md)
-- [Benchmarks](BENCH.md)
-- [REST](REST.md)

decision_toolkit-0.4.1/docs/ROADMAP.md

@@ -1,33 +0,0 @@
-# Roadmap (not in the catalog)
-
-The catalog only contains models that are **runnable today** (`m0`). The entries
-below are researched but not available through the toolkit: they have no ONNX
-export, no verified parity, a broken export, or an adapter still to be written.
-They are deliberately absent from `CATALOG`, `dtk models list` and `/v1/models`.
-
-| id | status | reason |
-|---|---|---|
-| `gliner2.5-multi-decide` | planned | ONNX split export exists (`DanKau/gliner2.5-multi-decide-onnx`), but the `[L]` prompt layout needs an adapter |
-| `intern-decision-0.8b` | planned | ONNX exists (`Wortou/Intern-Decision-0.8B-ONNX-FP16`), but it is split and needs the upstream prompt compiler; `parityVerified: false` |
-| `d3-lite` | phase2 | Web ONNX torso/readout files exist, but the repo declares no license/README; not eligible |
-| `kev-0.8b` | phase2 | ONNX exists (`midudev/kev-0.8b-ONNX`) but external data is missing and it uses custom ops |
-| `sieve-0.8b` | phase2 | No ONNX published; conversion planned |
-| `thx-01` | phase2 | No ONNX published; conversion planned |
-| `jevk5-lite` | waitlist | Export references external `.onnx.data` files that are not published |
-| `bosun-0.6b` | waitlist | No ONNX yet |
-| `d3-edge` | waitlist | Web ONNX files exist, but no license/README; not eligible |
-| `tev1-0.8b` | waitlist | ONNX exists, but the upstream licence is `other` and needs review |
-| `d1-omni-600m` | waitlist | ONNX exists, but the LFM Open License needs review and it is multimodal |
-| `bekko-400m` | waitlist | License pending |
-
-## How a model moves into the catalog
-
-1. A real, runnable ONNX export exists on the Hub.
-2. Its contract is understood and an adapter runs it (written only from the real files).
-3. Parity against the original is measured on a fixed sample.
-4. License is clean and CPU latency is measured.
-5. A model card is written; the variant carries `verified_at`, `latency_ms_p50`
-   and `parity`.
-
-Conversion work (safetensors → ONNX, quantization, manifests) is a separate
-project planned for the next phase.