Lovable sync 2026-08-19 — run summary

Deliverable: .local/lovable/sync-20260819/port-plan.md (51 KB, 208 lines) · 29 change units · PLAN-ONLY, nothing under apps/dashboard was touched.

Provenance — one clean run, git-first

Unlike 2026-08-18, git fetch succeeded on this machine (gh account fibretraceniafam logged in), so origin/main is genuinely current and no MCP-only fallback was involved. The prior run’s single follow-up item is therefore closed.

Window

0df29e60 (last verified ported) → f0d02a28 “Updated all button styling” — 781 raw commits, 118 files, +11 970 / −3 041 lines, spanning 2026-08-16 19:57 UTC → 2026-08-19 05:33 UTC. That is roughly 2.5 days of continuous Lovable autosave, which is why the raw commit count is so high relative to the 21 meaningful non-SKIP units.

Lovable’s commit stream alternates a code-bearing Changes commit (gpt-engineer-app[bot]) with an adjacent near-empty descriptive commit carrying the human-readable message. MCP list_edits surfaces one entry per descriptive commit — 185 edits for this window, fully paginated. The WHY layer is strong: 29/29 units carry an mcpSha, and 24/29 carry Malcolm’s verbatim triggering prompt.

The 29 units

Totals: PORT 7 · REVIEW 14 · SKIP 8 · ALREADY_DONE 0.

Design-system rebrands — do first (3 units, all high, all watchlist hits): design-system-button-tabs-pagination · completeness-button-tsx · completeness-tabs-tsx. All three land on the same surface, so the plan correctly merges them into 2 file-scoped execution units (button.tsx; tabs.tsx + a brand-new table-pagination.tsx) while the CHANGELOG still lists all three input ids.

Ready to port, no BE dependency (4): partners-nav-and-map-refactor (medium — Partners moves out of Settings into main nav) · sdu-page-refactor (low) · supply-chain-tier-explorer-polish (low) · plus the 2 merged rebrand units above.

Blocked / BE-gap (12 resolved API specs): claim-position-page-rewrite (PORT, PARTIAL) and 11 REVIEW units. See the API table below.

Pure UI/copy decisions with no apiSpec (3): credit-units-terminology-sweep · certificates-scope-modal-hig-refactor · table-toolbar-and-shared-pagination-rollout. Low risk to port, but each is sequenced behind another unit — 18 in particular must follow the button.tsx table variant port.

SKIP (8, all justified): five admin-console clusters (admin-billing-cms-misc, admin-companies-programmes-refactor, admin-evidence-fibre-production-verifications, admin-new-pages-home-guide-overrides-reservations, plus site-subtext-global-setting) fall under the hard DO-NOT-PORT src/pages/admin/ rule · chatbase-widget-logo-persistence-fix (already tracked in known_unported) · connections-directory-share-refactor · supabase-schema-migrations-and-types (reference-only by project rule).

API resolution — every data surface this window is BE-blocked

12 specs resolved, all with verified: true (11 high confidence, 1 medium):

Zero EXISTS is the headline: nothing data-touching in this window can be wired today. All portable value this pull is visual/structural. Each GAP/PARTIAL carries a beSpecForGap in the plan, ready to hand to BE.

Flagged highlights — the watchlist call was correct

The launch note predicted src/components/ui/button.tsx changed and that any watchlist hit should be treated as a global rebrand ranked above per-page edits. Both held, and I verified the design-system surface independently of the workflow before it reported back:

The button.tsx divergence is genuinely larger than the 781-commit window shows. The window itself only adds tertiary and restyles table; monet additionally carries older unported drift that only the window-independent completeness pass surfaced — and it did surface all of it:

• action variant: monet bg-action (green) vs Lovable bg-primary (near-black) — the exact class of silent rebrand that slipped through in 2026-06. Only 1 call site (ProgrammeDetail.tsx:281), but it is a brand decision, not a detail.
• base radius rounded-[8px] vs rounded-[10px]
• CTA size scale: monet mixes h-[42px] / h-8 / h-10 and pins lg:min-w-[150px]; Lovable unified everything to h-9 with no min-width
• tertiary and table variants absent from monet entirely (not merely restyled); Button renders in ~86 files under apps/dashboard/src

One concrete portability catch the judge made: Lovable’s table variant uses shadow-2xs, which is not a real Tailwind v3 utility in either repo — substitute shadow-sm when porting.

table-pagination.tsx does not exist in monet at all; Facilities.tsx (L344-364), Verifications.tsx, Collections.tsx, SDU.tsx and Scanners.tsx each hand-roll an older inline pagination block. Lovable also tried and dropped a variant="mac" prop mid-window — port the settled shape, not the intermediate.

Gaps worth knowing before porting

1. The index.css completeness pass regressed — this is the one real defect in the plan. The plan states “index.css and tailwind.config.ts show no diff”. tailwind.config.ts is right; index.css is wrong. Monet is missing, versus origin/main:

• the .tier0-quiet block — 6 font-weight demotion rules for the Tier 0 home (.font-black → 800, .font-extrabold → 700, and so on down to .font-normal → 300)
• Mona Sans weight 800 in the Google Fonts import (wght@400;500;600;700 vs ...;800), which .tier0-quiet .font-black depends on

Traced to 90b49e35 (2026-06-27), which pre-dates even the old baseline — so it appears in no commit window and is invisible to every future window diff. The window-independent completeness pass is the only mechanism that can ever surface it; last run’s pass did catch it (as completeness-index-css-admin-tier0-tokens), this run’s dropped it because it compared only colour tokens, not typography or rule blocks. It has no known_unported entry either, so it is currently at risk of permanent silent loss. It is small, purely visual, has no BE dependency, and pairs naturally with home-dashboards-t0-macos-refresh.

The remaining ~50 lines of the index.css delta are the admin-console reskin (.admin-shell, .admin-content, admin nav-active rules), which is correctly not portable under the hard DO-NOT-PORT src/pages/admin/ rule. That is likely what led the pass to wave the whole file through.

2. Zero EXISTS means the claims/PO/nominations domain cannot be finished, only restyled. Nine GAP units cluster in one domain. Porting their visuals will produce more toast-guarded empty shells on top of the five already tracked in known_unported since 2026-06-25. Worth deciding deliberately whether to add shell surface area now or wait for BE.

3. Last run’s two cautions still stand. The .legacy.tsx mid-rewrite signal in the claims/PO domain is consistent with this run leaving every one of those units at REVIEW rather than PORT — treat them as provisional. The Purchase Order → Nomination copy rename is now carried by nominations-purchase-orders-rewrite, but it is still spread across units; porting piecemeal will leave the UI half-renamed.

Cursor changes

• last_fetched_sha: 438d32cc → f0d02a28 (prev_fetched_sha correctly captured as 438d32cc before overwrite). This is the only advance the run was authorised to make freely.
• last_ported_sha: 0df29e60 → 2b52a608, auto-advanced by the flow over its evidence-verified prefix. I verified this independently:
  • git rev-parse f3b4bddd~1 = 2b52a6089ef7462bffb2cd845f256931d09be893 — confirmed in the clone.
  • Old cursor is an ancestor of the new cursor, and the new cursor is an ancestor of origin/main — a clean forward move, 67 commits.
  • No unit sits entirely before the new cursor. All 21 non-SKIP units have at least one commit at or after 2b52a608, so every one of them re-surfaces in the next window. The two entries the flow added to known_unported (public-claim-fibre-creation-copy, reservation-term-expiry-negative-allowance) straddle the boundary rather than falling behind it, so they are belt-and-braces rather than strictly required. The advance loses nothing from this window — the index.css item in gap 1 is a pre-baseline miss, not something this advance stepped over.
• known_unported: 8 → 10 entries.
• suggested_ported_sha = f0d02a28, to accept only once units 1-5 of the plan are actually ported:

  python3 .claude/skills/monet-dashboard-lovable/scripts/state.py set-ported f0d02a2866a25dbd15e9a055cb786e41b3c58405
  

• 20 nf-tasks-compatible task drafts emitted; CHANGELOG.md prepended with the 2026-08-19 entry.

Follow-ups for the next sync

1. Record the index.css drift so it cannot be lost — it belongs in known_unported (or ported outright, since it is ~7 lines with no BE dependency). Nothing else will resurface it.
2. Harden the completeness pass to compare whole rule blocks, not just colour tokens. The step 6b prompt says “semantically diff the DESIGN TOKENS / variant definitions” and lists colour-ish examples, which is what let a typography-only rebrand through two runs in a row. Widening it to “any divergent rule or @import” would have caught this, and it is the second time the completeness pass has been the only net that matters.
3. Decide the shell-surface question in gap 2 before porting any of the nine GAP units.
4. Confirm the final CU label (“Claim Unit (CU)” vs “credit units”) so credit-units-terminology-sweep can go platform-wide — still open from 2026-08-18.