adf-codec/todo.md

todo

Scoring

Score = 100 * V / (2**S + 8)

Bar = 0.1

Next ID = 177

Items

ID	Release	Blocked by	Exempt	Item	L	E	I	R	S	Moves	Score
153	0.4.0		question	Decide whether src/markdown/ root may import emit/, parse/ or portable/, and hold the answer in .oxlintrc.json and §The source parts by ADF and format.	1	2	2	3	2	code health by 3	26.2
175	0.4.0			Delete the previous === undefined branch, the 'a task holds blocks no list item spells' failure and the path arguments of tryTaskList, once a type makes a task list opening with a non-task unrepresentable.	1	1	1	1	2	code health by 2	17.4
176	0.4.0			Delete runEdge's standing === 'moves' give-up, and any code only it needs, once a test or an argument shows runEdge never reaches it, or else pin the case with a hand-worked line.	1	1	2	2	2	code health by 2	17.4
45	0.4.0		defect	Replace isAdfDocument with a reader returning Result<AdfDocument>.	2	1	3	4	3	1 by 2	13.1
127	0.4.0		principle	Hold what entity-references.test.ts, json-text.test.ts and result.test.ts check by asserting outcomes, so none reads source text.	2	2	1	3	3	code health by 2	13.1
159	0.4.0			Delete taskBlocks's branch that writes the task marker as its own paragraph, once a test shows no block task item reaches it.	1	1	1	1	1	code health by 1	10.5
144	0.4.0			Give every container docker-runner.sh and publish.sh start a memory limit and an equal swap limit, sized from the largest a gate leg measurably needs.	1	1	1	1	2	code health by 1	8.7
143	0.4.0		defect	Keep, in adfToPortableMarkdown, the link around a note, so a linked stored image writes every target.	2	1	3	4	3	7 by 1	6.5
169	0.4.0		owner	Investigate kovetskiy/mark in depth and record in a report how its formats differ from ours, where the two projects overlap and how they compare.	1	1	1	1	3	3 by 4	6.3
112	0.4.0			Pad a GFM pipe table's short rows with empty cells, read its body rows holding no pipe, and read a one-column table without leading pipes in portableMarkdownToAdf.	2	1	2	3	3	4 by 4	6.2
80	0.4.0			Name the node's path and the attribute key when not-an-adf-document refuses a node inside the document.	2	1	2	3	3	6 by 4	6.0
171	0.4.0		principle	Assert the facts an error message must carry, never its whole sentence, in markdown-to-adf.test.ts, document.test.ts and html-to-adf.test.ts.	3	1	1	3	6	code health by 3	4.4
173	0.4.0		principle	Move each CommonMark spelling's description out of commonMarkSpellings, beside the emitter's spelling it describes.	2	1	1	2	4	code health by 1	4.4
81	0.4.0			Fail the gate unless the README's error tables list exactly the codes ConvertErrorCode holds.	1	1	1	1	2	3 by 2	4.2
147	0.4.0			Ship adfToPlainText, writing a document's content with no markup, under a README goal for plain text.	2	2	1	3	4	3 by 4	4.2
141	0.4.0		decision	Write strike as HTML in adfToHtml and read it back in htmlToAdf, in place of the carry element.	1	1	1	1	2	6 by 2	4.0
135	0.4.0			Write media, media groups and captions as semantic HTML in adfToHtml and read them back in htmlToAdf, in place of the carry element.	2	2	2	4	4	6 by 2	2.0
136	0.4.0			Write underline, text and background colours, subscript, superscript, alignment and indentation as HTML in adfToHtml and read them back in htmlToAdf, in place of the carry element.	2	2	2	4	4	6 by 2	2.0
134	0.4.0		decision	Write panels, expands, task lists and decision lists as semantic HTML in adfToHtml and read them back in htmlToAdf, in place of the carry element.	2	2	2	4	5	6 by 3	1.8
46	0.4.0	134, 135, 136		Publish the bundle size in the README, failing the release pipeline when it drifts.	2	2	1	3	4	3 by 3	3.2
174	0.4.0		defect	Bound the passes trimmedEdges takes by a constant, so a line of alternating code and emphasized blanks trims in linear time.	3	1	2	4	7	9 by 3	2.3
51	0.4.0			Match a reference label to its definition under Unicode case folding.	2	1	2	3	4	4 by 2	2.1
148	0.4.0			Ship conversion between ADF and Jira wiki markup both ways, under a README goal for wiki markup.	3	2	2	5	7	3 by 5	0.9
149	0.4.0			Read any HTML into ADF, in a call or option a developer panel picks, under a README goal for foreign HTML.	3	2	2	5	7	3 by 5	0.9
172	0.4.0		principle	Build one test per reader or case in place of a loop inside a test body, in the 26 test files that loop.	3	1	1	3	8	code health by 2	0.8
77	0.4.0		decision	Read the input once into a plain copy, so an accessor property or a throwing Proxy yields not-an-adf-document and every call returns a result.	3	1	3	5	4		0.0
107	0.4.0		decision	Spell the block carry's JSON so its length stays linear in the node's JSON.	3	1	3	5	5		0.0
170	0.4.0		owner	Bundle the packed tarball with webpack, esbuild, Rollup and Bun in the gate, failing on any bundler warning, and run each bundle's round-trip in Node.	3	1	1	3	5		0.0
42	0.5.0		defect	Trim a text leaf's trailing blanks in linear time.	1	1	2	2	2	9 by 3	26.2
31	0.5.0			Make the branch-coverage figure repeat across runs of an unchanged tree.	2	1	1	2	3	code health by 3	19.6
57	0.5.0			Make each ci.sh leg build what it reads, so one leg run alone tests the current tree.	2	1	1	2	3	code health by 3	19.6
38	0.5.0		defect	Spell a lone surrogate in a text node so it survives a UTF-8 encode.	1	1	2	2	2	1 by 2	17.4
76	0.5.0			State the files the package does not ship once, so the build, the lint ceiling, the coverage exclusions and the mutation run cannot drift apart.	1	1	1	1	2	code health by 2	17.4
56	0.5.0			Give each piece of blocks.ts's block-walk state and inline-content.ts's Scan one owner that returns what it changes.	3	1	2	4	5	code health by 5	13.1
118	0.5.0			Spell a portable link's title through oneLine.	1	1	1	1	1	code health by 1	10.5
82	0.5.0			Give the directive tests and fixtures random UUID v7 ids in place of a-1, a, x and h.	2	1	1	2	3	code health by 1	6.5
58	0.5.0			Port ci.sh, publish.sh, docker-runner.sh and mutation-tests/run.sh to standalone Python scripts.	4	2	2	6	6	code health by 3	4.4
9	0.5.0			Ship an online sandbox: a web page with two textboxes converting between ADF and markdown on the library's browser build.	3	2	1	4	6	3 by 6	2.1
34	0.5.0			Read emphasis flanking by the whole character beside an astral symbol.	2	1	2	3	3	4 by 1	1.5
74	0.5.0		owner	Keep the release path publishing after npm retires bypass-2FA tokens.	3	2	4	7	4		0.0
8	0.6.0		owner	Ship a CLI.	3	2	1	4	7		0.0
139	0.7.0		decision	Ship the calls between markdown and HTML, composed through ADF and each named by its flavour.	2	1	2	3	4	3 by 3	3.2
137	0.7.0		decision	Read foreign HTML in htmlToAdf the three ways docs/decisions.md §Foreign HTML sorts three ways states.	2	1	2	3	4	4 by 2	2.1
138	0.7.0	137		Read raw HTML in every markdown reader through htmlToAdf's foreign-element mapping, in place of refusing it as unmappable-html.	3	1	3	5	5	4 by 4	2.5

Renamed

• 0.2.1 renamed to 0.3.0

Details

153. Decide whether src/markdown/ root may import emit/, parse/ or portable/, and hold the answer in .oxlintrc.json and §The source parts by ADF and format.

The writers read each escaped line back through src/markdown/inline-content.ts's scan. A writer question put back on that scan's path, such as the opening-link consult parseInlineContent takes as an argument today, would close the cycle writer → reader → writer and recurse without bound on a paragraph opening with a spellable !adf:link (§Nothing recurses unbounded). The override for src/markdown/** bans only html/; banning emit/, parse/ and portable/ for the root's own files, with its ci.sh probe, extends the decision.

175. Delete the previous === undefined branch, the 'a task holds blocks no list item spells' failure and the path arguments of tryTaskList, once a type makes a task list opening with a non-task unrepresentable.

reduceToPortable leaves every task list opening with a task. So adfToPortableMarkdown reaches neither the previous === undefined branch of tryTaskList (portable/block-spelling.ts) nor its failure. Deleting the branch made two panel seats rule the code harder to read: a block pushed before the first task would vanish silently. Delete it only together with a type that makes a task list opening with a non-task unrepresentable. Mutants of the path arguments to taskBlocks and placedBlock survived, so no test shows they reach an error. The headroom is not a guard: a list that cannot be a markdown list refuses by it, and a test pins that.

176. Delete runEdge's standing === 'moves' give-up, and any code only it needs, once a test or an argument shows runEdge never reaches it, or else pin the case with a hand-worked line.

In src/markdown/emit/unspellable-runs.ts, Stryker's mutants of runEdge's if (standing === 'moves') return undefined survive the hand-worked tests, and 50,000 fuzz runs comparing the code with the mutant found no output differ. Found on PR 145, 2026-11-09. The comprehension seats found runEdge hard to read (larv-review places); make it easier.

45. Replace isAdfDocument with a reader returning Result<AdfDocument>.

Lands with item 77: the reader returns item 77's copy. Goal 1 has every call return a result; the boolean guard is the one export that does not, and it cannot say which branch refused, where not-an-adf-document's message already does. Breaking: MIGRATION.md shows the guard's replacement.

127. Hold what entity-references.test.ts, json-text.test.ts and result.test.ts check by asserting outcomes, so none reads source text.

Technical principle "Assert the outcome", which bans scanning the source with ast or a regex. Three tests read source text:

• src/markdown/commonmark/entity-references.test.ts regex-extracts the packedReferences literal from entity-references.ts, to guard that each of the 2125 packed references parts into one name and one value the table decodes. Reformatting the literal fails it with the table unchanged.
• src/json-text.test.ts greps every shipped file for JSON.parse(, to guard that only parseJsonText reads JSON text. JSON.parse ( or JSON['parse']( escapes it.
• src/result.test.ts regex-parses the ConvertErrorCode union out of result.ts, and each failure('…' and code: '…' out of every shipped file, to guard that every declared code has a production cause and every call site names a declared code. A code held in a variable or passed through another helper escapes it.

Item 81 changes result.test.ts too. Found by the technical-principles audit of item 60, 2026-10-22.

159. Delete taskBlocks's branch that writes the task marker as its own paragraph, once a test shows no block task item reaches it.

taskBlocks (portable/block-spelling.ts) writes the marker as a paragraph of its own where a block task item's content opens with another block. Mutation testing finds that branch unreached: a block task item is left only when it holds two or more paragraphs. Found by the stability review of item 33, 2026-10-29.

144. Give every container docker-runner.sh and publish.sh start a memory limit and an equal swap limit, sized from the largest a gate leg measurably needs.

in_image, with_firefox and publish.sh run docker run with neither --memory nor --memory-swap, so a runaway leg can starve the machine instead of failing alone. Some tests build strings near V8's longest, about 512 MB each, so measure each leg's peak before picking the limit: one too small fails the gate.

143. Keep, in adfToPortableMarkdown, the link around a note, so a linked stored image writes every target.

noteLeaves (src/markdown/portable/inline-reduction.ts) drops the node's marks, so a stored image a link marks writes _(image not included)_ without the link's target, which Goal 7 counts as content. adfToHtml's carry element keeps it, <a href="…"><em>(image)</em></a>.

Move the choice of where a note's link goes into src/adf/shown-text.ts, beside shownImage's for an image, so both writers render one answer.

Leave what an HTML carry element shows unchanged, or ship the change as a breaking HTML change with a MIGRATION.md entry: htmlToAdf refuses any other content, so HTML stored by an earlier release would no longer read (docs/decisions.md §The formats are API).

169. Investigate kovetskiy/mark in depth and record in a report how its formats differ from ours, where the two projects overlap and how they compare.

https://github.com/kovetskiy/mark publishes markdown files as Confluence pages. Read its README, source and issues, and run its conversion on the corpus's documents where that works with no Confluence account; say which parts were only read. Write the report as one page under docs/ answering:

• Which format does it write: ADF, or Confluence's storage format? Can it read a page back?
• How does it spell the Confluence features we spell as directives, alerts and callouts: macros, panels, includes, labels, attachments and page metadata? Where do its spellings and ours differ, and which would the README's ## Audience expect?
• Which personas overlap, and what does a user of one need from the other?
• Does it belong in the README's "Other ADF converters" table? docs/decisions.md §The comparison table lists every converter that starts and converts covers npm packages, and this is a Go program. If it is not ADF at all, that is a question for the maintainer.

The report changes no spelling: one is changed only by a writer panel (AGENTS.md §7). A finding that asks for a change is filed as its own item.

112. Pad a GFM pipe table's short rows with empty cells, read its body rows holding no pipe, and read a one-column table without leading pipes in portableMarkdownToAdf.

GFM pads a short row with empty cells, reads a line holding no | after a table as a row, and reads a | over --- | as a one-column table. portableMarkdownToAdf refuses a short row as malformed-pipe-table, ends the table at a line holding no |, and reads the one-column shape as a paragraph. A long row stays refused (docs/decisions.md §The portable reader refuses what GFM's reading would drop), so the README's pipe-table guarantee keeps the long row for both readers, and the short row and the line holding no | for losslessMarkdownToAdf alone. The one-column shape needs isBareDelimiterRow in the writer's escaping to cover it too.

80. Name the node's path and the attribute key when not-an-adf-document refuses a node inside the document.

Lands with items 45 and 77, whose reader walks every node. Today every refused node, such as one with a Date in an attribute, reads an ADF document's content holds ADF nodes: one of them is not with an empty path. An export tool told this about one document in thousands then searches that document's whole tree. Give the node's path, and name its type and the attribute key as attributeNestingMessage does. Found by the product-owner review of items 66 and 67, 2026-10-05.

171. Assert the facts an error message must carry, never its whole sentence, in markdown-to-adf.test.ts, document.test.ts and html-to-adf.test.ts.

README → Audience says no persona may rely on a message's wording, yet content() in src/markdown/parse/markdown-to-adf.test.ts folds a failure into ${code}: ${message}, about 103 assertions compare that whole sentence, and document.test.ts (7) and html-to-adf.test.ts (3) pin message: the same way. A rewording fails tests with no behaviour changed, while the facts §message and path requires (the violation, the escape named) are never asserted on their own. Assert code, path and position exactly, and match message against the facts it must carry. Technical principle: "Pin wording only where the wording is the contract". Found by the principles audit of the release review of 0.3.0, 2026-11-08.

173. Move each CommonMark spelling's description out of commonMarkSpellings, beside the emitter's spelling it describes.

commonMarkSpellings (src/markdown/parse/markdown-to-adf.ts) restates in prose the spellings spellReadableBlock (src/markdown/emit/adf-to-markdown.ts) owns, such as "each item behind - , or behind * where …". A changed spelling leaves the refusal naming the old one, and a node missing from the map is refused naming no spelling. Put each description beside the spelling and export it through emit/spelling-questions.ts, the one import parse/ may take from emit/, and drop the comment at markdown-to-adf.ts:145 that keeps them apart. Technical principle: DRY. Found by the principles audit of the release review of 0.3.0, 2026-11-08.

81. Fail the gate unless the README's error tables list exactly the codes ConvertErrorCode holds.

src/result.test.ts checks the union against the call sites only, so a code added or removed leaves the README's tables a false contract under a green gate. Assert that the tables list exactly the union. Found by the technical-principles audit of item 78, 2026-10-05.

147. Ship adfToPlainText, writing a document's content with no markup, under a README goal for plain text.

Asked on the Atlassian developer community for search indexing, notifications, emails and CSV exports, where markdown's markers are noise, and with a mention as @Display Name rather than an account id. Add the goal to the README's ## Goals in the same change. Split into items before starting. The maintainer, 2026-10-28: ship what exists first, so this waits for 0.4.0.

Once it ships, answer developer-community topic 41858, which asks for ADF as plain text (the maintainer, 2026-10-28).

141. Write strike as HTML in adfToHtml and read it back in htmlToAdf, in place of the carry element.

GFM spells strike, and no other HTML item writes it: adfToHtml writes only the marks CommonMark spells. A writer panel picks <s> or <del>. A reader panel (docs/decisions.md, the feature-table entry) expected <s> or <del>, and the writer panel takes that as input.

135. Write media, media groups and captions as semantic HTML in adfToHtml and read them back in htmlToAdf, in place of the carry element.

A writer panel picks the spellings; a plain external image already has one, <p><img …></p>. Atlassian's editor puts localId, width and widthType on every mediaSingle, so every editor-written image rides the carry until this item.

136. Write underline, text and background colours, subscript, superscript, alignment and indentation as HTML in adfToHtml and read them back in htmlToAdf, in place of the carry element.

A writer panel picks the spellings.

134. Write panels, expands, task lists and decision lists as semantic HTML in adfToHtml and read them back in htmlToAdf, in place of the carry element.

An expand is <details><summary>, as htmlToAdf reads it (docs/decisions.md §Foreign HTML sorts three ways). A writer panel picks the rest. A reader panel (docs/decisions.md, the feature-table entry) expected a panel as <div class="adf-panel" data-panel-type="…"> and a task list as <ul class="adf-task-list"> of <li data-state="DONE|TODO"> with a checkbox input, and the writer panel takes these as input.

46. Publish the bundle size in the README, failing the release pipeline when it drifts.

The HTML items change the size. The quantity is what a consumer downloads and loads: the tarball npm pack produces, its unpacked dist, and the built JavaScript minified + gzipped — the figure the competitors advertise (marklassian's "12kb") and the only apples-to-apples one, since ours ships tsc's unminified output and no minifier yet (decide here whether to minify for the build or report the unminified gzip). The figure lands in README §The package beside the "no runtime dependencies" claim, and in a size column of the comparison table. Measured today, unminified: tarball 60.4 kB, unpacked 221.5 kB, JS gzipped 45.6 kB.

174. Bound the passes trimmedEdges takes by a constant, so a line of alternating code and emphasized blanks trims in linear time.

trimmedEdges (src/markdown/portable/inline-reduction.ts) repeats a pass until nothing changes, and each pass costs O(leaves) while leaf text is short; the trailing-blank scan inside one long leaf is item 42's. Leaves alternating ' ' marked code with ' ' marked em take ceil(n/4) + 1 passes: 26, 251 and 2 501 at n = 100, 1 000 and 10 000, so the trim is quadratic, against goal 9. The same holds for 'x' code beside ' ' em, and for [link, code] 'x' beside [link, em] ' '. The smallest line is six leaves ' ' marked code, em, code, em, code, em, which takes 3 passes. The passes in a random search of 20 000 short lines never went above 4.

The cause: under spanned grouping an em or link run crosses the code leaves, which can't hold it, so one run spans the line and only its two end leaves stand beside a neighbour outside it. Each pass strips em from those two, and the next pair is exposed. A single pass cannot take the fixpoint directly: every pass judges all leaves against one snapshot, and edgeOutcome (drop at a line edge while a run is shared, keep when not) isn't monotone. A hard break between [link, em] ' ' and [link, em] ' b' trims to b today, but to a linked space and b if the first leaf's drop is seen first. Two ways forward: make the edge rule monotone, which changes output in edge cases and needs a decision and a writer panel; keep run extents under truncation in runsSharedWithNext (emit/inline-runs.ts), exact but large. Count passes to measure, never time (§Cost fixes are measured, never timed). Found by the principles audit of the release review of 0.3.0, 2026-11-08.

51. Match a reference label to its definition under Unicode case folding.

link-syntax.ts normalizes a label with toLowerCase, so [ẞ] misses its [SS] definition (spec example 540). Fold with Unicode's full case folding, CaseFolding.txt's C and F mappings, shipped as data per §Standards ship as data: lowercasing and then uppercasing would also merge ı with I, which folding keeps apart, so add that pair as a fixture. Breaking: MIGRATION.md's Readings table gains its row. Its pending exceptions go, and the README's guarantees drop the case-folding exception and count one fewer.

148. Ship conversion between ADF and Jira wiki markup both ways, under a README goal for wiki markup.

Jira's REST API v2 and its changelogs carry wiki markup, and Atlassian's WikiMarkupTransformer is the only converter; community threads report its round-trip losing expands, layouts and some marks. Add the goal to the README's ## Goals in the same change. Split into items before starting. The maintainer, 2026-10-28: ship what exists first, so this waits for 0.4.0.

149. Read any HTML into ADF, in a call or option a developer panel picks, under a README goal for foreign HTML.

htmlToAdf reads only what adfToHtml writes. The most asked need on the Atlassian developer community is HTML from emails, CMSs and editors into ADF. The maintainer, 2026-10-28: reading foreign HTML is a separate goal, added to the README in this change. Before building, a panel of the developer personas picks the API among those the goals allow: a parameter on htmlToAdf or a separate call; Goal 8's "nothing to configure" and Goal 6.1's naming bear on the choice. Split into items once the API is picked.

172. Build one test per reader or case in place of a loop inside a test body, in the 26 test files that loop.

134 loops run several readers or cases inside one test body, for example for (const read of [commonMarkToAdf, losslessMarkdownToAdf, portableMarkdownToAdf]) at lines 97, 279, 771, 917, 926, 962 and 1118 of src/markdown/parse/markdown-to-adf.test.ts; the most are in adf-schema.test.ts (18), limits.test.ts (12) and portable-markdown-to-adf.test.ts (10). The first failing reader stops the test and the failure never names the reader. Write for (const read of readers) test(`… in ${read.name}`, …) at the top level. Technical principle: "No loops or branches in a test body; parametrize the cases". Found by the principles audit of the release review of 0.3.0, 2026-11-08.

77. Read the input once into a plain copy, so an accessor property or a throwing Proxy yields not-an-adf-document and every call returns a result.

Lands with item 45, in one chunk: item 45's reader returns the copy. The maintainer's call, 2026-10-04: Goal 1 promises a result for any value, getters and Proxies included. Today the guards and the conversions read the caller's object many times, so a throwing getter, a revoked Proxy or a getter minting a fresh child per access throws or walks forever. Item 45's reader copies the document once inside one tight try/catch, reading property descriptors so no getter runs: an accessor property, or a Proxy trap that throws, is not-an-adf-document. The copy keeps own enumerable keys, as JSON.stringify does, so an inherited or non-enumerable key drops. The reader reads arrays by index, so no custom iterator runs. Every later step works on the copy. A Proxy can still claim unbounded width, which the 500-level limit does not cover, so the copy reads at most 10 million JSON values, each object, array, string, number, boolean and null counting one (the maintainer, 2026-10-05): roughly 50 to 100 MB of JSON, far past any real page, where 10 000 would refuse a 500-row table. Before reading an array, the reader checks its length against the values left, so a Proxy claiming 2³² elements is refused at once. The chunk asks whether that refusal shares the 500-level guard's unsupported-nesting-depth or takes another code. Found by the technical-principles audit, 2026-10-04.

107. Spell the block carry's JSON so its length stays linear in the node's JSON.

The block carry indents two spaces a level, so about 1 MB of deeply nested arrays in a carried node's JSON spells past V8's longest string, which the writers refuse as unsupported-node-shape; the inline carry's compact JSON would fit. docs/decisions.md §The code list calls a refusal no spelling recovers from a gap in the flavour. Respelling the carry changes the format (§The formats are API). Found by the README-goals audit of item 83, 2026-10-11.

170. Bundle the packed tarball with webpack, esbuild, Rollup and Bun in the gate, failing on any bundler warning, and run each bundle's round-trip in Node.

Consumers bundle the package, Forge apps with webpack among them, and one reported a "Critical dependency" warning from another converter. Nothing in the gate bundles it today. Install the tarball as package-tests does, bundle an entry file that round-trips one document through adfToLosslessMarkdown and losslessMarkdownToAdf, and run each bundle in Node. A warning fails the leg as an error does: webpack's "Critical dependency" is a warning. The bundlers live in their own package, as Stryker does (§Stryker in its own package), with pinned versions. Bun builds from the image the gate already pulls. No leg needs an account or a network beyond the install.

Once the leg holds, the README may say the package bundles with those four; it says nothing about Forge's runtime, which no leg tests. Item 46 may take its bundle size from this leg.

42. Trim a text leaf's trailing blanks in linear time.

portable/inline-reduction.ts's leafEdges finds the trail with an unanchored /[ \t]*$/, quadratic in a run of blanks inside one leaf: a paragraph of a, 80 000 spaces, b takes 6.5 s in adfToPortableMarkdown. Scan backward, as the expand title's trim does.

31. Make the branch-coverage figure repeat across runs of an unchanged tree.

Three Node test legs over one unchanged tree reported emit/inline-line.ts at 95.83%, 96.23% and 96.23%, and the total at 98.80%, 98.84% and 98.84% (2026-09-21). --experimental-test-coverage counts branches off V8's own coverage, which the runner's parallel files and V8's optimization make run-dependent, so the number the floor is read against is not the code's alone. The floor of 98 holds today on 0.8 points of slack and docs/decisions.md §The coverage floors says it only ever moves upward, so the first raise to the measured figure reddens a run that changed nothing. Make the measurement repeatable, or state the number the floor may be raised to and why it is not the measured one.

57. Make each ci.sh leg build what it reads, so one leg run alone tests the current tree.

Technical principle "Compose, do not entangle": Deno, Bun and the tests read the install leg's node_modules, the floor and consumer legs the pack leg's, and the browser leg whatever dist/ the build leg left, so a leg run alone can pass on stale output.

38. Spell a lone surrogate in a text node so it survives a UTF-8 encode.

adfToLosslessMarkdown emits it verbatim, so markdown stored as UTF-8 reads back U+FFFD; attribute values already escape it.

76. State the files the package does not ship once, so the build, the lint ceiling, the coverage exclusions and the mutation run cannot drift apart.

Technical principle DRY: the set is written in tsconfig.build.json's exclude, .oxlintrc.json's ignorePatterns and the test script's two --test-coverage-exclude flags, and §The size ratchet assumes they agree. A helper added to one list lints test code under the shipped ceiling, or drops shipped code from coverage, silently. Derive the others from one, or assert in a test that they agree. Found by the technical-principles audit, 2026-10-04.

mutation-tests/run.sh's pathspec restates the test files and test helpers, and mutation-tests/stryker.config.mjs's testFiles the test files, so a helper missing from the pathspec has its mutants read as survivors. Found by the architecture review of item 152, 2026-10-28.

56. Give each piece of blocks.ts's block-walk state and inline-content.ts's Scan one owner that returns what it changes.

Technical principle "One owner per value": the Walk record passes through functions that mutate it and return void; walk.leaf alone is written in seven places, and openedList reads the endedLists that closeLeaf fills. ContainerStack in the same file shows the shape to follow. inline-content.ts's Scan has the same shape: about fifteen functions write pending, pieces, brackets, linkOpenersFrom and openingSpellableLink on the Scan they are handed, most of them returning void, and parseInlineContent reads a flag scanInline leaves on it. The same shape recurs in mintTaskIds (portable/task-ids.ts), which writes attrs on the document readDocument built; takeFallback (emit/inline-line.ts), which writes the record the next lineSegments pass reads; writeUnpaired and markHighlights (inline-content.ts), whose order changes the output. Once Scan's pieces have owners, the bracket resolution — linkedAlt through linkTo, holdsLink, and closeBracket through altText — can leave inline-content.ts (about 690 lines) for a module of its own. That module resolves an image once, into a node holding its alt text and its link, so imageAlt, resolvedPieces and directiveMarkPieces no longer filter and regroup ImageLink pieces. Found by the architecture pass of the release review of 0.3.0, 2026-11-08.

118. Spell a portable link's title through oneLine.

portableLink (src/markdown/portable/inline-reduction.ts) repeats oneLine's replacements by hand, so a change to what a portable line drops would miss link titles. Technical principle: DRY. Found by the principles audit of item 99, 2026-10-14.

82. Give the directive tests and fixtures random UUID v7 ids in place of a-1, a, x and h.

Technical principle: tests use random UUID v7 ids; portable-markdown-to-adf.test.ts already does. Found by the technical-principles audit of item 78, 2026-10-05.

58. Port ci.sh, publish.sh, docker-runner.sh and mutation-tests/run.sh to standalone Python scripts.

Technical principle "Prefer a standalone Python script over a shell script": AGENTS.md §3 teaches bash footguns (&& chaining under ||, tee /dev/stderr) the port removes. A wrong line range in run.sh mutates the wrong lines without failing, so the port tests how it reads diff hunks.

34. Read emphasis flanking by the whole character beside an astral symbol.

Check whether delimiterRunToken (inline-tokens.ts) and resolveEmphasis (line-escaping.ts) read one UTF-16 unit beside an astral symbol. A lone surrogate is neither punctuation nor symbol, but CommonMark reads 😀 as punctuation. Where they do, read the code point, with a fixture per direction.

74. Keep the release path publishing after npm retires bypass-2FA tokens.

Lands after 2027-01-01, or after a release run fails on the token, whichever comes first: the maintainer chose on 2026-10-02 to wait and see whether the retirement bites. It holds back no release: when the rest of its release is done, it moves to the next. 0.1.0 published only once the npm token carried Bypass 2FA: the account requiring no 2FA on writes was not enough, and npm answered EOTP until the token itself bypassed. npm retires bypass-2FA tokens for direct publishing around January 2027, leaving them npm stage publish, which a maintainer approves with 2FA. Trusted publishing over OIDC replaces the token, and GitHub-hosted Actions support it. The release has run there since 2026-10-04, still with the organization's NPM_TOKEN: the maintainer chose to keep the token for now. The trade is the maintainer's: move publish.sh to trusted publishing, or to the staged publish, which fits badly with publish-on-merge. Its Moves cell is empty: no README goal covers the release path.

8. Ship a CLI.

Its Moves cell is empty: no README goal or persona covers a CLI yet. The chunk proposes both (the maintainer, 2026-10-02), and they land in the README's ## Goals and ## Audience with the CLI.

139. Ship the calls between markdown and HTML, composed through ADF and each named by its flavour.

Waits until a consumer needs it. Which flavours get a composed call is this item's to settle; the CommonMark spec suite runs against the one reading CommonMark, checked against the spec's expected HTML. Where a composed call naming a markdown Flavour sits is this item's to settle.

137. Read foreign HTML in htmlToAdf the three ways docs/decisions.md §Foreign HTML sorts three ways states.

Waits until a consumer needs htmlToAdf to read HTML that adfToHtml did not write.

138. Read raw HTML in every markdown reader through htmlToAdf's foreign-element mapping, in place of refusing it as unmappable-html.

Waits until a consumer needs it. The change touches inline-content.ts and markdown-to-adf.ts, places the comprehension seats found hard; leave each place it touches easier to read.