Ship Geist as WOFF2
Both faces are preloaded on every route, so the format is paid on every first visit. 134,132 B of WOFF became 112,252 B of WOFF2 — 21,880 B, a 16% cut. That is...
A retail market should not download the mortgage tabs
The market settings page imported all 24 tabs statically, so every market shipped every tab — a retail market carried the seven mortgage tabs, the four personal...
Show the first wallpaper without waiting for the slowest
The slideshow preloaded every wallpaper behind a single Promise.all, so nothing painted until the LAST image in the set had settled. On a tenant with a dozen ba...
Every context and hook now reads user-info through the coalescer
Seven independent /api/user-info calls fired on a cold authenticated load, each one asking for the member id a sibling was already asking for at the same moment...
One refresh timer for the whole app, not one per useUserInfo()
useUserInfo has 194 call sites, roughly a dozen of them mounted for the entire session — the nav, the shell providers, the launchpad. Each instance armed its ow...
Server-render the authenticated app, and fix the fallback org that 500s
MemberListProvider returned null until its mount effect ran, and it sits above NavTop AND {children} — so that one line kept the ENTIRE authenticated applicatio...
A checkbox click should not re-render the folder sidebar
Six call sites read the email store with a bare `useEmailStore()`. In zustand that subscribes to the WHOLE store, so any change anywhere in it re-renders the co...
Memoize the app-wide unread value, fix a self-retriggering effect
UnreadContext published a fourteen-member object literal as its context value, so React saw the context change on every render of the provider and woke every co...
Stop polling for badges nobody is looking at, and stop waking every consumer
Two always-mounted providers ran a one-minute poll with no visibility gate. LegalPackageCount had none at all; TrainingPendingCount had a visibilitychange re-re...
A phone-less org should not open a call socket
The softphone opened its SockJS/STOMP connection to media.<domain> unconditionally at boot. getCapabilities() is awaited a few lines above and is the backend's ...
Stop re-merging the whole catalogue on every request
getRequestConfig runs per request, and for any non-English member it called mergeWithFallback over the entire 23,252-key catalogue every time — rebuilding a res...
One arriving message should not re-render the whole history
Batch 5 of the client-side performance audit. MessageList maps every loaded message to a MessageBubble, and MessageBubble was not memoized. With 300 messages o...
Take AG Grid's 187 KB of CSS off the critical path of every route
Batch 4 of the client-side performance audit, and the last of the big ones. app/globals.css opened with two @imports of ag-grid's stylesheets, and the root lay...
Stop sending the whole message catalogue to the sign-in handoff
Batch 3 of the client-side performance audit. The root layout passed every message to NextIntlClientProvider on every route. The catalogue is 1.05 MB compact —...
Split the tool windows out of the shell, stop the idle timer storm
Batch 2 of the client-side performance audit. As with batch 1, nothing a member can see or do changes. The tool registry imported all ten tool windows eagerly,...
Stop shipping the shell code no route can reach
Batch 1 of the client-side performance audit. Every change is behaviour-preserving; nothing a member can see or do is different afterwards. I18nProviders impor...
Drop hasDesignVersion and its per-row template lookup
Every document list ran an esign_templates lookup for EACH row — 100 per page — to answer "is this an original that happens to carry a design". That was how the...
Stop scanning every market x product after the answer is known
/leads/available-summary returns three things — hasAcceptable, hasAcceptableCredit and one example product — but visited every market x product pair regardless,...
Index the two shapes geolite_locations is actually queried by
Country alone was the only index on an 80,429-row table, and neither real query used it well. Measured on production: city resolution (country + subdivision)...
Count in the database instead of hydrating every commit
GET /changelog/public/stats loaded all 16,571 commit logs as entities and tallied them in a Java loop to produce two small maps and a total. It takes no authent...
Index member_rights_applied, bound the intake claim, aggregate changelog stats
Three findings from an endpoint-by-endpoint audit, each measured against production rather than inferred. member_rights_applied declared no indexes at all whil...
Fetch one server-filtered page instead of the whole organisation
The grid fetched every lead in the org and narrowed it in a ~300-line useMemo. Measured on production 2026-08-20: 1,427 leads fetched so seven could be shown, 2...
Page the grid endpoint and take the PHI audit write off the request thread
GET /api/security/leads returned every lead in the organisation. It now takes page/size/sort plus the filters the grid actually uses, and returns a page with a ...
Filter and page the grid in the database, batch the PHI audit trail
Loading /leads took 23.3 s at 1,427 leads. 23.3 s of that was the HIPAA audit trail: recordLeadList writes one phi_access_log row per lead returned, and each on...
Compress every response, not just the document
server-nonce.js asks upstream for `identity` so the HTML can be stamped with a CSP nonce without a decompress/recompress round trip, then recompresses on the wa...
Drain deferred work one item per idle period
Four components waited for the main thread to go quiet before doing their work — the toast/consent bundle, the scroll chrome, the touch target sweep and the ref...
Read the scroll position from an observer, not from layout
The nav's scrolled/at-top state came from a scroll handler that read window.scrollY, and the mount effect called that handler synchronously during React's commi...
Serve static assets as build-time brotli
The flagged chunk is react-dom, and I checked it properly before concluding anything: it is the production build (no dev markers, and the profiling hooks `injec...
Drop the unused preconnect, and fix the 565 ms of brotli I added
Two findings, and the second one was mine. THE DOCUMENT TOOK 1,259 ms BECAUSE OF THE NONCE WRAPPER. Measured against the live site: the HTML answered in ~900 ...
Move every looping animation onto the compositor
All 26 elements Lighthouse listed under "avoid non-composited animations" were the same mistake in four places: animating a paint property. Measured with Chrome...
63 HTTP requests down to 22
Implements the request-budget report. Measured the same way it was written — Chrome over the debugging protocol, mobile emulation, sixteen-second window. 6...
The reel was never using hls.js, and phones paid for it
Chasing "make fewer HTTP requests" found two real things, and the second is much larger than the audit that led to it. THE HLS BRANCH NOBODY WAS TAKING. HeroSt...
Stop re-measuring the document on every scroll frame
Two scroll listeners on the homepage read `scrollHeight` per event. That is a layout query, and this page is the worst place to make one repeatedly: its section...
Drop backdrop-filter on phones
The homepage renders 94 elements carrying `backdrop-filter: blur(20px)`. Every one is a separate compositing layer that has to read back everything painted behi...
Take the footer and the scroll chrome off the client
Three changes, all removing work from the window where the page is being judged. `(pages)/layout.tsx` carried `"use client"` and never needed it — no state, no...
Stop three resources from outranking the LCP element
The LCP breakdown is where the remaining time is, and its two loading subparts — 270 ms of resource load delay, 500 ms of load duration — are a priority problem...
Remove gsap, the second animation runtime on the homepage
Framer was the chunk Lighthouse named. GSAP is the one it did not, because it arrives late enough to fall outside "unused JavaScript" — but it is 69 KB in its o...
Remove framer-motion from the site
"Reduce unused JavaScript" has named the same 35.4 KB chunk in every report, and it is Framer Motion. It is not an eager script on the homepage, which is why it...
Resolve the org's icons on the server and preconnect the origin
Lighthouse named theme.kamocrm.com as an unconnected origin worth 300 ms while also reporting no origins preconnected, and last round I read that as a stale com...
Stop three route bundles being prefetched onto the homepage
Lighthouse's 103 KiB of "unused JavaScript" named four chunks, two of them essentially 100% unused. None of them appear in the homepage's HTML — they are other ...
Name the backdrop widths, and read the theme config on the server
Two findings, both of which were doing something other than what the code said. THE BACKDROP WAS FETCHING THE 1920 CANDIDATE. 149 KB, for a decorative picture ...
Compress the images harder, hold the cards still, drop duplicate requests
Images. Every quality setting on the page goes down a step, and the allow-list in next.config.ts goes to [30, 50] so nothing can quietly ask for more: hero...
Split the hero, and take the last two libraries off the page
Step C. Neither GSAP nor Framer Motion is on the homepage any more, and the `home` message namespace is no longer sent to the browser at all. JS 24...
Seven sections stop hydrating
Step B. TrustBar, WhatReplacesWhat, HowItWorks, UseCases, Features, CTAStrip and Pricing are server components now. They render to HTML and ship no JavaScript. ...
Take the animation runtime off the homepage entirely
Step A of the hydration work. Framer Motion is gone from the homepage's eager bundle: 798,723 -> 687,593 bytes raw, 245,367 -> 209,422 gzip. That is 35,945 fewe...
Stop rendering eight off-screen sections before the hero can paint
LCP was 4.4 s, and its breakdown said where: 120 ms to first byte, 290 ms of load delay, 390 ms actually downloading the image — and then 1,810 ms of "element r...
Undo the two changes that cost 380ms of blocking time, and quiet the phone
The last two rounds took the score from 87 to 67. Total Blocking Time went from 80 ms to 460 ms and LCP from 3.9 s to 4.4 s. Two of my own changes did that, and...
Move the headline sweep onto the compositor, and the chat widget off load
Two things left over from the last round, both flagged rather than done because each trades something. Doing them now. THE HEADLINE'S LIGHT PASS. It was a seco...
Inline the stylesheet, and drop polyfills nothing can reach
Two Lighthouse findings, both of which stayed open after the last round because I had concluded they were not addressable. They were; I had not looked hard enou...
Like what you see shipping?
All of it arrives in your workspace on its own. Start on the free plan and read this page again in a month.