blog: every post carries a widget, as a t27 rule the work report enforces (Closes #1576)

[gHashTag]

Closes #1576

The owner's rule of 2026-10-09, that every t27.ai post carries a widget, is now a t27 spec, and the required T27 work report check enforces it on every PR that adds or changes a post. The 14 posts that broke it now have widgets.

The rule: apps/website/specs/policy/post_widget.t27, next to course_post.t27 and in the same style.

• It holds the rule sentence, WIDGET_KINDS = ["terminal", "figure"], the cutoff WIDGET_SINCE = "2026-09-27", and GRANDFATHERED: the 72 posts dated before the cutoff that carry no widget, listed by name.
• It has 5 test blocks. t27c test-report (0.5.2) passes 5 of 5 with 0 vacuous passes.

The glue stays at the Only-t27 cap: 40 + 32 + 8 = 80 added foreign lines.

• pr-blog-report.yml saves each changed post body as text from the PR head, through the contents API. It never checks out or runs the PR.
• scripts/pr_blog_report.py validate --posts reads the spec's constants with the same regexes as course_post.t27. It refuses a body that has no widget in body or ruBody, and fails closed when the spec is broken or a body was not saved.
• docs/PR_BLOG_AUTOMATION.md gets a short section that points at the spec.

Posts and casts. Every cast already exists under public/term/. I read each cast's text before writing its caption, and the Russian body gets the same blocks.

Post	Widget(s)
three-dies-one-verdict	tri-fpga-jtag
an-eye-on-the-board	tri-fpga-jtag, t27c-eth-crc
buses-and-peripherals	t27c-verilog-uart, t27c-eth-crc
clocks-resets-and-cdc	t27c-mtbf, t27c-mmcm-mutate
course-three-verifying-hardware	tri-fpga-wire, t27-tri-seals
course-three-goldenfloat	t27c-tnf17
nine-modules-of-three-again	t27c-gft-xornet
course-zero-t27-basics	t27c-hello-world
tri-test-now-runs-the-tests	t27c-gft-relu
two-courses-twenty-seven-lessons-each	t27c-bitnet-mlp, t27c-gft-smul
a-course-pr-carries-its-post	tri-blog-list
one-regional-clock-twelve-pull-requests	perf2-knockouts
the-seed-moved-it-the-pins-moved-it-more	tri-fpga-jtag, tri-fpga-tools
mask-over-scale-ternary-drafts-that-compile	a figure drawn from the post's own numbers (no recording of tri tc-draft exists)

One text fix: the clocks post said one MMCM test fails when the VCO floor is lowered. Its recording shows two, vco_below_the_window_is_rejected and vco_out_of_range_is_an_error, so the sentence now says two, in both languages.

{
  "version": 1,
  "head_sha": "b4d6e9ffdc9c79b3084e49bdae5a56669e733dbf",
  "summary": "The owner's rule that every t27.ai post carries a widget is now a t27 spec, post_widget.t27. The required work-report check enforces it on changed post bodies, and the 14 widgetless posts dated on or after 2026-09-27 get casts or a figure, in English and Russian.",
  "changes": [
    "apps/website/specs/policy/post_widget.t27: the rule sentence, WIDGET_KINDS terminal and figure, WIDGET_SINCE 2026-09-27, 72 grandfathered post names, and 5 test blocks.",
    "scripts/pr_blog_report.py: post_widget_policy reads the spec's constants and fails closed; post_widget_refusal refuses an added, modified or renamed body with no widget in body or ruBody; validate takes --posts (40 added lines).",
    "scripts/test_pr_blog_report.py: 3 tests that cover the real spec, a broken spec, both languages, grandfathered, removed and unsaved bodies (32 added lines).",
    ".github/workflows/pr-blog-report.yml: a step saves each changed post body from the PR head as text through the contents API, and validate gets --posts (8 added lines).",
    "apps/website/src/data/blog/bodies/*.ts: 14 posts get existing terminal casts (one gets a figure) in English and Russian, and the clocks post's MMCM sentence is corrected from one failing test to two.",
    "docs/PR_BLOG_AUTOMATION.md: a short section that points at the spec."
  ],
  "tests": [
    {
      "command": "t27c test-report apps/website/specs/policy/post_widget.t27 (t27c 0.5.2, zig 0.16.0)",
      "status": "passed",
      "result": "5 of 5 pass, 0 vacuous; WIDGET_KINDS[1] changed to chart fails a_widget_is_a_cast_or_a_figure, and a changed last grandfathered name fails the_grandfathered_list_is_closed",
      "evidence": "local run on this head, 2026-10-09"
    },
    {
      "command": "python3 -m unittest discover -s scripts -p 'test_pr_blog*.py'",
      "status": "passed",
      "result": "74 tests OK; five planted bugs in the new glue (ruBody skipped, grandfathering off, length check off, removed bodies checked, wrong segment matched) each fail one test",
      "evidence": "local run on this head, 2026-10-09"
    },
    {
      "command": "post_widget_refusal over every body file, main against this branch",
      "status": "passed",
      "result": "on main's 98 bodies it refuses exactly 14, the 11 named in the issue plus 3 from 09-29 to 10-02; on this branch it refuses 0, with 72 grandfathered",
      "evidence": "local run on this head, 2026-10-09"
    },
    {
      "command": "validate --posts on trinity#1569's real files and head (the fetch step simulated with gh api)",
      "status": "passed",
      "result": "exit 1, refused: three-dies-one-verdict.ts has no widget in body or ruBody; without --posts it exits 0 as before",
      "evidence": "local run, 2026-10-09"
    },
    {
      "command": "esbuild bundle of src/data/blog/posts.ts, widget count per post dated on or after 2026-09-27, and a check that each terminal src exists under public/",
      "status": "passed",
      "result": "23 posts, every one has at least one widget in both languages, the same number in each, and 0 missing recordings",
      "evidence": "local run on this head, 2026-10-09"
    },
    {
      "command": "npm run typecheck:ratchet and npx vite build in apps/website",
      "status": "passed",
      "result": "179 errors across 26 files, the same as the baseline; the build finished in 25 s",
      "evidence": "local run on this head, 2026-10-09"
    },
    {
      "command": "vite preview, #/blog/three-dies-one-verdict and #/blog/mask-over-scale-ternary-drafts-that-compile in a browser",
      "status": "passed",
      "result": "the tri-fpga-jtag player mounts and types the recording; the figure renders four bars with its caption",
      "evidence": "local run on this head, 2026-10-09"
    },
    {
      "command": "the website-checks gate on this PR",
      "status": "not_run",
      "result": "not run locally beyond typecheck and build",
      "evidence": "the website-checks workflow on this pull request runs it"
    }
  ],
  "limitations": [
    "The new rule runs from main (pull_request_target), so it binds PRs after this one merges, not this PR.",
    "The PR-time check cannot see whether a terminal block's recording exists, because the workflow never checks out the PR; the site's player shows a missing one.",
    "Widgets are matched by the text kind: 'terminal' or \"kind\": \"figure\"; a block built by a function call would not be seen.",
    "72 older posts stay without widgets, by name; the list only shrinks.",
    "Owner rule says each post also needs its og-art triptych; this PR does not touch triptychs.",
    "t27c cannot lower != on two str values yet (Zig: operator != not allowed for []const u8), so the spec does not assert that the two widget kinds differ."
  ],
  "tags": [
    "Blog",
    "Policy",
    "t27"
  ],
  "blog": {
    "title": "Every post carries a widget",
    "summary": "A rule that was said once and written nowhere let 11 posts ship as text only. It is now a t27 spec, and the required check enforces it on every post a PR changes.",
    "outline": [
      "Eleven posts in a row shipped with no recording, because the widget rule lived only in a conversation.",
      "post_widget.t27 holds the rule, the two widget kinds and the grandfathered posts, and pr_blog_report.py reads it on every PR.",
      "Fourteen posts got casts that show the command they discuss; one got a figure, and one caption corrected its own post."
    ]
  }
}

🤖 Generated with Claude Code

[tri-claw]

Summary by t27-bees

Checked by the t27 compiler

_{t27c parse, parse-complete and typecheck on each changed .t27 file at b4d6e9f. Nothing was generated, built or run. A pass means the file is well formed, not that it is right.}

• apps/website/specs/policy/post_widget.t27: parses and typechecks.

Read by the model

This PR codifies the owner's rule that every t27.ai blog post dated on or after 2026-09-27 must carry a widget (terminal cast or figure) in both English and Russian bodies. It adds a t27 spec (post_widget.t27) with the rule, constants, a grandfathered list of 72 pre-cutoff posts, and 5 tests; extends the PR workflow to fetch changed post bodies via the GitHub Contents API and validate them; updates the validation script to enforce the widget requirement; and adds widgets to 14 previously non-compliant posts (plus a text correction in one post).

Worth a look

• .github/workflows/pr-blog-report.yml:54-60 — The POST_DIR is extracted from the spec file using a fragile sed regex (^pub const POST_DIR : str = "\(.*\)";$). If the spec's formatting changes (whitespace, quotes, line breaks), the extraction fails silently and the validation runs against an empty or wrong directory. The workflow runs on pull_request_target (base branch), but validates PR-head bodies; if a PR modifies the spec itself, the base-branch spec and head-branch bodies could mismatch.

• apps/website/specs/policy/post_widget.t27:68-70 — The validation only checks that a widget block exists in the body text (kind: 'terminal' or kind: 'figure'). It does not verify that the referenced cast recording actually exists under public/term/<id>/session.cast, nor that the SVG in a figure is well-formed. The spec acknowledges this gap ("the check cannot see whether a terminal block's recording exists"), but a post could pass the check with a broken or missing cast reference.

• apps/website/specs/policy/post_widget.t27:87-158 — The GRANDFATHERED list of 72 post names is hardcoded and manually maintained. The spec states "The list only shrinks: a post that gains a widget may leave it, and no post joins it," but there is no automated enforcement of this policy; a future edit could accidentally add entries or fail to remove a post that has gained a widget.

• scripts/pr_blog_report.py (not fully shown) — The description says the script "reads the spec's constants with the same regexes as course_post.t27" and "fails closed when the spec is broken or a body was not saved." If the regex parsing of the spec drifts from the actual t27 syntax (e.g., multi-line arrays, comments, trailing commas), the validation could misread constants and either pass invalid posts or reject valid ones.

• apps/website/src/data/blog/bodies/course-3-clocks-resets-and-cdc.ts:47,55 — The text fix changes "the one test that fails" to "the two tests that fail" and names them (vco_below_the_window_is_rejected and vco_out_of_range_is_an_error). This aligns the post with the recorded cast, but the cast caption in the same file (line 55) says "14 of 16 tests still pass, and the two that fail..." — ensure the English and Russian bodies both reflect the same two-test count (the Russian body at line 134 appears correct).

---

_{Read: 19 files, +512 -3, the whole diff up to 60000 characters (the rest was not read). Head b4d6e9f. Model: openai-compatible/nvidia/nemotron-3-ultra-550b-a55b. Commands: @t27-bees review | summary | help | pause | resume. Free for every repository.}

[github-actions]

X Brain Health Check

Score: 100.0/100
Status: 🟢 HEALTHY
Threshold: 80/100

X Brain is above merge threshold

[github-actions]

🧪 Brain Stress Test Results

Tests: 261/261 passed, 0 leaked
Status: 🟢 PASS
Gate: every test in zig build test-brain-stress passes and none leaks