fix(converter): see grids, forms and link labels - 100% design copy captured #110

Merged

peterswimm merged 1 commit from fix/toilville-source-regressions into main 2026-09-09 23:45:59 +00:00

Conversation 1 Commits 1 Files changed 31 +2840 -1739

peterswimm commented 2026-09-09 23:33:05 +00:00

Owner

Copy link

First step toward 1:1 on #102. Design copy reaching the ForgeBlock tree goes
from 97/100/98/92/80% per page to 100% on all five.

Four converter fixes — three are the same underlying bug

A container's own text was dropped whenever the walker only recursed into child
elements.

1. _grid_columns() only matched repeat(<digit>. Every grid in the design
is repeat(auto-fit, minmax(Npx, 1fr)) or an explicit track list
(150px 1fr auto, 208px 1fr, 1.35fr 1fr) — so none were detected and
every grid flattened. Now resolves auto-fit/auto-fill from the child count
and counts explicit tracks paren-aware, so minmax()'s comma doesn't split.

2. A column that IS its own text vanished. Table cells and teaser fields are
<span>2026-07-14</span> with no child elements, so _convert_children
produced nothing and the column was skipped. That is how the news titles and the
console's intent/handler/p95 values disappeared the moment grids started being
detected.

3. A text-only <a> emitted nothing — losing every plain link label ("All
news →", "Enter the portal →", "Browse RFCs →", "Read the post →"). Now emits a
one-button actions run, which is what it is.

4. input could not be produced at all — no heuristic, no data-block
case. Adds both, plus bare <button>. The design's controls carry no
name/id/value, so the action_id is synthesised from the label text and
select options are read from <option> text.

The Contact form now yields 5 input blocks (2 text, email, a 4-choice
select, textarea) where it previously yielded nothing.

Two source regressions fixed — both mine

• The Spelwork Documentation grid was broken markup. My CTA-annotation pass
  in #101 classified the six full-bleed card links as buttons (padding +
  background matched is_button()) and the line-based wrapper assumed
  single-line anchors, so </div> closed inside the still-open <a>,
  orphaning every card's content. All five sources now parse with zero tag
  mismatches (checked with html.parser).
• background-color: #cccccc left on the three Home section headers — the grey
  bars visible in the review renders.

And a silent test hole

test_forgeblock_grammar located the sibling Rituals checkout as
../Toilville_Rituals relative to the source tree. That's wrong inside a git
worktree, so 18 grammar tests silently skipped instead of running. Now asks
git for the real repo root — 26 pass in both the main checkout and a worktree.

Verification

• just check green; pytest generator/tests/ tools/wp_theme_ingestor/test_*.py — 267 passed
• All five sources well-formed, 0 mismatches, 0 unclosed tags
• Cross-brand: zero copy lost on any of the other five brands. fux-capacity,
  pardner and tctd each gain a column_set; peterswimm and tctd gain an
  actions run; fux-capacity gains 5 words it was dropping.

What this does not do

Structure is now captured; it is not yet rendered faithfully. column_set
renders as generic wp:columns and input renders as plain paragraphs, so the
patterns are still not 1:1. The block work (generalized card-grid, the
terminal, console, ritual demo, and wish-form on Contact) follows in
toilville_rituals, with adapters here after.

https://claude.ai/code/session_01Kbzfscp4w5enUrZxMnDt78

First step toward 1:1 on #102. **Design copy reaching the ForgeBlock tree goes from 97/100/98/92/80% per page to 100% on all five.** ## Four converter fixes — three are the same underlying bug A container's own text was dropped whenever the walker only recursed into child *elements*. **1. `_grid_columns()` only matched `repeat(<digit>`.** Every grid in the design is `repeat(auto-fit, minmax(Npx, 1fr))` or an explicit track list (`150px 1fr auto`, `208px 1fr`, `1.35fr 1fr`) — so **none** were detected and every grid flattened. Now resolves `auto-fit`/`auto-fill` from the child count and counts explicit tracks paren-aware, so `minmax()`'s comma doesn't split. **2. A column that IS its own text vanished.** Table cells and teaser fields are `<span>2026-07-14</span>` with no child elements, so `_convert_children` produced nothing and the column was skipped. That is how the news titles and the console's intent/handler/p95 values disappeared the moment grids started being detected. **3. A text-only `<a>` emitted nothing** — losing every plain link label ("All news →", "Enter the portal →", "Browse RFCs →", "Read the post →"). Now emits a one-button `actions` run, which is what it is. **4. `input` could not be produced at all** — no heuristic, no `data-block` case. Adds both, plus bare `<button>`. The design's controls carry **no `name`/`id`/`value`**, so the `action_id` is synthesised from the label text and select options are read from `<option>` text. The Contact form now yields **5 `input` blocks** (2 text, email, a 4-choice select, textarea) where it previously yielded nothing. ## Two source regressions fixed — both mine - **The Spelwork Documentation grid was broken markup.** My CTA-annotation pass in #101 classified the six full-bleed card links as buttons (`padding` + `background` matched `is_button()`) and the line-based wrapper assumed single-line anchors, so `</div>` closed *inside* the still-open `<a>`, orphaning every card's content. **All five sources now parse with zero tag mismatches** (checked with `html.parser`). - `background-color: #cccccc` left on the three Home section headers — the grey bars visible in the review renders. ## And a silent test hole `test_forgeblock_grammar` located the sibling Rituals checkout as `../Toilville_Rituals` relative to the source tree. That's wrong inside a git worktree, so **18 grammar tests silently skipped instead of running**. Now asks git for the real repo root — 26 pass in both the main checkout and a worktree. ## Verification - `just check` green; `pytest generator/tests/ tools/wp_theme_ingestor/test_*.py` — **267 passed** - All five sources well-formed, 0 mismatches, 0 unclosed tags - **Cross-brand: zero copy lost on any of the other five brands.** `fux-capacity`, `pardner` and `tctd` each gain a `column_set`; `peterswimm` and `tctd` gain an `actions` run; `fux-capacity` gains 5 words it was dropping. ## What this does not do Structure is now *captured*; it is not yet *rendered* faithfully. `column_set` renders as generic `wp:columns` and `input` renders as plain paragraphs, so the patterns are still not 1:1. The block work (generalized `card-grid`, the terminal, console, ritual demo, and `wish-form` on Contact) follows in toilville_rituals, with adapters here after. https://claude.ai/code/session_01Kbzfscp4w5enUrZxMnDt78

peterswimm added 1 commit 2026-09-09 23:33:07 +00:00

fix(converter): see grids, forms and link labels — 100% design copy captured ... a1bed787e0

Design copy reaching the ForgeBlock tree goes from 97/100/98/92/80% per page to
100% on all five. Four fixes, three of which are the same underlying bug: a
container's own text was dropped whenever the walker only recursed into child
ELEMENTS.

1. _grid_columns() only matched `repeat(<digit>`. Every grid in the toilville
   design is `repeat(auto-fit, minmax(Npx, 1fr))` or an explicit track list
   (`150px 1fr auto`, `208px 1fr`, `1.35fr 1fr`), so NONE were detected and every
   grid flattened. Now resolves auto-fit/auto-fill from the child count and
   counts explicit tracks (paren-aware, so minmax() commas don't split).

2. A column that IS its own text vanished. Table cells and teaser fields are
   `<span>2026-07-14</span>` with no child elements, so _convert_children
   produced nothing and the column was skipped — which is how the news titles
   and the console's intent/handler/p95 values disappeared once grids started
   being detected at all.

3. A text-only <a> emitted nothing, losing every plain link label ("All news",
   "Enter the portal", "Browse RFCs", "Read the post"). Now emits a one-button
   actions run, which is what it is.

4. `input` could not be produced at all — no heuristic, no data-block case. Adds
   both, plus <button>. The design's controls carry no name/id/value, so the
   action_id is synthesised from the label text and select options are read from
   the <option> text. The Contact form now yields 5 input blocks (2 text, email,
   a 4-choice select, textarea) instead of nothing.

Two source regressions fixed, both mine:

  - The Spelwork Documentation grid was BROKEN markup. My CTA-annotation pass in
    #101 classified the six full-bleed card links as buttons (padding +
    background matched) and the line-based wrapper assumed single-line anchors,
    so the closing div landed inside the still-open anchor, orphaning every
    card's content. All five sources now parse with zero tag mismatches.
  - background-color #cccccc left on the three Home section headers — visible as
    grey bars in the review renders.

Also: test_forgeblock_grammar located the sibling Rituals checkout relative to
the source tree, which is wrong inside a git worktree — 18 grammar tests
silently SKIPPED instead of running. Now asks git for the real repo root.

Cross-brand: zero copy lost on any of the five other brands; fux-capacity,
pardner and tctd each gain a column_set, peterswimm and tctd gain an actions run.

Claude-Session: https://claude.ai/code/session_01Kbzfscp4w5enUrZxMnDt78

peterswimm merged commit 49689e9efc into main 2026-09-09 23:45:59 +00:00

peterswimm deleted branch fix/toilville-source-regressions 2026-09-09 23:45:59 +00:00

peterswimm referenced this pull request from a commit 2026-09-09 23:46:00 +00:00

Merge pull request 'fix(converter): see grids, forms and link labels - 100% design copy captured' (#110) from fix/toilville-source-regressions into main

peterswimm commented 2026-09-10 00:11:17 +00:00

Author

Owner

Copy link

Reworked: the vocabulary decides, the client renders

Peter's steer — ritual blocks are universal, clients like Gutenberg decide how
to render them — meant the adapter approach in this branch was backwards. It
pattern-matched runs of ForgeBlocks and bypassed the vocabulary entirely.
Rewritten.

Vocabulary (udg/config/forgeblock.yml)

• code — a monospaced literal. of: [text], enums: kind: [terminal, source].
  There was no way to say "terminal transcript"; diagram renders to wp:code
  but means Mermaid.
• column_set — variant: [plain, cards, table, list] declared, and flipped
  to used: true. It was the most important composite in the design while
  marked unused, and a client had no way to ask what kind of grid it had.

Adding one block here is deliberately not a one-line change. The repo demanded,
and this supplies: a renderer in the WP pattern, email and ASCII
clients; text · BBS · SMS · RCS · Apple Business Chat reductions; a mock in
the Ritual Blocks catalog; and the count tripwire bumped 32 → 33.
wcag.py report again shows all 33 blocks reducing to every channel.

The client boundary is now real

forgeblock_pattern takes a client and no longer imports one — brand-agnostic
in fact, not just intent. build_surfaces looks it up per brand, so a brand
without one gets the universal rendering. Verified: no toilville block leaks
into the other five brands.

render_sequence gives the client first refusal at every depth. A choice
that only held at the top level meant the Contact form — which lives inside a
column — silently reverted to the universal rendering.

forgeblock_toilville is now two levels:

render_block(block)     one universal block, rendered richer
                        code            -> toilville/terminal
                        column_set{cards} -> toilville/card-grid + card
render_run(blocks, i)   composites that are not any single vocabulary block
                        (a hero is context + header + section + actions),
                        still keyed on TYPES

Converter

<article> joins the container tags — the design puts the grid on the
article for news rows, so those never registered as a column_set at all.
column_set now records its variant and its chrome, so the client doesn't
guess. Card fields are read in order (a short run before the heading is the
mono slot; prose after it is the body) — reading purely by type put the tag in
the body and left the real body on the floor.

Guards worth keeping

The card renderer refuses a column carrying inputs, a terminal, a nested grid,
or a long repeated run. Without them the Contact form rendered as a card, and
the ritual demo's step list and signed history were dropped past the first
paragraph. Both were caught by measuring rendered output, not by reading code.

Result

98% of design copy survives to rendered HTML — landing and rituals 100%,
spelwork 98%, news 94%, contact 92% (the remainder is a handful of
contractions). Every page emits toilville blocks. Contact renders the real
wish-form with Name, Email, Organization, Interested in, and the custom
message label.

just check green, 270 tests pass.

Still not wired

toilville/ritual-demo and toilville/intent-console exist (toilville_rituals#368)
but nothing emits them. Doing that properly means annotating the design source
so the demo's steps become status blocks and its history becomes
lineage_entry — the converter can't produce lineage_entry at all today. That
is the vocabulary-first path and it is the next piece, rather than another
shape-matcher.

## Reworked: the vocabulary decides, the client renders Peter's steer — *ritual blocks are universal, clients like Gutenberg decide how to render them* — meant the adapter approach in this branch was backwards. It pattern-matched runs of ForgeBlocks and bypassed the vocabulary entirely. Rewritten. ### Vocabulary (`udg/config/forgeblock.yml`) - **`code`** — a monospaced literal. `of: [text]`, `enums: kind: [terminal, source]`. There was no way to say "terminal transcript"; `diagram` renders to `wp:code` but means Mermaid. - **`column_set`** — `variant: [plain, cards, table, list]` declared, and flipped to `used: true`. It was the most important composite in the design while marked unused, and a client had no way to ask what *kind* of grid it had. Adding one block here is deliberately not a one-line change. The repo demanded, and this supplies: a renderer in the **WP pattern**, **email** and **ASCII** clients; **text · BBS · SMS · RCS · Apple Business Chat** reductions; a mock in the Ritual Blocks catalog; and the count tripwire bumped 32 → 33. `wcag.py report` again shows **all 33 blocks reducing to every channel**. ### The client boundary is now real `forgeblock_pattern` takes a `client` and no longer imports one — brand-agnostic in fact, not just intent. `build_surfaces` looks it up per brand, so a brand without one gets the universal rendering. **Verified: no toilville block leaks into the other five brands.** `render_sequence` gives the client first refusal at **every depth**. A choice that only held at the top level meant the Contact form — which lives inside a column — silently reverted to the universal rendering. `forgeblock_toilville` is now two levels: ``` render_block(block) one universal block, rendered richer code -> toilville/terminal column_set{cards} -> toilville/card-grid + card render_run(blocks, i) composites that are not any single vocabulary block (a hero is context + header + section + actions), still keyed on TYPES ``` ### Converter `<article>` joins the container tags — the design puts the grid *on* the article for news rows, so those never registered as a `column_set` at all. `column_set` now records its variant **and** its chrome, so the client doesn't guess. Card fields are read in **order** (a short run before the heading is the mono slot; prose after it is the body) — reading purely by type put the tag in the body and left the real body on the floor. ### Guards worth keeping The card renderer refuses a column carrying inputs, a terminal, a nested grid, or a long repeated run. Without them the **Contact form rendered as a card**, and the **ritual demo's step list and signed history were dropped** past the first paragraph. Both were caught by measuring rendered output, not by reading code. ### Result **98% of design copy survives to rendered HTML** — landing and rituals 100%, spelwork 98%, news 94%, contact 92% (the remainder is a handful of contractions). Every page emits toilville blocks. Contact renders the real `wish-form` with Name, Email, Organization, Interested in, and the custom message label. `just check` green, **270 tests pass**. ### Still not wired `toilville/ritual-demo` and `toilville/intent-console` exist (toilville_rituals#368) but nothing emits them. Doing that *properly* means annotating the design source so the demo's steps become `status` blocks and its history becomes `lineage_entry` — the converter can't produce `lineage_entry` at all today. That is the vocabulary-first path and it is the next piece, rather than another shape-matcher.

peterswimm referenced this pull request 2026-09-10 14:47:33 +00:00

feat(udg): code primitive + client boundary (fixes toilville blocks leaking into 4 brands) #111

Sign in to join this conversation.

Reviewers

No reviewers

Labels

Clear labels   

priority

P1

Urgent — no working icon yet (custom art pending) or a shipping asset

  

priority

P2

High — has an accessible stand-in; bespoke art wanted

  

priority

P3

Medium — stage visual headers

  

priority

P4

Low — behavioral dimension family

  

bug

Something is not working

  

duplicate

This issue or pull request already exists

  

enhancement

New feature

  

help wanted

Need some help

  

invalid

Something is wrong

  

question

More information is needed

  

wontfix

This won't be fixed

No labels

priority

P1

priority

P2

priority

P3

priority

P4

bug

duplicate

enhancement

help wanted

invalid

question

wontfix

Milestone

Clear milestone

No items

No milestone

Projects

Clear projects

No items

No project

Assignees

Clear assignees

No assignees

1 participant

Notifications

Subscribe

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference

Toilville/toilville-design!110

No description provided.