` or `` in every frozen artefact, because a static mockup has nothing to click, and all three already carried `cursor: pointer` - the design saying it always meant them as controls. Phase 5 makes them real `` elements, which `docs/design-targets.md` requires of the Tool Shed by name so `?tool=X&video=Y` can compose, and Bootstrap's reboot underlines a link. On `.spine` the label is `writing-mode: vertical-rl`, so the underline ran VERTICALLY and read as an extra hairline beside every tool name. NO MECHANICAL LAYER COULD SEE IT. The token check reads colour, radius, font size and typeface and does not read `text-decoration`, so the Tool Shed passed all four D67 layers at 64 of 64 assertions and zero off-token while drawing it. Found by looking at the page, then proven by probing computed style on the frozen artefact against the Django page: `none` there, `underline` here. Recorded as a gap in the ORACLE, not only as a defect in a page. IT CANNOT MOVE A SIGNED PAGE AND THAT WAS CHECKED, NOT ARGUED: no frozen artefact carries an `` on any of the three, so the full sweep returns the same thirteen element counts (12, 30, 64, 91, 85, 104, 147, 155, 100, 170, 186, 78, 170) and the same twelve negative-control counts (2, 12, 51, 22, 28, 50, 49, 43, 48, 64, 55, 65) before and after. NO NEW TOKEN and NO new contrast pair, so `contrast.py` is untouched at 77 pairs. The file already did this seven times over - `.shell__nav a`, `.btn-cw`, `.trackcard`, `.crumb a` and three more - so this is the file's own practice reaching three components that had never been links. v17 · 22 Aug 2026 wave 2i, surface 11 (System health). SECTION 17 ADDED, and it is the LAST surface section: the active `More` (the first page in the build that lives under the overflow control), the dark console, the history table with a trend chart per row, the targets table and the System Architect's targets form, the cross-team pair, forecast accuracy, tracking freshness and triage pace. TWO NEW COLOUR TOKENS, and both are a MEASUREMENT rather than a choice: Ben asked for green and red direction arrows on the console and the frozen pair CANNOT be read on `--itk-black` - `--text-success` is 1.99 and `--text-danger` is 1.84 against D68's 4.5 floor. `--text-success-dk` (#739987) is `--itk-sage`, already a brand token, at 6.11; `--text-danger-dk` (#CD6F88) is oxblood's exact hue and saturation lifted in lightness, at 5.73. BOTH ARE DECLARED INSIDE `.csl` RATHER THAN `:root`, which is a guard and not tidiness: both FAIL on white, so a future session reaching for them on a light surface resolves to nothing instead of to an unreadable pass. THIS IS ALSO THE FIRST SURFACE THAT BOTH REPORTS AND EDITS WHAT IT REPORTS AGAINST (D73's targets, Ben ruling 22 Aug) - the Dashboard's "the page writes nothing" gate is deliberately NOT inherited here, and the two halves are separated by the frozen `.btabs`. v16 · 21 Aug 2026 wave 2h, O52. SECTION 4, TWO DECLARATIONS, AND IT IS A DEFECT FIX RATHER THAN A DESIGN: `.shell__nav` gains `overflow-x: auto` with `overflow-y: hidden` written explicitly, because the topbar nav OVERLAPPED the account block below a 1106px viewport - `More` painted on top of `Ben Bielby` - and had done since the wave-1 freeze on 6 Aug. Ben ruling DP-108 A, taken against a first-layout-breakpoint option he declined. THE SAME RULING NARROWS DP-16 C: More being "always present" means always IN the nav, not always on screen. NO SIGNATURE MOVED AND THAT WAS PROVEN RATHER THAN ARGUED - above the breakpoint there is no overflow, so the full eleven-artefact `compare.py` sweep is byte-equal before and after (every element count identical, all PASS), `render.py` passes all eleven, and the topbar measures 52px at every width from 1400 down to 900. One Phase 5 consequence is recorded at the rule: a scroll container clips, so More's future dropdown must not live inside `.shell__nav`. v15 · 19 Aug 2026 wave 2g, surface 9 (How-It-Works). SECTION 15 ADDED: the pipeline rail and its stage card, the gate markers with the travelling light that lights each one as it passes, the confidence ladder aligned under the stages it belongs to, both track rows and their titles, the progression arrows, the hover tip, the reveal panel's body, the step list, the role cards and the narrative plate. NO NEW TOKEN. TWO NEW CONTRAST PAIRS, both declared before they were drawn rather than after: --itk-gold on --itk-charcoal at 5.40 (the narrative plate, Ben's ruled colour) and --itk-smoke on --itk-charcoal at 7.21. 68 pairs pass. THE MODAL IS REUSED, NOT REDEFINED: section 13's .gvpop__hd, __t and __x are exactly what this surface's nine panels wear, so the only new rule for them is the body layout and ONE WIDTH, and that width is SCOPED to .hiw-pop-dlg rather than applied to .modal-dialog - because .modal-dialog is Bootstrap's own selector and the SIGNED governance.html renders through it, so a global rule would have redrawn a frozen artefact from another page's instruction. Also here, and unrelated to this surface: the header's layer list stopped at 13 while section 14 existed, and two changelog entries both claimed v13. Both corrected above. v13 · 18 Aug 2026, Ben ruling closing O49: "extend the charcoal badge to project detail too". NO RULE CHANGED - `.badge-cw--ended` already existed - but TWO COMMENTS DID, and both had stated a superseded position as settled. Section 7 carried a flagged note saying the badge was deliberately NOT extended; section 11 said `--off` is the Stopped fill. Freeze amendment A-3, four files, and it reached the PATTERN SET as well as Project Detail because the catalogue's own sample decision log drew a `Stop` in oxblood - the identical case, on the one page whose job is to be the reference. THE PALETTE'S LINE IS NOW CLEAN AND FITS IN A SENTENCE: oxblood means something is WRONG - late, off track, in error - and charcoal means something has ENDED. Oxblood did not shrink to nothing: `--status-off` is still the RAG fill, still the overdue fill, still `--text-danger`. Section 11's correction KEEPS THE EARLIER HISTORY on purpose. The concern that an oxblood Stop contradicted DP-18 was wrong and is still wrong, since DP-18 draws its line at button-versus-badge. The colour moved for a reason DP-18 never addressed: one fill carrying two meanings on one screen. v14 · 19 Aug 2026 wave 2f, surface 8 (Portfolio Dashboard). SECTION 14 ADDED: RENUMBERED FROM v13 AT THE HOW-IT-WORKS FREEZE, because two entries carried that label: this one and the 18 Aug O49 entry below. Both were real changes on different days and neither is wrong; only the label collided. This is the later change, so it takes the higher number, and the happy result is that section N now matches version N from 13 onward. The decisions register at v64 refers to this entry as "app.css v13", and that reference is corrected in the register entry landed with this freeze rather than left to drift. the headline gauge and its year toggle, the confidence mix with its under-segment labels and vertical legend, the rung-stacked split rows with their verified Total rows, the confidence chip, the centred cell, the width-independent top-10 table cap with a permanent scrollbar, the two-track count bands, and the dashboard grid. THE FIRST PALETTE EXPANSION SINCE --text-eg: three value-confidence rungs join the token set (--conf-verified, --conf-evidenced, --conf-banded; --conf-tracked aliases --itk-teal), one hue at four strengths per DP-87 A, adopted by Ben's own instructions. FIVE new rows in tools/contrast-pairs.tsv: the four rung text pairs (11.13 / 8.04 / 12.53 / 16.37) and the band label pair --itk-charcoal on --sec-band (6.88), which had been used unpaired since the Idea Board. v12 · 18 Aug 2026 wave 2e, surface 7 (Governance). SECTION 13 ADDED: the month grid and its day cells, a scheduled review inside a cell, the today marker, the overdue tab count and DP-69's reveal panel. No new token and no new contrast pair, the fourth surface running to manage both. THESE RULES SPENT A ROUND LIVING IN THE MOCKUP'S OWN `