Answer stability round 1: 0 🔴, 3 🟡; pin the blank line's spaces in an item's code block, drop the timing-only tests, and file tab-indented lines as item 120

docs/decisions.md

@@ -828,8 +828,9 @@ changed, and a before-and-after figure in its PR; the gate times nothing. Measur
 `adfDocumentFault`'s shape and depth walks stay two — the parting gives depth its own code — at
 52 ms for a 9 MB document the emit takes 314 ms over. The 500-level guard runs after the block walk,
 so it bounds nothing in it: `matchBlankLine` continues a blank line through the open list items in
-one step, and an item reads a line's indentation only up to its own width, which took 20000 nested
-items and as many blank lines from 10.7 s to 39 ms.
+one step, which took 20000 nested items and as many blank lines from 10.7 s to 39 ms; and an item
+reads a line's indentation only up to its own width, though a tab an item's width splits still
+costs the whole line at each level (`todo.md` item 120).
 
 ## Only the hard break holds a raw newline
 

src/markdown/parse/blocks.test.ts

@@ -123,7 +123,7 @@ test("releases an indented code block's held blank lines as a walk rather than a
   assert.deepEqual(kinds(`    a\n${'\n'.repeat(200000)}    b\n`), ['code'])
 })
 
-// A level is one list item. The gate times nothing; a quadratic walk outlasts Bun's 5 s test timeout.
+// A level is one list item.
 function deepestItem(markdown: string): { depth: number; kinds: string[] } {
   let blocks = parseBlocks(markdown, flavourClaims.lossless).blocks
   let depth = 0
@@ -134,19 +134,21 @@ function deepestItem(markdown: string): { depth: number; kinds: string[] } {
   return { depth, kinds: blocks.map((block) => block.kind) }
 }
 
-test('continues a blank line through a deep run of list items in time linear in the lines', () => {
-  const depth = 20000
+test('continues a blank line and an indented line through list items nested past the 500 levels the parser carries', () => {
+  const depth = 1000
   assert.deepEqual(deepestItem(`${'- '.repeat(depth)}a\n${'\n'.repeat(depth)}`), { depth, kinds: ['paragraph'] })
   assert.deepEqual(deepestItem(`${'- '.repeat(depth)}a\n${'\n'.repeat(depth)}${'  '.repeat(depth)}b\n`), { depth, kinds: ['paragraph', 'paragraph'] })
+  assert.deepEqual(deepestItem(`${'- '.repeat(depth)}a\n${'  '.repeat(depth)}b\n`), { depth, kinds: ['paragraph'] })
 })
 
-test('stops a blank line at the first blockquote and at an empty item', () => {
+test('stops a blank line at the first blockquote', () => {
   assert.deepEqual(deepestItem('- > a\n\n  b\n'), { depth: 1, kinds: ['blockquote', 'paragraph'] })
-  assert.deepEqual(deepestItem('- - \n\n  a\n'), { depth: 2, kinds: [] })
   assert.deepEqual(deepestItem('- > - a\n\n  > b\n'), { depth: 1, kinds: ['blockquote', 'blockquote'] })
 })
 
-test('continues a line through a deep run of list items in time linear in the line', () => {
-  const depth = 20000
-  assert.deepEqual(deepestItem(`${'- '.repeat(depth)}a\n${`${'  '.repeat(depth)}b\n`.repeat(3)}`), { depth, kinds: ['paragraph'] })
+test('stops a blank line at an empty item, and gives a code block in an item the blank line without its spaces', () => {
+  assert.deepEqual(deepestItem('- - \n\n  a\n'), { depth: 2, kinds: [] })
+  assert.deepEqual(parseBlocks('- ```\n      \n  x\n  ```\n', flavourClaims.lossless).blocks.map((block) => block.kind === 'bulletList' && block.items[0]?.[0]), [
+    { kind: 'code', language: '', position: { line: 1, offset: 0 }, text: '\nx' },
+  ])
 })

todo.md

@@ -6,12 +6,13 @@
 
 `Bar = 0.1`
 
-`Next ID = 120`
+`Next ID = 121`
 
 ## Items
 
 | ID | Release | Blocked by | Exempt | Item | L | E | I | R | S | Moves | Score |
 |---|---|---|---|---|---|---|---|---|---|---|---|
+| 120 | 0.3.0 |  | defect | **Read a tab-indented line under nested list items in time linear in the line, where today a tab an item's width splits copies the whole line at each level.** | 2 | 1 | 2 | 3 | 3 | 9 by 2 | 6.0 |
 | 115 | 0.3.0 |  | defect | **Build text holding no `border` mark in `losslessMarkdownToAdf`, and have `adfToLosslessMarkdown` carry text holding one.** | 2 | 1 | 2 | 3 | 3 | 4 by 3 | 4.5 |
 | 100 | 0.3.0 |  | decision | **Put the reading of an image's title as its caption back to a reader panel, now that a writer panel reads a title as hover text.** | 1 | 2 | 2 | 3 | 3 | 6 by 4 | 4.0 |
 | 33 | 0.3.0 |  | defect | **Emit a line in time linear in its mark runs, in `adfToLosslessMarkdown` and `adfToPortableMarkdown`.** | 3 | 1 | 2 | 4 | 5 | 9 by 3 | 3.6 |
@@ -23,7 +24,7 @@
 | 102 | 0.3.0 |  | decision | **Place a block a portable task list refuses at its path in the returned document without replaying `splitTaskList` on stand-in nodes.** | 2 | 1 | 2 | 3 | 4 | 10 by 4 | 2.0 |
 | 59 | 0.3.0 |  | decision | **Group the directive grammar into `src/markdown/directive/`, move `Read<T>` to `result.ts`, and move `Flavour` and `flavourClaims` to `markdown/flavour.ts`.** | 3 | 1 | 2 | 4 | 5 | 10 by 5 | 1.5 |
 | 62 | 0.3.0 |  | decision | **Move the portable flavour's reading out of `parse/` and its writing out of `emit/` into `markdown/portable/`, so `emit/` no longer imports `portable/`.** | 3 | 1 | 2 | 4 | 5 | 10 by 5 | 1.5 |
-| 73 | 0.3.0 | 33, 59, 60, 62, 91, 92, 95, 100, 102, 109, 115 |  | **Announce the package where someone needing an ADF converter already reads: JRACLOUD-77436, the Atlassian developer community and Stack Overflow's ADF-to-markdown questions.** | 1 | 2 | 2 | 3 | 2 | 3 by 8 | 32.0 |
+| 73 | 0.3.0 | 33, 59, 60, 62, 91, 92, 95, 100, 102, 109, 115, 120 |  | **Announce the package where someone needing an ADF converter already reads: JRACLOUD-77436, the Atlassian developer community and Stack Overflow's ADF-to-markdown questions.** | 1 | 2 | 2 | 3 | 2 | 3 by 8 | 32.0 |
 | 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 | 8.0 |
 | 45 | 0.4.0 |  | defect | **Replace `isAdfDocument` with a reader returning `Result<AdfDocument>`.** | 2 | 1 | 3 | 4 | 3 | 1 by 2 | 6.0 |
 | 6 | 0.4.0 |  | decision | **Specify the HTML dialect.** | 3 | 2 | 1 | 4 | 6 |  | 0.0 |
@@ -57,6 +58,17 @@
 
 ## Details
 
+### 120. Read a tab-indented line under nested list items in time linear in the line, where today a tab an item's width splits copies the whole line at each level.
+
+`removeColumns` in `src/markdown/parse/blocks.ts` returns `' '.repeat(n) + line.text.slice(index)`
+where an item's width splits a tab, a fresh string, so the next level's `slice` copies the whole
+line instead of sharing it. Measured on 2026-10-18 on 20 tab-indented lines under 20000, 40000 and
+80000 nested list items: 176, 460 and 1812 ms, where the same lines indented with spaces took 73, 90
+and 173 ms. Keep the split tab's leftover columns as a number on `Line` beside `column` and `text`,
+so a cut never builds a string. `docs/decisions.md` §Cost fixes are measured, never timed names
+this exception (cites item 120); delete that clause with the fix. Found while reviewing item 98's
+chunk, 2026-10-18.
+
 ### 115. Build text holding no `border` mark in `losslessMarkdownToAdf`, and have `adfToLosslessMarkdown` carry text holding one.
 
 `!adf:border[x]{color="#091e42" size=2}` builds text holding `border`, which