Skip to content

docs+build: truth-align public claims and complete discovery metadata - #176

Merged
niveousdragon merged 4 commits into
mainfrom
docs/marketing-truth-and-metadata
Jul 29, 2026
Merged

docs+build: truth-align public claims and complete discovery metadata#176
niveousdragon merged 4 commits into
mainfrom
docs/marketing-truth-and-metadata

Conversation

@levi770

@levi770 levi770 commented Jul 29, 2026

Copy link
Copy Markdown
Contributor

What

Pre-publication gate fixes agreed in #xlog-marketing (operator-authorized 2026-07-29). Four commits:

docs commits — truth alignment:

  • The type-inference claim ("rejects ill-typed programs before any GPU kernel runs") stays as written — it is the spec. The observed gap (cross-predicate schema mismatches currently surface as Kernel("Union requires compatible schemas...") execution errors on paper/examples/02_types_err.xlog, while arithmetic mismatches are correctly frontend-rejected) is tracked as maintainer bug Reject cross-predicate schema mismatches before GPU execution #177 per operator direction; an earlier commit in this PR reworded the claim downward and commit 0ebc1b5d8 reverts that.
  • The 2.74× MNIST-addition speedup is now attributed to its actual experiment: the cached-vs-uncached circuit-compilation ablation within XLOG's own pipeline (paper/sections/08_evaluation.tex), not "a CPU-resident baseline"; the properly scoped 2.80×-vs-Scallop steady-epoch comparison is added (README + docs/index.mdx).
  • Zero-copy claims now name DLPack and the Arrow C Device interface only; Arrow IPC is labeled host-side interchange (5 locations).
  • Source-install snippet leads with cargo build --release -p xlog-cli --features host-io; the previous order produced a first-run xlog prob failure ("Host output is disabled").
  • Flag for reviewer: updates the ROADMAP Current release: marker v0.10.0 → v0.11.0. Repo agent rules say agents never edit release markers; this corrects a stale marker toward the already-published state (release-plz published v0.11.0 on 2026-07-29) and was explicitly operator-authorized. Reject this hunk if you read the rule strictly.

build commit — discovery metadata (takes effect at next release):

  • Workspace keywords: drop self-referential xlog, add logic-programming + neural-symbolic.
  • xlog-cli: categories were EMPTY on crates.io; now 5 set. Sharper description.
  • pyxlog: PyPI keywords (were none), full classifiers (were 3), and a verified quick-start in the landing README (condensed from examples/python/01_dlpack_reachability_torch.py, runs green on RTX PRO 3000 / CUDA 13.1).
  • release-plz git_release_body for xlog-cli: v0.10.0 and v0.11.0 GitHub releases shipped 0-byte bodies (suspected cause: duplicate per-crate ## [x.y.z] headings in the shared CHANGELOG leave {{ changelog }} empty). The template adds a stable install/docs footer so the body is never blank.

Verification

  • cargo metadata --no-deps passes; pyproject.toml and release-plz.toml parse cleanly (tomli).
  • CUDA certification suite run on this branch's worktree 2026-07-29: 207/207 pass (RTX PRO 3000).
  • No version, tag, changelog, or release-commit changes. No behavior changes.

Related: #177 (type-checking gap this PR deliberately does not paper over).

levi770 added 2 commits July 29, 2026 15:57
…runtime behavior

- Replace the 'single-pass type inference rejects ill-typed programs before
  any GPU kernel runs' claim with what the runtime actually does: arithmetic
  type mismatches are rejected at compile time (Compilation error with cast()
  hint), while cross-predicate schema violations fail closed at execution.
  Verified against paper/examples/02_types_err.xlog and
  examples/xlog/90-negative-tests/03_arithmetic_type_mismatch.xlog on v0.11.0.
- Lead the source-install snippet with the host-io CLI build; a default
  workspace build produces a CLI whose 'xlog prob' cannot print results.
- Update the stale ROADMAP 'Current release' marker from v0.10.0 to the
  actually published v0.11.0 (release-plz released it 2026-07-29); explicitly
  operator-authorized correction toward published state.
…body template

- Workspace keywords: drop the self-referential 'xlog', add
  'logic-programming' and 'neural-symbolic' (crates.io max 5).
- xlog-cli: add the previously empty categories list and a description that
  names the actual capabilities.
- pyxlog: add PyPI keywords and the full classifier set (GPU/CUDA environment,
  AI topic, per-version Python), sharpen the summary, and give the PyPI
  landing README a verified quick-start example and feature list (condensed
  from examples/python/01_dlpack_reachability_torch.py).
- release-plz: add git_release_body for xlog-cli so GitHub release bodies are
  never blank again (v0.10.0 and v0.11.0 shipped with 0-byte bodies; the
  shared workspace changelog's duplicate per-crate version headings can leave
  {{ changelog }} empty, so the template keeps a stable install/docs footer).
Copilot AI review requested due to automatic review settings July 29, 2026 13:57

Copilot AI left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

…ro-copy to device interfaces

- README and docs/index.mdx claimed the 2.74x MNIST-addition speedup was
  'over a CPU-resident baseline'; paper/sections/08_evaluation.tex defines it
  as the cached-vs-uncached circuit-compilation ablation within xlog's own
  pipeline (242.9s -> 88.9s, 3 seeds, 95% CI [2.29, 3.18]). Reworded to state
  the ablation and added the properly-scoped 2.80x steady-epoch Scallop
  comparison at matched accuracy.
- Zero-copy claims now name DLPack and the Arrow C Device interface only;
  Arrow IPC is labeled host-side interchange (README intro, feature bullet,
  interop table row, docs-table row, docs/index.mdx).
Copilot AI review requested due to automatic review settings July 29, 2026 14:09

Copilot AI left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

…ap as an issue

Per maintainer direction: the cross-predicate schema-violation gap between the
README claim and current runtime behavior is a product bug to fix, not a doc
to weaken. Reverts the earlier rewording; the behavior gap is now tracked in
a dedicated issue with reproduction against the repo's own ill-typed example.
Copilot AI review requested due to automatic review settings July 29, 2026 14:53

Copilot AI left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

@niveousdragon
niveousdragon merged commit c63dea7 into main Jul 29, 2026
16 checks passed
@niveousdragon
niveousdragon deleted the docs/marketing-truth-and-metadata branch July 29, 2026 18:19
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants