Reduce reader startup work after merged Lighthouse improvements

[abhi1693]

What changes and why?

Signed-out readers currently download authenticated reading tools, and the closed Dev Card invitation mounts its large SVG and requests artwork images during initial page load. Load Must Reads and the streak panel together after sign-in, and mount the invitation content when its existing dialog opens. Keep the current navigation, 30-second invitation delay, dialog coordination, editing, animation, and dismissal behavior.

Latest and article previews remain the first priorities from the production Faro sample used for the preceding performance work (137 and 121 normalized navigation events respectively over seven days; sampled events, not pageview counts). This follow-up measures against the newly merged master, 8b25ac2f, rather than the earlier feature branches. No API bottleneck was identified that requires a backend change for these improvements.

Validation

Hosted Lighthouse comparison: merged master 8b25ac2f (baseline run) → final PR f76300a1 (run and reports, automated Lighthouse comment). Three cold mobile loads per page, unchanged fixtures/targets, median timings, largest transfer sizes:

Page	LCP before → after	TBT before → after	JS before → after	Total before → after
Latest	2.74 → 2.58 s	215 → 107 ms	252 → 247 KiB	930 → 525 KiB
Article preview	4.39 → 2.58 s	266 → 116 ms	256 → 251 KiB	943 → 542 KiB
Topics	2.60 → 2.37 s	64 → 37 ms	267 → 231 KiB	403 → 361 KiB
Sources	2.96 → 2.53 s	64 → 38 ms	277 → 267 KiB	413 → 398 KiB
Public profile	2.54 → 2.23 s	60 → 31 ms	257 → 221 KiB	368 → 327 KiB
Leaderboard	2.22 → 1.97 s	182 → 69 ms	225 → 188 KiB	319 → 278 KiB

Hosted advisory warnings fall from seven to three, and every page passes the 200 ms TBT target. Sampled performance scores rise from 92 → 96 on Latest and 79 → 96 on article previews. Timings vary between runs, particularly the hosted article baseline; the local comparison below shows smaller paint/score improvements and mixed TBT. Transfer savings are consistent in both environments. Latest/article LCP remains 2.58 s and Sources 2.53 s against the 2.5-second target.

Local comparison against the same merged master, before the subsequent error-boundary fix, using the same three-load method:

Page	LCP before → after	JS before → after	Total before → after
Latest	2.67 → 2.58 s	252 → 247 KiB	930 → 525 KiB
Article preview	2.72 → 2.57 s	256 → 251 KiB	943 → 542 KiB
Topics	2.59 → 2.32 s	267 → 231 KiB	403 → 361 KiB
Sources	2.72 → 2.60 s	277 → 267 KiB	413 → 398 KiB
Public profile	2.52 → 2.21 s	257 → 221 KiB	368 → 326 KiB
Leaderboard	2.21 → 2.04 s	225 → 187 KiB	319 → 278 KiB

Startup transfer falls 44% on Latest and 43% on article previews; both shed 437 DOM elements. Advisory LCP warnings fall from five to three. Median TBT remains under 200 ms on all six pages but is mixed: Latest 170 → 161, article 180 → 189, Topics 37 → 52, Sources 40 → 49, profile 40 → 37, leaderboard 109 → 129 ms. Latest, article previews, and Sources still exceed the 2.5-second LCP target. Deferred invitation assets load when the invitation opens.

• npm run web:lint: passed.
• npm run web:test: 1,171 tests passed on the final commit in hosted CI, including deferred content, account switching, and rejected-chunk regressions. Commit hooks also reran web lint, types, and unit tests.
• npm run extension:check and npm run extension:test: passed, 29 tests.
• npm run ci:browser-measure: passed; no fixture or budget changes.
• npm run reader:test:parity: web and all five browser suites for each built Chrome/Edge extension passed, covering invitation timing/animation/editing, authenticated reading tools, responsive layouts, and accessibility. Hosted reader parity also passed on the final commit.
• Fixed and resolved the review finding about rejected optional reading-tools downloads: a local error boundary now keeps the reader and other header actions available; a rejected-import regression passes.
• Final commit f76300a1: CI required, CodeQL, SonarQube, hosted reader parity, backend checks, security scans, and ARM64 image checks all passed. No unresolved review conversations; auto-merge remains disabled.
• Native GitHub attachment upload is unavailable in this environment. Browser screenshots stay local and untracked; no visible design changes are intended.

Documentation does not need an update: public behavior, configuration, and API contracts remain the same; this description records the performance measurements and remaining gaps.

Ready for review

• The description explains the change and the validation results, including any gaps.
• Relevant documentation and regression tests are updated, or the description explains why they aren't needed.

[github-actions]

PR reader API performance

Base 8b25ac2 → PR f76300a

Not measured: public API inputs are unchanged. This PR changes only recognized frontend or browser-test files.

Full HTML/CSV reports and logs · attempt 1

[gitar-bot]

Code Review ✅ Approved 1 closed / 1 findings

🟡 Medium risk · Defers authenticated reading tools and invitation artwork until account or dialog activation.

Reduces reader startup work by deferring hidden reader tools and invitation artwork until needed, and contains optional reading-tools chunk download failures. Median total transfer size drops 44% on Latest and 43% on article previews; LCP warnings fall from five to three across cold mobile Lighthouse loads.

✅ 1 closed

✅ Edge Case: Lazy reading-tools chunk failure can take down the whole shell

📄 apps/web/src/components/reading-tools.tsx:6-16
AccountReadingTools used to be a static import. It now loads through React.lazy, wrapped only in Suspense. If the chunk fails to load (flaky network, or a deploy that removes old chunk hashes while a tab is open), the rejected import throws during render and goes to the nearest error boundary. That replaces the page or header with the error UI, when the only loss should be the optional Must Reads and streak buttons. Wrapping the lazy component in a small error boundary that renders null keeps the failure contained.

Review coverage

📋 Rules No rules evaluated

🧪 Functional validation Not enabled · Set up

Options

Auto-apply is off → Gitar will not commit updates to this branch.
Display: compact → Counting what did not apply, without listing it.

Comment with these commands to change the behavior for this request:

Auto-apply	Compact
gitar auto-apply:on	gitar display:verbose

_{Was this helpful? React with 👍 / 👎 | Gitar}

[github-actions]

Backend unit coverage

PASS · minimum combined coverage 67.60%.

Tests: 3508 passed, 0 failed, 0 skipped. Source files: 276.

Workspace	Lines	Branches	Combined
All backend code	72.17%	54.31%	68.33%
apps/admin-api	65.12%	36.07%	60.29%
apps/aggregator	73.94%	57.18%	70.13%
apps/api	72.77%	58.45%	70.13%
apps/article-enrichment-worker	80.00%	50.00%	71.43%
apps/cli	68.76%	42.93%	64.81%
apps/images-worker	80.00%	50.00%	71.43%
apps/mcp	82.99%	73.88%	81.28%
apps/notifications	83.33%	76.67%	81.75%
apps/search-indexer	85.48%	85.71%	85.53%
apps/source-discovery-worker	80.00%	50.00%	71.43%
apps/user-api	59.37%	33.33%	54.50%
packages/core	74.09%	57.12%	70.06%
packages/http	94.94%	87.77%	93.29%

Unit coverage includes every backend workspace. PostgreSQL/Redis integration tests run separately.

Commit: f76300a · HTML, JSON, XML reports and job logs · attempt 1

[github-actions]

Lighthouse · ⚠️ Targets exceeded

⚠️ 3 target warnings · Bold values need attention.

Page	FCP	LCP	CLS	TBT	JS	Total
Latest (/latest)	1.07 s	2.58 s	0.001	107 ms	247 KiB	525 KiB
Article preview (/articles/…)	1.06 s	2.58 s	0.005	116 ms	251 KiB	542 KiB
Topics (/topics)	1.06 s	2.37 s	0.000	37 ms	231 KiB	361 KiB
Sources (/sources)	1.07 s	2.53 s	0.000	38 ms	267 KiB	398 KiB
Public profile (/users/…)	1.06 s	2.23 s	0.000	31 ms	221 KiB	327 KiB
Leaderboard (/leaderboard)	0.91 s	1.97 s	0.000	69 ms	188 KiB	278 KiB

Mobile reader · Three cold loads per page · Median timings · Largest transfer size.

Needs attention

Page	Finding	Measured	Target / limit
Latest (/latest)	⚠️ Target · LCP	2.58 s	2.50 s
Article preview (/articles/…)	⚠️ Target · LCP	2.58 s	2.50 s
Sources (/sources)	⚠️ Target · LCP	2.53 s	2.50 s

Full reports and logs · f76300a · Attempt 1

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues
0 New dependency risks

Measures
0 Security Hotspots
100.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud