Skip to content

Releases

Puckworks v0.3.0 — Guided Espresso Pull

Choose a tag to compare

@trbrewer trbrewer released this 18 Jul 15:30

Puckworks v0.3.0 — Guided Espresso Pull

The Guided Espresso Pull — a rights-independent, runnable guided mechanism explorer — under the
supported puckworks.product API. Enter a bounded recipe and run one coherent, model-backed espresso
pull on CPU, stage by stage, with evidence-labelled traces, a visual report, deterministic exports,
and a guided Google Colab notebook. Also adds the caller-owned input normalization boundary.

What you get

  • Package API: load_pull_preset, simulate_pull, render_pull_report, pull_run_to_json /
    pull_run_to_markdown; records PullRecipe / PullConfig / PullRun / PullSeries / PullTrace
    / ComponentCoverage / PullReportArtifacts.
  • CLI: puckworks-pull presets, puckworks-pull run …, and --report-dir/--figures.
  • Authoritative traces copied from the solver at full precision, each labelled by value role
    (prescribed input · simulated output · derived quantity).
  • Visual report: evidence-badged figures + captions + JSON/Markdown (matplotlib is an optional
    [viz] extra; the core stays matplotlib-free).
  • Google Colab: notebooks/guided_espresso_pull_colab.ipynb with native form controls, installing
    this exact release wheel after SHA-256 verification.

One coherent model — and its honest limits

The one executed chain is cameron2020.extraction_bdf (grind → machine/flow → extraction → cup). It
is an EXPLORATORY_SIMULATION, code-verified against the source paper, not independently validated
against a measured cup. Active recipe controls: dose, target beverage, prescribed constant pressure,
EK43 grind, coffee profile, domain policy. Bean label is metadata only. Temperature is
recorded-only
(no thermal transient). It does not model physical first drip or puck
wetting
(the model begins from a saturated bed), does not simulate a dynamic pressure
profile
(pressure is prescribed constant), reports chemical composition (never sensory flavor),
and never executes or averages a parallel lens into a consensus. Alternative components get an
explicit, non-executed coverage disposition.

Install (GitHub-only — not on PyPI)

Download puckworks-0.3.0-py3-none-any.whl from this release and pip install the file. Puckworks is
not published on PyPI.

Provenance

  • Source commit: c5ab770b76ea2fb876c348ca48d802d604c112ca
  • puckworks-0.3.0-py3-none-any.whl — SHA-256 4a88931fc0bbc069365242fc5e3ef4d226da9225c80384b6d013028e4b7a3ec0 (1,036,725 bytes)
  • puckworks-0.3.0.tar.gz — SHA-256 dc89d1f40c2e8329ac34b7a474f7b03cb7f17c5c8952f9232a2a2daa3954b1ca (1,064,496 bytes)

No private or rights-gated fixture is distributed. See CHANGELOG.md for the full change list.

puckworks v0.2.0

Choose a tag to compare

@trbrewer trbrewer released this 17 Jul 01:08
458ee70

puckworks v0.2.0

First verifiable post-0.1 release: canonical status source, complete (non-short-circuiting) gate
evaluation, a deterministic Paper 3 archive, a hardened CI supply chain, and a fully-adjudicated
per-claim evidence graph.

Source commit: 458ee70bb2eedd79231385fa2d2dad2e1d457aaf (annotated tag v0.2.0 peels exactly to this commit).

Supported environment

  • Python 3.10–3.13; 3.12 is the primary / release interpreter (the canonical wheel and sdist are built on 3.12).
  • Dependencies: numpy>=2.0 (uses np.trapezoid), scipy>=1.13. The attached
    requirements-release.lock.txt is the fully-resolved, hash-enforced py3.12 environment.
  • The wheel is py3-none-any (pure Python) — one artifact serves all supported interpreters.

Public API & compatibility

The supported public API is puckworks.__all__: evaluate_all_gates, components, get,
load_builtin_components, Component, GateStatus/GateResult/GateSuiteResult, and the
contracts / registry / validate namespaces. Everything else (harness, analysis, paper3,
paper_a, paper_b, figures, lib, viz, inventory) is internal research tooling and may
change without notice
. No public numerical behaviour change vs. 0.1; existing examples reproduce.

Gate result

All gates evaluated on the frozen commit: 50 PASS + 1 ACKNOWLEDGED_EXCEPTION (51 total), suite passed=True.

The one acknowledged exception — brewer2026.lb_taichi: its code_verification evidence is a
0.003% cross-check to lb_reference that runs in the Colab notebook, not CI — Taichi is an
optional dependency and the package must import without it. Physics anchor:
lb_reference.gate_lb_channel. This is a documented, expected exception, not a failure.

Attached assets

asset what it is
puckworks-0.2.0-py3-none-any.whl canonical wheel (tag-run build, py3.12)
puckworks-0.2.0.tar.gz canonical sdist (tag-run build, py3.12)
release_manifest.json build provenance: commit, environment, per-artifact SHA-256
requirements-release.lock.txt fully-resolved, hash-enforced py3.12 dependency lock
paper3_archive.tar.gz (+ .sha256) deterministic, source-checkout-independent Paper 3 evidence archive (built twice byte-identical)
gate_report.json authoritative gate report (51 results)
evidence_report.json Paper 3 evidence-graph summary
release_provenance.json machine-readable release provenance record
SHA256SUMS.txt checksums for every asset above

Artifacts were built by the tag-triggered release.yml run; the canonical files' extracted contents
are byte-identical to the independent workflow-dispatch preflight build and a local build (only the
archive-container hashes differ). Every asset's SHA-256 is listed in SHA256SUMS.txt and mirrored in
release_provenance.json.

Publication status

No package-registry publication has occurred. This release has not been published to PyPI or any
other package index. Installation is from the attached wheel/sdist or the source tag.