test: mainnet readiness sign-off — Lighthouse regression check, governance latency SLO, trust-critical walkthrough, launch notes

[thegreatfeez]

Summary

This PR resolves four issues from the Final SCF/Mainnet Frontend Readiness Sign-off category. Each issue has its own commit.

• Run the full Lighthouse CI suite against this batch's changes and address any regression #959 (Lighthouse regression check): Ran the repo's own .lighthouserc.json on this batch and on the last pre-batch dev commit (579706e). There is no Core Web Vitals regression beyond run-to-run noise. The check found two real problems, both fixed here:
  1. next build was broken on dev by two type errors from chore: soroban.ts contract-ABI audit, OracleBadge oracle states, defer updateLPWhitelist, FUNDING.yml (#450 #783 #784 #785) #965.
  2. The Lighthouse workflow never ran on dev PRs, because it only triggers on main/develop.
• Confirm SLOs still hold under this batch's new real contract-call latency profile #960 (SLOs vs. real governance latency): Governance writes are still mocked: castVote, executeProposal, vetoProposal, and createProposal wait on a fixed setTimeout and return a fake tx hash. The full write path therefore can't be timed yet. Instead, a new script times every RPC stage a real governance write will go through, against live testnet: the read, the account fetch, prepareTransaction, and a real submit → confirmed. Those numbers are recorded as a new SLO 5 in docs/slos.md. All measured stages meet their targets.
• Conduct final maintainer sign-off walkthrough of every dark-feature and admin-surface change in this batch #961 (maintainer walkthrough): Added the live walkthrough script (a real vote, a dark-flag flip, and an admin action through the confirmation/audit-log flow) and a blank sign-off record. I traced each step through the code while preparing it, and several steps can't currently show what they claim. These are recorded as findings F1–F8, and the checklist item is marked Blocked.
• Update mainnet-launch-notes.md's overall readiness narrative to reflect this batch's completion #962 (launch notes): Added a Launch Readiness Status section to mainnet-launch-notes.md, tied to the readiness checklist. It covers governance (read-only; writes simulated), dark features, what admin hardening is real and what doesn't reach the chain yet, the Lighthouse outcome, and the remaining blockers. It also fixes the claim that governance is a live contract interaction.

Related Issue

Closes #959
Closes #960
Closes #961
Closes #962

Maintainers — about #961: this PR adds the walkthrough script and the sign-off record. It does not record a sign-off, because the live session hasn't happened and the table is intentionally empty. Please run the walkthrough and fill in docs/trust-critical-surface-walkthrough.md §4 before merging. If you'd rather keep #961 open until then, change its line above to Refs #961.

Complexity

• Trivial
• Medium
• High

Type of Change

• Feature
• Fix
• Docs
• Perf
• Test

Detailed Changes & Rationale

1. Key Changes

• Build fix (Run the full Lighthouse CI suite against this batch's changes and address any regression #959): UPDATE_LP_WHITELIST_SUPPORTED is now typed boolean, not the literal false, so the runtime feature check type-checks. encodeReferralCodeBytes now passes a Buffer to xdr.ScVal.scvBytes. next build and tsc --noEmit pass again.

• Lighthouse workflow (Run the full Lighthouse CI suite against this batch's changes and address any regression #959): now also triggers on dev. docs/LIGHTHOUSE_CI.md now:

  • lists the page actually audited: /lp, not /dashboard/lp;
  • states that every budget is warn-level;
  • includes a baseline-vs-candidate results table with the verdict and residual risks.

• Governance latency (Confirm SLOs still hold under this batch's new real contract-call latency profile #960): scripts/measure-governance-latency.mjs measured the following (Soroban testnet, n=20; submit n=10):

  Stage	p50	p95
  Governance read	323 ms	498 ms
  Account fetch	268 ms	347 ms
  prepareTransaction	283 ms	583 ms
  Submit → confirmed	4,977 ms	5,830 ms

  The targets are: read p95 ≤ 1 s, pre-signature preparation p95 ≤ 1.5 s, and submit → confirmed p95 ≤ 10 s. docs/slos.md also recommends a distinct "Confirming on-chain…" UI state once writes are wired, because a real vote takes about 5–6 s against the mock's 2 s.

• Walkthrough findings (Conduct final maintainer sign-off walkthrough of every dark-feature and admin-surface change in this batch #961): these came from reading the code, not from running a live session.

  #	Finding
  F1	Vote casting is mocked
  F2	Proposal execution is mocked
  F3	Protocol pause/unpause is mocked, although the dialog says "will call the contract"
  F4	Token approve/remove is signed but never submitted, yet the UI reports success
  F5	NEXT_PUBLIC_GOVERNANCE_CONTRACT_ID is unset on testnet, so governance reads fall back to mock proposals
  F6	DelegationPanel uses hard-coded values
  F7	Dark flags are build-time only
  F8	The admin audit log only covers signer rotations and parameter updates

2. Implementation Details

• The Lighthouse baseline commit doesn't type-check (an error the batch later fixed). It was built with type checking skipped, which doesn't change runtime output. Both builds ran back-to-back on the same machine: Node 20.20.2, @lhci/cli 0.15.1, 3 runs per URL, medians reported.
• / shows a higher median LCP (2,046 → 2,343 ms), but the individual runs overlap (1,650–2,051 vs 1,858–2,420 ms), and the LCP element is the same hero image in every run.
• The governance read and cast_vote preparation targeted the address the app actually resolves as the governance contract on testnet. Neither function exists there, so the RPC round trip is real but the simulation returns a contract error.
• Submit → confirmed was measured with a friendbot-funded throwaway account, submitting a read-only native XLM contract balance call. It goes through consensus but changes no state.
• The readiness checklist gains a Blocked row for the walkthrough, and its Lighthouse, SLO, and launch-notes rows are updated. The repo's pre-commit hook ran prettier over the checklist, which is why that file's diff is larger than the rows changed.

Files Changed Summary

• Created:
  • __tests__/lighthouse-config.test.ts
  • __tests__/governance-latency-measurement.test.ts
  • scripts/measure-governance-latency.mjs
  • docs/trust-critical-surface-walkthrough.md
• Modified:
  • .github/workflows/lighthouse.yml
  • src/utils/soroban.ts
  • docs/LIGHTHOUSE_CI.md
  • docs/slos.md
  • docs/mainnet-launch-notes.md
  • docs/mainnet-frontend-readiness-checklist.md
• Deleted: none

Breaking Changes

• No breaking changes
• Yes (describe impact and migration steps below):

Test Coverage

• Unit tests added/updated (npm test):
  • lighthouse-config.test.ts checks the workflow triggers on dev, that every audited URL maps to a real route, and that the doc's page list matches.
  • governance-latency-measurement.test.ts covers the percentile, summary, and timing helpers.
• E2E tests added/updated (npm run test:e2e)
• Storybook stories added/updated for new/changed components
• Manual testing completed: two Lighthouse CI runs (baseline and candidate), and a live testnet latency run including 10 real transaction submissions.
• Coverage thresholds maintained (90% lines/functions, 80% branches)

Local CI mirror (Node 20.20.2, pnpm 9.0.0):

• Build: pnpm run build passes. It fails on dev.
• Type check: tsc --noEmit passes.
• Tests: 2,362 passed and 44 failed. The same 44 fail on current dev, and none fail only on this branch.
• Lint, format, env checks: these fail identically on dev. ESLint shows the same 140 errors / 247 warnings on both, prettier issues went from 45 files to 43, and the file-size and env checks fail on both. Nothing new comes from this PR.
• Note: ci.yml also only triggers on main/develop, so the main CI workflow doesn't run on PRs to dev. I left that alone because it's outside these issues.

Accessibility

• Keyboard tested: Yes
• Keyboard tested: No / Not applicable

Checklist

• Code follows project style guidelines (npm run lint)
• Self-review completed
• Comments added for complex logic
• Documentation updated (README, DESIGN.md, architecture docs) if needed
• No new console warnings or errors
• All tests passing locally (44 failures pre-exist on dev; see above)

Residual / accepted risk

• Governance and admin writes still don't reach the chain (F1–F4, F8). This PR documents them and marks them as launch blockers. Wiring them is Add a prominent 'not yet live' UI indicator for any governance action still backed by a mock #850/Publish a governance write-path launch-readiness gate tying this whole category to a single sign-off #852.
• The governance write-path SLO is unverified end to end until a governance contract is deployed and configured. The script can then be re-run against it.
• Pre-existing Lighthouse breaches: home CLS ≈ 0.13 (budget 0.1), and every page is above the 200 KB byte budget. The biggest contributors are the 1.1 MB icon font and the home page's Horizon transaction-history fetches.
• Contract drift: the deployed testnet invoice contract no longer exposes get_invoice_count, which src/utils/soroban.ts still calls.

[vercel]

@thegreatfeez is attempting to deploy a commit to the Nursca's projects Team on Vercel.

A member of the Team first needs to authorize it.

[drips-wave]

@thegreatfeez Great news! 🎉 Based on an automated assessment of this PR, the linked Wave issue(s) no longer count against your application limits.

You can now already apply to more issues while waiting for a review of this PR. Keep up the great work! 🚀

Learn more about application limits