Where the merchant lane stands, end of 2026-08-28

One line: everything I could finish is committed and pushed on fix/merchant-portal-988, 21 commits ahead of the portal branch, and the only work left in merchant territory is blocked on two decisions that are hers.

What is waiting on her?

Decision	State	Where the evidence is
The merchant design overhaul direction, A, B or C	Three directions rendered with the app's real CSS at 1440x950. Not picked. I did not pick for her.	merchant-design-brief.md, and the mockups served at http://localhost:8099/nav.html while this session lives
The two failing colour tokens	Already approved by her, directly, on 2026-08-28. Not started: the code review was taken up first and the day closed	axe-sweep-0828.md R1 and R2, with measured ratios

• The token split is approved work, not a pending question. Anyone picking it up should not re-ask her.

What is done?

• The footer defect (#988), root-caused to the previous day's own M5 fix, with the computed styles that proved it, and the admin rail cleared as unaffected.
• Ten per-screen defects across the merchant surfaces, and three reported rather than fixed.
• #980, the switcher, which needed her word for one shared file and got it.
• #993, venue addresses, fixed forward at the importer rather than formatted at the screen.
• The offer builder, driven end to end for the first time. Four fixes, including a title that could publish ten characters over what its card renders while every indicator said it was fine.
• The first accessibility sweep the merchant portal has had. Nineteen surfaces, 12 clean, and everything left is the two shared tokens.
• A code review of the portal branch, 182 commits, none of it my own work. See branch-review-0828.md.

What is open elsewhere?

• #995, the admin Banned stat counting expired bans. Located, not fixed, auth territory.
• #996, a rejected offer having no home while the reason the API sends is read by nothing.
• The branch review's findings, all routed to the lanes that own them and none fixed here.

What should whoever picks this up NOT redo?

• --muted-dark-2 is SPLIT and gone (2026-08-30). Four tokens now: --text-secondary for the 14 text uses, --glyph-dim for 5 aria-hidden marks, --border-strong for 4 control rings, --chart-neutral for the chart series. The earlier count of 19 text / 5 non-text folded the five decorative marks into text; they are not text. --secondary-accent-blue-500 is still unsplit. See scenario 23 in the test run.
• The set-state-in-effect warnings in the offer builder are not a reverting field. Tested by typing, navigating away and back. Do not spend the afternoon on that hypothesis again.
• The per-placement pages rendering no text inputs is the inheritance model working, not a broken page.
• M5 in the design survey is closed, not open. Its first option caused #988 and she has ruled on the empty rail: leave it.

What is worth distrusting in my own record?

• The axe sweep's BEFORE column undercounts the report-only routes. It waited for a loading word to vanish rather than for content, so Account reads clean before and shows six failures once its card exists. Before and after are comparable for the eleven builder surfaces only.
• Six readings this session were false and were caught, five by a harness fault and one by me. They are named in the project docs rather than tidied away, because the pattern is worth more than the individual results.