# Backlog - what this stage found and deliberately did not fix Opened at the molecule round of step 5, earlier than the pack schedules it, for one reason: the findings were arriving faster than the steps that own them, and a finding with no home is a finding that gets lost. Stages 09 to 13 read this and close what their own material closes. ## Where this file stands **Counted on 2026-08-23 by `design/kit/screens/handoff13.cjs`, section 7, and not by hand: 114 table rows, 94 saying closed, 30 open by the reading that a row with anything unfinished in it is an open row, and 0 addressed to a stage that has ended.** The instrument is the owner of these five numbers from today; everything below is the history of how the file got here, each block keeping the corpus it was counted on. Re-run it rather than trusting this paragraph a month from now. **Four rows landed after that run, on the evening of 2026-08-23, with node 6.17**, and a re-run of the same instrument answers **111 rows, 80 saying closed, 41 open, 3 addressed to stage 13**. The paragraph above keeps the corpus it was measured on, which is the rule; this sentence is the recount, in the step that changed the file. **Plus 1 row on 2026-08-26**, the first one this file has taken from outside its own repository: the CSS pipeline of the product build, which refuses a `url()` it cannot resolve. Re-run of the same instrument in the same step: **115 rows, 94 closed, 31 open, 5 naming the founder as owner and 1 of those still open.** The two paragraphs above keep their own corpora. ## How the count got here, counted on 2026-08-12 after step 9 **54 rows. 20 closed. 34 open. 0 rows dropped at verification.** **Plus 4 rows added on 2026-08-13 by the self-sufficiency test of stage 09, 1 of them closed the same day: 58 rows, 21 closed, 37 open.** They are in their own section below rather than merged into the tables above, because they were found by a different instrument: not an audit of what stands, but the first screen built out of the system after it was declared finished. **Plus 2 rows added the same day by the verification pass of step 6, 1 of them closed on arrival, and 1 row above closed by it: 60 rows, 23 closed, 37 open.** The pass ran on two instruments and produced twelve findings; eleven of them were one defect repeating, so they are carried as one row about the mechanism rather than as eleven about its symptoms. Counted by walking every table below, not estimated: 2 blockers, 4 decisions, 6 holes, 19 from the molecule and organism rounds, 5 still open on the components, 6 from the second control census, 5 owned by Voice, and 7 added today by step 9. The dry-run table is a measurement rather than a queue and its five screen rows are not in the 54. **One row is closed in part and is counted as open**, because a row with anything unfinished in it is not a closed row: `kit.html`, closed on the variables and open on the style block that was scheduled out of the page at step 4. The second such row, the chip tones, closed the rest of the way on 2026-08-13 when step 6 corrected the comment that was its open half. One further row, the button that may not wrap, is marked **corrected** rather than open or closed: the rule stands, the numbers were withdrawn, and it is counted as open. What WAS dropped at verification in this stage never had a row here and is named in the last section instead, so that the next audit does not raise it again. **Every row was re-read against the current files today**, after the source pass by Codex, the browser pass, and the three batches of fixes in `e694d2c`, `2ce7fe5`, `437943b`, `43cc778` and `ce9771d`. A row that only carried a state written weeks ago now carries the state it measures today, and where the two disagree the disagreement is written into the row rather than quietly overwritten. **No row was dropped whole today, and two measurements inside surviving rows were withdrawn.** The 213 against 225 overflow pair was withdrawn on 2026-08-12 and its row survives on the principle. Today the four-atoms row lost three of its four numbers to a re-read: `amount.css` sets four host rules but only two distinct sizes now, `logo.css` sets five host-keyed sizes and not four, and `muted-line.css` sets one and not two. The principle under that row is untouched, so the row stays open and its evidence is corrected in place. **Nothing open here is a defect on a shipped screen.** One was, for part of 2026-08-12, and it is recorded in the section on the second control census with its measurement and its repair rather than deleted: this file exists to be trusted, and a defect that is written down only while it is unfixed teaches the next reader that the file is a to-do list rather than a record. Everything else is either a decision that needs a person, or work whose correct owner is a later step. Each row says which. **Closed rows are kept and marked rather than deleted.** A row that vanishes on the day it is fixed leaves the next reader unable to tell a solved problem from one nobody ever found. --- ## Blockers, and both are now closed | What | State on 2026-08-12 | Who owned it | |---|---|---| | **Nothing in `design/system/` established a container.** `container-type: inline-size` was declared in exactly two places, `wireframes/_wf.css:35` and `design/kit/kit.css:145`, both on `.stage`, which is the reviewer's frame and travels to no build. Three of the product's layouts are container queries: the tab bar's rail at 760, the dashboard head at 900, the groups column set at 1340 | **CLOSED at the organism round, and closed twice over, because the stand needed its own answer.** `base.css:141` declares it on `body`. Not on `.app`, which was the obvious answer: a query matches an ANCESTOR container and never the element itself, so `container-type` on the shell cannot style the shell, and measured, the tab bar went from a 240px rail beside the content to 965px below it. Zero pixels move, checked at 33 widths in 8px steps through all three breakpoint bands before the change was made. **The stand takes the same rule at a different level**: `_page.css:572` puts `container-type: inline-size` plus `contain: layout` on `.kit-stage`, because `body` is right for a build and wrong for a stand, where every specimen was showing its desktop form inside a narrow plate. `design/_screen.css:149` carries the third, on `.stage`, for the review frame around a product screen. **Re-read after step 9 and still closed**: `design/kit/kit.css` no longer exists, so the second of the two original sites went with the file at step 8 | Closed | | **Two placeholders still ship as literal text in colour.** `history-trends-loading.html:67` renders `[chart]` and `history-trends-empty.html:61` renders `[chart: waiting for a third month]` | **CLOSED at step 6, verified in a browser rather than in the diff.** Both now draw the same three-line frame the other two chart screens draw, and the empty state keeps its accessible name ("An empty chart frame, waiting for a third month of history"). Walked at 360 and 1280 on all four History and Trends screens: **no `[chart` string renders on any of them**, and the frames measure 328x168 at 360 and 740x220 at 1280. It also corrects the row on the stage 07 hub which records "3 found, 3 fixed, 0 carried"; that row was wrong and is now true | Closed at step 6 | --- ## Decisions that need a person | What | The options, honestly | Who decides | |---|---|---| | ~~Should a card that cannot be clicked answer a pointer at all? `.plan-opt` is a `div` holding a `button`: it cannot be focused, cannot be pressed and has no accessible role, and yet its edge strengthens under the pointer~~ | **OPEN, narrowed, and measured again today.** The token half is decided: `--line-container-hover` is declared on `--hairline-strong` with a computed dark pair, because dropping a state is a visual change and none of the stage's named sources asked for one, so carrying it was the only move that needed permission from nobody. Measured on `design/upgrade.html` at 1280: the card's edge goes `rgb(228, 233, 234)` to `rgb(205, 215, 217)` under the pointer, and nothing else on the card moves. What is left is the taste question underneath, and it is smaller: a hover that promises a click the card does not deliver. **Step 9 touched nothing here** | **CLOSED 2026-08-20. No.** The hover is gone; `--line-container-hover` had no other reader so it went too, and `--hairline-strong` with its dark pair had no role so they went as well. Three deletions, one decision. Measured before removal on `design/upgrade.html` at 1280: the edge went rgb(228,233,234) to rgb(205,215,217) and nothing else on the card moved | | **Non-text contrast on the remaining surfaces.** Card, panel and divider edges sit at 1.23:1 against the 3:1 of WCAG 1.4.11 | Already decided once, on 2026-08-11: the darker edge is scoped to controls and surfaces keep the hairline, because raising `--line` outlines the whole product and the language was chosen against that. Recorded here so a later audit does not reopen it as new | Closed, kept for the record | | **~~The icon set is optically unbalanced.~~ CLOSED 2026-08-21.** Painted boxes differ by 22 percent and INK by 106 percent: `.ic-save` carries 2.06 times the paint of `.ic-you` in the same tab bar | **OPEN, unchanged.** Measured, not eyeballed, at the icons page, and not re-measured at step 9 either, because nothing in step 9 redrew a glyph: the step's only icons work was accounting for the twelfth mark on `icons.html`, which added a row and no geometry. The repair is redrawing a glyph, which is `design/system/` work and belongs with whoever owns the set | **CLOSED 2026-08-21, and it took two moves, only one of which was a drawing.** The finding opened at 106 per cent of ink spread over four icons and was carried as the founder's for eleven days. **Move one was a deletion:** the Save tab retired the same morning, `.ic-save` went with it, and the spread fell to **81 per cent** with no line redrawn - which is what said the problem was the SET rather than any one mark. **Move two was the founder choosing candidate B** off four drawn and measured options: `.ic-trends` gave up its frame for an axis corner, **23.82 to 14.71 per cent**, and the spread is now **61 per cent** with the house as the heaviest mark. Ink over the set as it ships: home 21.12, alerts 15.62, trends 14.71, you 13.14, rasterised at 40 samples per module of the 24 cell with an alpha floor of 10. **The candidate drawn to be rejected is on record and it is the obvious one:** simply deleting the box takes the chart to 10.60, makes it the lightest thing in the bar and the spread **99 per cent**. The four candidates and their numbers are in `design/kit/screens/icon-candidates.html`. Option E, which also drops the house's door and measures 25 per cent, stays available as one decision if the row ever needs to come tighter | **Closed** | ~~The Trial chip has a rule and no wearer in colour. `chip.css:135` declares `.chip.trial` on `--bg-trial` and `--text-trial`, which closes the census's oldest hole: the variant used to have no CSS rule anywhere. But the one place in colour that says Trial does not wear it~~ | **OPEN, new shape of an old row, and step 9 raised it independently.** Measured on `design/home.html`: the chip on the Peloton row renders `rgb(233, 238, 239)` on `rgb(79, 94, 98)`, the quiet tone, because its markup is `class="chip"`. `chip.css:135` still carries the rule. The trial pair is therefore two semantic roles read by exactly one rule that nothing in colour wears, and one grey wearer on the landing. `pixel-proof.html` carries the same row as "A role that has never rendered", found by the browser pass. Two honest answers: the Home row wears `.chip.trial` and the pair earns its place, or the pair goes the way `.tone-attention` went and a trial is a quiet chip like every other status | **CLOSED 2026-08-20. Retired.** D-Concept: a status is a quiet gray badge and a trial is a status. Re-counted the same day - **zero `trial` in any class attribute on any of the 110 pages**, so this row's claim of a grey wearer was wrong too. `--bg-trial`, `--text-trial` and the four teal primitives under them went with the rule. Zero pixels: the Peloton row already rendered quiet | | ~~The landing has two doors and the second step has one. `index.html` offers "Get started free" (to `path-choice`) and "Sign in" (to `sign-in`), in the header and again at the foot, and `sign-in` leads back with "No account yet? Start here". `path-choice` leads back to neither: a person who taps the wrong door on the landing has the browser's back button and nothing on the screen~~ | **OPEN, raised by the founder on 2026-08-19** ("чуть смущает, что у нас sign in одно, а get started другое") and verified before it was written down: the pair IS deliberate and it is the product's model - value before an account, so "Get started free" starts using Tendd and "Sign in" picks up a list you already have. `voice/docs/microcopy.md` owns all three strings and node 1.6 was added precisely because the product had a door in and a door out of an account nothing in it created. **What is not decided is whether `path-choice` owes a way across.** Two honest answers: it does, and gains one quiet line under the two doors, which is a new product string and therefore Voice's and the founder's; or it does not, because the fork is one tap from the landing and a second exit on the first screen of a chain is the kind of clutter this product refuses. Nothing was changed on either ground | **CLOSED 2026-08-20. It owes one and now has it.** `path-choice` carries one quiet line under the retreat: "Already have an account? Sign in". Under the retreat rather than beside the doors, because it is for the person who is in the wrong place and not an option for the person who is in the right one. Both corpora; the founder's decision is written into both files | --- ## Holes in the system, each with the reason it was not filled | What | State on 2026-08-12 | Who closes it | |---|---|---| | ~~The numbered-steps marker is 26 by 26 and no primitive holds that size. `--tap` and `--row` are the only size primitives and it is neither; a width is not a margin, a padding or a gap, so the 8px grid does not govern it~~ | **OPEN, and step 9 has now run without creating a size scale.** Re-read today: `numbered-steps.css:93` still declares `width: 26px; height: 26px`, and `:86` still carries the reason at the line. `tokens.css:175` and `:176` are still the only two size primitives, `--tap` 44 and `--row` 56. Step 9's geometry triage answered the neighbouring question instead, `--ring`, and refused this one on the same ground the row already gives: a size primitive for one marker is a token nobody else reads. Worth knowing before it is answered: `.steps` stands on **zero** of the 28 coloured screens and on **3** grey pages (`cancel-guide`, `cancel-guide-blocked`, `cancel-guide-no-guide`), counted today, so nothing a person can open today is affected either way | **CLOSED 2026-08-20. No primitive.** One wearer, on 3 coloured screens and 3 grey. A size primitive one file reads is a token nobody else can be wrong about, and the reason already stands at the line | | **The spacing fold moved the step marker's text.** `padding-left: 36px` goes to `--space-40`, which opens the marker-to-text gap from 10px to 14px and shifts every step's text 4px right | **OPEN, and step 8 could not see it, exactly as this row predicted.** The arithmetic re-checked today in the file: `numbered-steps.css:67` reads `var(--space-40)`, `tokens.css:258` declares `--space-40: 40px`, and the marker is 26px, so the gap is 40 minus 26, which is 14 against the old 36 minus 26, which is 10. The pixel comparison at step 8 ran on the 28 coloured screens and `.steps` occurs on none of them, so the difference is absent from all three rounds on `pixel-proof.html` by construction rather than by oversight. It still has to be checked on `design/kit/numbered-steps.html` or on the grey, by hand. This row and the one above it were carried twice in the first draft of this file, because two agents found the same marker independently; they are one row each now, and that two instruments found it is kept, because it is the evidence that the marker is worth a decision rather than a shrug | **CLOSED 2026-08-18, as recorded rather than repaired**, the third row in this file to close that way. It has shipped and it is measured: `padding-left: 40px` against a 26px marker box on `design/cancel-guide.html`, so the marker-to-text gap is exactly the 14px this row predicted. `numbered-steps.css` carries the same sentence at the rule - "THAT MOVES A PIXEL AND IT IS EXPECTED" - so the change is named where it lives. A row kept open for a change that was named, expected and delivered implies an undone fix | | **A seam is two hairlines thick.** The last `li` of a charge list keeps its bottom rule and the `.gate` that follows it inside the same panel draws a top rule | **CLOSED at the founder's review of the components.** `charge-list.css:104` gives `.charges li:last-child { border-bottom: 0 }`, applied as a principle rather than to the seam that exposed it, which is what found the second seam under it. The panel is 1px shorter on the five subscription-detail screens and nothing else moves | Closed | | **~~Eleven ARIA and form-semantics findings from stage 07.~~ CLOSED 2026-08-20, and the count was wrong because the corpus had doubled.** Founder: "давай чинить дыры какие есть". The eleven were measured on 28 coloured pages and the product is 55, so they were not recounted, they were **re-asked**: `design/kit/screens/aria13.cjs` puts six questions to all 110 pages - a fieldset with no name, an `aria-label` on an element ARIA discards it on, a wait that does not declare `aria-busy`, a form control with no accessible name, a link or button with no accessible name, a heading level skipped. | **18 findings in 18 pages, 9 per corpus, and 2 of the 6 questions came back clean on the first run**: zero form controls and zero controls without an accessible name, across 110 pages. The other four: 6 fieldsets with no name (the three states of Add a subscription in both corpora), 8 `p.total` carrying an `aria-label` that ARIA discards because `paragraph` takes no name from the author, 2 waits not declaring `aria-busy` (`home-loading`, the twelfth finding), 2 heading levels skipped (the landing footer, h2 to h4). | **All 18 fixed, all 6 questions now 0.** The fieldsets are named with `aria-label` and not a `legend`, because a fieldset is role `group` and takes a name from the author, and a legend is a VISIBLE heading this form was designed without. The discarded labels came off rather than being moved: the sentence above the figure and the one under it already say what it is. The footer headings became h3 and `site-footer.css` stopped binding the look to the LEVEL, which is the defect `group-head.css` was broken by twice. | **Closed.** rollout12 none, width12 none over 55 x 58, quality13 at its 24 baseline, pages13 0 of 128 | | ~~The `.switch` checkbox is 20px against the 24px of WCAG 2.5.8. Its wrapping label is the real pointer target at 44px or more, so the raw control is not the target~~ | **OPEN, unchanged, and step 9 counted it a third time.** Recorded rather than resized on 2026-08-11: a kit value changes only by a named decision, and this one has a defensible reading. The row's argument got stronger this round rather than weaker: the focus ring now lands on the label too, so the 20px box is no longer the target for a pointer, for a keyboard or for the eye. `pixel-proof.html` reports "tap targets under the floor: 0 in the product", and names this checkbox as the one flag it set aside for exactly this reason | **CLOSED 2026-08-20. The reading is recorded and the control stays 20px.** WCAG 2.5.8 measures the TARGET and the target is the wrapping label at 44px or more, which since stage 09 also carries the focus ring. Resizing a control that is not the target, to satisfy a rule about targets, helps nobody | | **Four atoms are sized by their hosts, which is the undeclared variant `CLAUDE.md` forbids by name.** `amount.css` sets four different sizes keyed by `.row`, `.charges li`, `.plan-opt` and `.hero`; `logo.css` sets four keyed by `.hero`, `.alert`, `.tile` and `.rgroup li`; `muted-line.css` two keyed by `.tile` and `.row`; `big-total.css` one keyed by `.sharecard` | **OPEN, the principle holds, and THREE OF THE FOUR NUMBERS DID NOT SURVIVE RE-READING.** Corrected in place rather than carried: `amount.css:70, 75, 80, 88` are four host rules but only **two distinct sizes**, `--type-body` on `.row` and `.charges li` and `--type-figure` on `.plan-opt` and `.hero`, the rest of each rule being weight and line-height. This is the same correction the step 9 first batch made to the registry, "amount 13.5/14/30/32 (it is 14/32)". `logo.css` sets **five** host-keyed sizes and not four: `.hero` 52, `.card .top` 36, `.alert` 32, `.tile` 30, `.rgroup li` 22, at lines 82, 90, 83, 84 and 95. `muted-line.css` sets **one** and not two, `.tile` at line 148; there is no `.row .muted` rule in any component file. `big-total.css:56` on `.sharecard` is the only one of the four that measured as written. Step 9 did not execute the rename: naming these means declaring a size axis and writing the class into the markup of both corpora. The visible cost is unchanged: seven of the eighteen atom pages host a higher rung because an atom page cannot show its own size axis without importing a molecule | **CLOSED IN PART on 2026-08-14, at the rollout, and the part that closed is the one that forced the question.** Founder: the axis exists. `amount.css` now declares `.amt` at `--type-body` on the base and `.amt.figure` at `--type-figure`, and all four host font-size declarations are gone; what stays with the hosts is `flex: none` on the row (placement), weight 400 on a past charge (history is a texture, not a size) and the two leadings, 1.05 in a plan card against 1.08 in the hero, which is what a container asks for. **Zero pixels:** all 66 amounts in the coloured corpus snapshotted before and after on size, weight, leading, tracking, display and box, identical. What forced it was stage 12 needing a FIFTH host, cancel-win's freed figure, and adding one more occurrence of a rule the system forbids is not a way to build a screen. **Still open: `logo.css` (five host sizes), `muted-line.css` (one) and `big-total.css` (one).** The same answer applies and the same evidence is owed | **Founder** decided the axis; **logo, muted line and big total** remain | --- ## Found at the molecule and organism rounds, and each one is a decision rather than a bug | What | State on 2026-08-12 | Who owns it | |---|---|---| | **The switch row has no hover of its own.** `kit.css` gives `label.switch` none, and the focus ring lands on the checkbox inside it because a label is not focusable | **CLOSED, both halves, at the founder's review of the components.** `switch-row.css:159` gives the row `--bg-hover`, the same role every other row in the product answers with, and the four rows on Settings keep their heights to the hundredth. Measured on `design/settings.html`: the row's fill goes transparent to `rgb(244, 247, 248)` under the pointer, the same as the subscription row and the nav row | Closed | | ~~`.cut` is now `.btn` minus two things. Measured on the rendered page: identical `min-height` 44, `padding: 0 16px` and `font-size: 14px` after the fold. It differs only in `font-weight`, 400 against 600, and in a hover that changes edge and ink but not fill~~ | **OPEN, and step 9 did not merge it.** Re-read today: `save-focus-candidate.css:37` still says in its own header "THE CLASS NAMES ARE NOT TOUCHED", and the six rules at `:64`, `:88`, `:101` and `:126` still stand under `.cand .cut`. The census still shows `.cut` as its own family, 2 controls, one value on all seventeen properties. Whether it merges is still a rename question, and a rename with no row on the map is a rename nobody executes. The map still owes it a row | **CLOSED 2026-08-20. Merged.** `.btn.soft` is declared in `button.css`: font-weight 400, and a hover that keeps the edge and the ink and refuses the fill, because a fill under the pointer on the one control that ends a subscription reads as encouragement. Six rules left `save-focus-candidate.css`; two stayed, and both are the host's. **The modifier was called `.quiet` for an hour and `counts.cjs` caught it**: `.quiet` is already the Quiet line component on 14 pages. Zero pixels, measured at 360 and 1440 in both themes | | ~~"Sign out" wears the nav row class with no modifier. It is the one row in the product that leaves rather than goes deeper, and nothing in the markup says so~~ | **OPEN, unchanged.** `design/settings.html:123` still reads `` with the same chevron every other nav row carries. Census case 9. A variant, or a deliberate sameness with a reason written down | **CLOSED 2026-08-20. It has one.** `.navrow.leave`, and what the modifier does is drop the chevron. A chevron means "there is more this way" - it opens a disclosure twice on that same screen - so drawing it on the control that ends the session says the opposite of what the control does. The mark left the markup rather than being hidden by CSS | | ~~`.cand` is recorded as a molecule while hosting a molecule, which the ladder calls an organism~~ | **OPEN, unchanged, and step 9 re-read the inventory without moving it.** `inventory.md:146` still files Save-focus candidate as a molecule and the level is noted on its page. Moving it changes the `@import` order in `index.css`, which is the reason it was left alone mid-round and is still the reason | **CLOSED 2026-08-20. It is an organism.** The ladder says a thing that hosts a molecule is one, and the reason it stayed was never the ladder - it was the `@import` order that moving it changes. Moved in `index.css`, in `inventory.md` (22 molecules, 28 organisms, total unchanged at 70) and in the registry. Measured after: the 460px reflow still fires between 460 and 461 | | **Two link hovers are dead in the product today, and it is not the system that killed them.** `.trust a:hover` and `.notice a:hover` are 0-2-1; `.app .trust a` and `.app .notice a` are also 0-2-1 and written 22 lines LATER in `kit.css`, so the later rule wins and the ink never changes | **CLOSED, both, and measured on the product rather than on the stand.** `trust-block.css` rebuilt the four rules as two, rest at 0-2-1 and hover at 0-3-1, so the hover wins on arithmetic and no later file can take it away by being written afterwards. Measured on `design/home.html` at 1280, the trust link's underline goes `rgb(231, 237, 238)` to `rgb(28, 106, 118)` under the pointer; **its ink does not change, and that is the design**, the link being petrol already. On `design/home-error.html` the wash link goes `rgb(154, 88, 66)` to `rgb(28, 106, 118)`, and the attention link on `design/home.html` `rgb(138, 92, 12)` to the same petrol. A first pass at this measurement read `color` and `text-decoration-line` and reported the trust link as still dead; the property that moves is `text-decoration-color`, and a probe that does not read the property a rule declares reports the rule as absent | Closed | | **A host repaints an atom, carried on purpose.** `.plan-opt .p` sets the Muted line to `--text-body` where the atom gives `--text-muted` everywhere else | **CLOSED at the founder's review of the components.** The host paints nothing and the atom's `--text-muted` stands, which was the second of the two answers the file carried. Shipped under a condition and the condition was checked: 10.16:1 to 5.78:1 light and 10.44:1 to 6.40:1 dark, both clearing the 4.5:1 an ink owes at 14px. Three paragraphs on `design/upgrade.html` | Closed | | **A third host override of the action row has no owner yet.** `.panel .gate .actions { margin: 0 }` at `kit.css:890` is byte-identical to `.locked .actions`, before the 8px grid as well as after, so the grid did not cause it | **CLOSED at the organism round.** All three now live in the file of the container that owns them: `panel.css:210` for `.panel .gate .actions`, `card.css` for `.card.prose .actions`, `reveal-step.css:110` for `.rstep .actions`. Written in the molecule's file they would have been a molecule reaching up into an organism | Closed | | **The stand was showing every list component at more than twice its real width.** `.kit-stage` handed a subscription row 898px where Home renders it at 425 | **CLOSED at the molecule round**, listed here as the record. `--kit-measure` carries the number, defaulting to the product's row column, with `.wide` and `.bleed` as the two named exceptions. Same class of error as the reading measure that came unbound past 900px on two accepted screens: nobody widens the window, so it stays quiet | Closed | | **`.field input` and `.field select` are not a ladder problem, and they are counted separately so the number is honest.** `select.css` names `.field` 38 times and `text-input.css` 26 | **CLOSED at step 6.** `.input`, `.select` and `.check` are written on 83 elements that had no class attribute at all. Zero pixels: 83 controls fingerprinted on 17 properties across 90 pages, 0 differences, and the test is not vacuous, because without the class the same 61 stand controls show 349. The second census sees the same move from the other side: **controls with no class of their own went from 126 to 104 per viewport, exactly 22**, which is the 14 inputs, 4 selects and 4 checkboxes on the coloured corpus | Closed at step 6 | | **The back arrow is still a character, and one class is why.** `a.back` is `‹ Back` 36 times and `× Close` twice, and CSS cannot select by the words inside an element | **CLOSED at step 6.** `.back` and `.close` are two classes and both marks are drawn, over `currentColor`, same 24 unit cell and same 16px box as `nav-row.css`'s chevron and `text-input.css`'s clear cross, so three marks are one family with one place to change them. Moves the label's left edge 6.67px on `.back` and 3.73px on `.close`. The census confirms the split landed: `.back` 16 controls and `.close` 2, both families clean on all seventeen properties | Closed at step 6 | | **The panel's focus ring is clipped by the panel's own `overflow: hidden`.** The `summary` sits 1px from the panel's edge and `base.css` draws the ring 2px wide at 2px offset, so 3px of it falls outside the box that clips it | **CLOSED at the founder's review of the components**, and the backlog understated it: the disclosure ships closed, so the summary is the panel's only child and the bottom edge went too. The ring was 100 percent invisible on all four edges in both themes and at both viewports. The negative offset loses nothing and the corners take the panel's own curve; `overflow: hidden` was kept because it is the only thing holding three fills inside the card's corners. **Worth knowing where it lands**: `` occurred on zero of the 28 coloured screens when this was written, so the repair was true in the system and invisible in the product. It became visible on 2026-08-13, one stage earlier than predicted, when the self-sufficiency test coloured Alerts: `design/alerts.html` is the only page in the product carrying a ``, and the ring was walked on it in both themes. The literal `-2px` went with it at step 6 of stage 09 and reads `calc(var(--ring) * -1)`, because a state that names the ring's own width should not spell it a second time | Closed | | **The panel's `summary` has no `--tap` floor.** It clears 44px at 53, by arithmetic rather than by declaration | **CLOSED at the founder's review of the components.** `panel.css:134` declares `min-height: var(--tap)`. It measured 53px at both viewports in both themes, so the floor never applies and the panel renders byte-identically with the line and without it, which is the point rather than an objection | Closed | | **The two-hairline seam is REAL, and it was measured rather than argued.** The last charge `li` and the `.gate` below it touch at 0.00px, two pixels of `--line-divider` at one seam | **CLOSED.** Same row as the seam above, kept here because two agents found it independently and the second one opened it in a browser. The principle went in, not the patch | Closed | | **`card.css` carries `.app > .screen > .locked { max-width: 560px }` and the App shell must not repeat it.** Its two neighbours in `kit.css`, `.form-col` and `.rows-col`, ARE the shell's | **CLOSED at round B, with the split named.** `app-shell.css:284` and `:285` take `.form-col` at 560px and `.rows-col` at 620px; `card.css` keeps the card's own 560px reading measure because the card is the subject of that rule. Three reading measures, two owners, and the reason each one has the owner it has is written in both files | Closed | | **Four sibling `order` declarations belong to the App shell and are homeless until it is built** | **CLOSED at round B.** `app-shell.css:343-346` takes all four: `.head` order 1, `nav.actions` order 2 with the 24px margin that migrated from the component, `.groups` order 3, `.trust` order 4. Each had been correctly refused by the component that contains the element, and each was named at the line where it was refused, which is the only reason they were findable | Closed | | ~~The system has no z-index scale. The whole product's stacking order is three literals: `1` on the field's currency prefix, `5` on the app bar, `6` on the tab bar. And `--tap-rail` is separate and still open, the rail's 48px anchor floor being neither `--tap` (44) nor `--row` (56)~~ | **OPEN on the z-index half, and the `--tap-rail` half is now its own row below.** Re-counted today across `design/system/`: still exactly **three** live `z-index` declarations, `app-bar.css:94` at 5, `tab-bar.css:71` at 6 and `text-input.css:122` at 1, every other occurrence of the word being a comment. The money field is a grid rather than a positioned box, so `text-input.css:122` reads "a grid item takes z-index without being positioned"; the value and the count are the same. The dialog sheet added no fourth, having no `z-index` and no `position` anywhere. So the question is still the narrow one, whether three literals, two of them adjacent, earn two tokens, and the recommendation is still not yet. **Step 9's geometry triage answered the neighbouring question and refused this one**: it created `--ring` for the focus ring, which was 16 literals in 17 files, and left both size literals alone. `--tap-rail` moves to the row "Two `min-height` literals" below, because step 9 found a second wearer and a two-wearer debt is a different row from a one-wearer debt | **CLOSED 2026-08-20. Two tokens, not a scale.** `--z-appbar` and `--z-tabbar`, because 5 and 6 are not two values but one RELATIONSHIP - where the two chromes meet, the bottom one wins - carried until today in prose in two files. The money field's `1` is local and stays a literal: naming it would have made this a scale, which is the thing being refused. Zero pixels. **The `--tap-rail` half of this row was already done in the code and nobody had closed it**: `tab-bar.css:219` reads the token and has since 2026-08-12 | | **~~`DESIGN.md` disagrees with the tab bar in two places.~~ CLOSED 2026-08-20** | It recorded the label at 11px against a `tab-bar.css:109` of `--type-meta`, which `tokens.css` declares at 12px, and the current destination at weight 700 against a `tab-bar.css:131` of 600 | Both corrected in the document, code wins, the same ruling already applied to the subscription row. The other two claims in that paragraph were re-checked at the same time and both hold: the rail's left border is 3px of petrol and its fill is paper | **Closed** with the fifteen above, in the same edit | | **A button that may not wrap has to go somewhere when it does not fit.** At 360px `.actions` measured 213px of room and its button 225: a 12px overflow, and `.plan-opt` overflowing by 2px the same way | **CORRECTED. The rule stands, the numbers do not reproduce, and the row is open on the principle only.** Swept on 2026-08-12 at 360 and 1280 across all 28 coloured screens: **zero buttons cross their container's right edge, and zero pages overflow the document.** The reason is in the component: `action-row.css:47` declares `flex-wrap: wrap`, so the row breaks BETWEEN buttons and never inside one. Measured on the worst case, `upgrade-payment-failed` at 360, where "Try another payment method" is 230.56px and "Maybe later" 113.91px in a 328px row: the two stack, both at x=16, and nothing is clipped. **Held again through step 9**: the rail comparison round on `pixel-proof.html` reports horizontal overflow anywhere as 0, and every commit of the step closes with 376 renders and no overflow. The principle is still real, because `white-space: nowrap` means a label longer than the room a single line can give it must overflow, shrink or stack, and the three honest answers are unchanged: a shorter label (Voice owns the string), a declared compact padding below 400px, or the action row stacking vertically at the narrow end. The 213/225 pair is withdrawn rather than carried, because a number nobody can reproduce is worse than no number. **REOPENED WITH REPRODUCIBLE NUMBERS ON 2026-08-13, by the type scale moving to `rem`, and the principle this row kept alive is now the live case.** The three answers it listed all assume a label can be authored once and measured once. A `rem` scale hands the text size to the READER, so no label exists that fits every root, and "a shorter label" has nothing left to shorten. Measured at a 24px root, which is what Chrome offers by name under **Very large**: on `design/upgrade.html` at **390px**, the plan card's track is 358 and "Start Tendd Pro - $7.99 a month" is 356 as a non-wrapping button, so the card is 390 in a 358 track and the page scrolls sideways by 16px. **Fixed**, one scoped exception in `button.css`: `.app .plan-opt .btn` takes `white-space: normal`. It is written in `button.css` and not in `plan-option.css` because white-space is the button's property, and it is scoped because this is the only button in the product whose label is a sentence carrying a price and a period. **Zero pixels move at the default**, checked across 8 596 element records on all 32 coloured screens. **What is left is one corner and it is named: 320px viewport with a 24px root, and nothing else.** Swept live and RECOUNTED on 2026-08-14 against the true corpus: **672 renders, 32 pages x 7 widths (320 to 1600) x 3 root sizes (16, 20, 24)**; the first pass ran 609 on a list of 28 that predated the four Alerts screens. Everything at **360 and above is clean at every root size**. At 320 with a 24px root **eight rows** cross the edge and they are three things: `.btn.primary` "Try another payment method" on `upgrade-payment-failed`, 329px in a 288px zone; the app bar's plan `.chip` on the five History and Trends screens, pushed out 25 to 32px by a brand lockup that grew with the text beside it; and a `.btn` on `alerts` and `alerts-empty` by 7px, which only the recount could see. **Not fixed, and deliberately, because the general answer is the founder's:** `white-space: nowrap` on `.btn` came from a founder finding on the etalon page, and the honest question now is whether that rule can survive a text size the product no longer controls. Two answers, both real: keep it and add an exception per host as each one appears (two so far), or drop it from `.btn` entirely, which stage 09 measured as zero pixels at the default and which makes the whole class of finding go away. **Founder** | **Founder**, with **Voice** for the string | | **The switch row's focus ring lands on the checkbox, not the row.** A `label` is not focusable, so the ring is drawn 20px wide inside a 44px target | **CLOSED at the founder's review of the components, as one paired write across two files.** `switch-row.css:162` draws the ring on the row through `:has(input:focus-visible)`, and `checkbox.css` stands its own down inside `.switch`. Either half alone is visibly wrong: the row's half alone gives two concentric petrol rings, the atom's half alone gives none. Measured: 2px at 2px offset around 328px at 360 and 620px at 1280, exactly one row of four, heights unchanged. `base.css` also lost `label:has(input)` from its focus-visible list, which read exactly like somebody meaning to solve this and was dead, because a label can never match `:focus-visible` | Closed | --- ## Still open on the components, and each needs a person or a step Every row in this section named **step 9** as its owner. Step 9 has now run and executed none of them, which is the honest state rather than a failure: each one turns on a value or an axis, and a value moves only by a named decision from the founder. The owner column is corrected below. | What | The question, and why it is not answered here | Who owns it | |---|---|---| | ~~The `.plans` fold and the gap question are one decision. The map folded `.plans` into `.grid` and the agent executing it refused, with a measurement: past a 760 container the plan row and the tile row were the SAME rule, 3 columns, `--space-8`, identical 201.328px tracks, and at the narrow end they are 1 column against 2, because a chip survives a 159px column and a card holding a price, a sentence, a button and a four-line list does not. THE SAMENESS ENDED ON 2026-08-18 and the case for folding went with it: the plan row declares `repeat(3, minmax(0, 1fr))` where the tile row auto-fills, its cap is 57rem where the tile row has none of its own, its gap is the door row's 16 where the tile row keeps 8, and the tracks that were identical at 201.328 are now 293.33 against 201.328 at a 1440 window. Two rules that agree on one number are a folding candidate; two rules with different counts, different caps and different widths are two rules~~ | **OPEN, and step 9 declared no modifier.** The column count is not a container property, it is the child's narrow floor, and `inventory.md` gives the component the axis "1 to 2 / 2 to 3" and never names this one. Folding it needs a modifier nobody has declared, and the gap question sits on the same component and cannot be answered separately. Counted today, `plans` stands on **1** coloured screen, `design/upgrade.html`, and **2** grey ones, `index.html` and `upgrade.html`. It is the only old name left standing in colour on purpose | **CLOSED 2026-08-20. It does not fold.** The two rules stopped being the same on 2026-08-18: 3 fixed columns against auto-fill, a 57rem cap against none, a 16px gap against 8, and tracks of 293.33 against 201.328 at a 1440 window. Two rules that agree on one number are a folding candidate; two rules with different counts, caps and widths are two rules | | ~~A size modifier is owed to `.muted`. Nine names fold onto it and they rendered at two type steps, not one. Only `.legal` has a token in the markup to hold the quieter step, `.ruled`, so only `.legal` keeps it; `.consequence` and `.freshness` fold onto a bare `

` and grow 2px~~ | **OPEN, and the map is incomplete here rather than wrong.** Measured out of `kit.css`: `.consequence` 12.5px and `.freshness` 12px, both now 14px, on 11 coloured places and 2 grey ones. Re-read today, `muted-line.css:95` gives the atom `--type-body` and the only host-keyed size left is `.tile` at `--type-meta`, so the two steps are still not declared as an axis. What is owed is a declared size modifier plus the class in the markup. A modifier no element wears is a rule documenting a decision nobody took, which is why one was not invented to hide the gap | **CLOSED 2026-08-20, and most of it had been closed in the code for four days.** `muted-line.css:284` declares `.muted.fine` at `--type-meta` and has since 2026-08-16. What was left was six lines that grew 2px in the fold, the old `.consequence`, and **reading the six sentences answered it the other way**: they are the reassurance under an error or an empty state ("Nothing about your plan or your data changed"), and making the reassurance smaller than the problem is the opposite of what this product is for. They stay at body size. The founder had approved putting `.fine` on them sight unseen; this is the reading after opening the file, and it is the founder's to overrule | | **`.muted.spaced` is a modifier that modifies nothing.** After the foundations review its `margin-top: 18px` and the base's 12px both fold to `--space-16`, so it is byte identical to the base | **OPEN, and it is not deleted the way `.btn.compact` was, on a difference of authority rather than of merit.** `.compact` is a row of the renaming map; this is not, so removing it means taking the class off `design/upgrade-processing.html`, which no step has owned. Zero pixels either way. Re-counted today: `spaced` stands **1** time in colour and **3** times in the grey, and `muted-line.css:192` still declares it | **DROPPED AT VERIFICATION 2026-08-18, and the row was wrong rather than stale.** It says the modifier is byte identical to the base. It is not: the base is `.app .screen > .muted`, a DIRECT CHILD rule, and of the three wearers in colour **two are not direct children of a screen** - the consequence line on `cancel-guide-no-guide` and the delete warning on `data-privacy`. Measured, all three compute `margin-top: 16px`, and for those two this modifier is the only thing supplying it. Deleting it would have moved two screens | | ~~The sixth host-scoped chip placement lives in the wrong file. Five are in `chip.css` (`.row .chip`, `.appbar .chip`, `p.chip`, `.alert .what .chip`, `.card .top .chip`) and the sixth is `plan-option.css`'s. Re-read 2026-08-18, when that sixth rule was rewritten: it no longer sets `align-self` and a bottom margin, it takes the chip OUT of the flow (`position: absolute; margin: 0`) and places it in the card's top right corner, because a badge on one card of three was pushing that card's price 32px below the other two. The split is unchanged and so is the question~~ | **OPEN, and both placements are defensible where they stand, which is the problem.** Re-read today and the split is unchanged: `chip.css:193, 194, 195, 196, 217` carry the five and `plan-option.css:106` carries the sixth. The rule states where a chip SITS in that card, which is a fact about the card; the other five state the same kind of fact and live with the chip. One of the two readings has to win, and whichever wins, the other five or the one moves | **CLOSED 2026-08-20. The sixth moved.** All six placements of the Chip are in `chip.css`. The argument for the rule stays in `plan-option.css`, where it belongs, because it is about that card; the four declarations left. Zero pixels: same selectors, and `chip.css` is imported before its old home | | ~~The `.divlist` bottom margin is keyed to the shell's column. `divided-list.css:99` declares `.app .rows-col > .divlist { margin-bottom: var(--space-24) }`~~ | **OPEN, with three honest homes and no decision.** Still at `divided-list.css:99` today. One of the two old names carried a bottom margin and the other did not, and a component does not own the gap under itself, so the gap is keyed to the container instead. The file names the alternatives at lines 89 and 90: a declared `.divlist.spaced` modifier, which needs a class added to two product screens and **collides with `muted-line.css`'s `.spaced`**, where the same word would mean a different thing; or a rule in `app-shell.css`, which is what `action-row.css` did with its own 24px margin | **CLOSED 2026-08-20. It is in `app-shell.css`.** The `.divlist.spaced` modifier lost on a NAME COLLISION rather than on taste: `muted-line.css` already uses `.spaced` to mean a margin above. `action-row.css` had already set the precedent with its own 24px. **Written into the 900 block for one round and measured out of it**: `.rows-col` is in the markup at every width, so scoped it would have dropped 22px off four screens on a phone | --- ## Found on 2026-08-12 at the second control census Six rows. The first three were new findings, the next two were corrections to work that round believed it had finished, and the last was the open half of a repair that had already landed. **Two of the six are closed by step 9 and two more are closed in part**, and where a row closed in part it is counted as open below, because a row with anything unfinished in it is not a closed row. | What | The measurement | Who owns it | |---|---|---| | **A live defect on a shipped screen, found and closed on the same day.** Two coloured screens were drawing their chart with variables the system does not declare. `design/history-trends.html:75` read `var(--line)` and `var(--mid)`, `design/history-trends-locked.html:69` read `var(--line)`, all three inside SVG presentation attributes. Those are `kit.css` names, and `design/system/tokens.css` declares neither | **CLOSED, and the scan re-run today across the whole system.** The computed stroke had been `none`: read in a browser at 360 and 1280 on both screens, the three grid lines and the trend path were not drawn at all, so the chart on the two screens the Pro tier is sold on was an empty box. Repaired within the round and re-measured: all four now stroke `rgb(228, 233, 234)` for the frame and the trend path strokes `rgb(56, 67, 73)` from `var(--text-body)`, byte-identical to what `kit.css` held under the old names. **Re-run today over 136 declared custom properties, all 29 files in `design/`, `base.css` and all 57 component files: zero undeclared `var()` in any live declaration.** One hit exists and it is prose, `wash-block.css:45`, quoting what `kit.css` used to carry inside a comment. **The one thing still owed is the instrument, not the result**: `screens/` holds `fp.cjs`, `ledger.cjs` and `brandproof.cjs` and no var scan, so the check that found the worst defect of the stage is the only one with no re-runnable script beside it | Closed on the defect. **Stage 12** owes the scan a script in `screens/` before the rollout | | **`design/` has no `index.html`.** `CLAUDE.md` states the rule by name: opening a stage folder lands on the product screen, node 1.1, and the list of all pages of that stage is `overview.html`, and it says this holds for `wireframes/` and `design/` alike | **OPEN, verified again today**: `design/index.html` does not exist, while `design/overview.html` does. `wireframes/index.html` is the landing and carries 60 distinct class tokens. So the coloured half of the project has a hub and no home page, and the rule is broken today. The screen behind it is the landing, which is one of the 27 screens not yet coloured, so this is not a five-minute copy. **The step 9 dry run put a number under it**: 36 of the landing's 60 tokens are outside the system | **CLOSED 2026-08-18.** `design/index.html` exists and is the coloured public landing, 52KB of it; the rule this row quotes is satisfied and has been since the landing was coloured in August. Re-read before closing rather than assumed | | ~~The `.landing` scope is carried by 46 of the 55 component files for a page that exists only in the grey~~ | **OPEN, and the count in the heading of this row is STALE and corrected here.** Measured today: **48 of the 57** component files name `.landing`, and `base.css` names it once more, in a comment. The row was written when the system had 55 files; the brand added two and the scope grew with them. `pixel-proof.html` carries the corrected 48 of 57. **One further correction, and it runs the other way**: the proof page says "no page wears it", and that is true of colour only. `wireframes/index.html:14` reads `

`, so the grey landing DOES wear it and every one of those 48 files is already live on that page; what is unworn is `.landing` on the coloured corpus, where it stands **0** times across 29 files. It is not waste, because stage 12 colours the landing and the scope starts paying at once; it is recorded so that whoever reads a component file and wonders why every rule is written twice gets the answer from here rather than from a guess | **CLOSED 2026-08-20. It stays.** The premise it was written on is gone - `design/index.html` is the coloured landing, so all 48 of those files are live in both corpora - and the doubled selector is what lets a component be lifted into either host without knowing which one it landed in | | **The three chip tones were cleared from `kit.html` and two button classes were not.** The deletion of six declared things counted two corpora when there are three: the stand was not swept, and `kit.html` was the last place in the repository writing `tone-attention`, `tone-error` and `cancelled` | **CLOSED on the classes at step 9, second batch, and OPEN on the comment.** Re-counted today as class attributes across all three corpora: `tone-attention`, `tone-error`, `cancelled`, `inverse` and `is-disabled` all return **0 / 0 / 0**. `design/kit/kit.html:110` and `:120-121` no longer carry `is-disabled` or `inverse`; line 117 is a plain `.btn`. The step 9 commit records the same result from the other side: "the census re-counted two rows that credited the frozen stand with 2 `inverse` and 1 `.is-disabled` deleted underneath them. All six tracked forms are now clean in all three corpora". **The stale comment closed on 2026-08-13**, at step 6 of stage 09, and Codex was the instrument that found it: `chip.css:88` was still telling the reader that "`design/kit/kit.html` carries `chip tone-attention` on lines 133 and ...", a claim that named line numbers in another file and had been false since the repair. It now records the repair instead, in the past tense it belongs in. A comment that names a line number in another file is a claim, and this row is the reason the file no longer counts anything by hand | **Closed 2026-08-13** | | **CLOSED 2026-08-17 on both halves.** `design/kit/kit.html` rendered with no chrome at all: its `