Velocity Forge – Core Web Vitals, Page Speed & Performance
| 开发者 | avakodeforge |
|---|---|
| 更新时间 | 2026年11月9日 11:35 |
| PHP版本: | 7.4 及以上 |
| WordPress版本: | 7.1 |
| 版权: | GPLv2 or later |
| 版权网址: | 版权信息 |
详情介绍:
Velocity Forge measures Core Web Vitals on your own site, from your real visitors: LCP, INP, CLS, TTFB and FCP. Every measurement is stored in your own database. The measuring script is served from your site, nothing is loaded from a CDN, and no field reading leaves the site.
It does not stop at the number. A slow interaction is reported with the control that was hit, and whether the time went on a busy main thread, on your own handler or on the repaint afterwards. A slow first paint is reported with the element and the file it loaded. Two cards on the Monitoring screen list the ten controls that cost the most interaction time and the five elements that delay the first paint, each with the file and the plugin behind it. A figure is shown only once an element has at least three measurements.
When an update slows the site down, the plugin tells you what changed. Before an update replaces any file it takes a snapshot of the scripts and stylesheets your front page loads, and afterwards the regression row names the assets the update added, removed or resized. You can set budgets for LCP, INP, CLS and TTFB, and the site owner gets an e-mail when a regression lands.
The changes the free plugin makes are small and reversible. It adds LCP hints to the detected hero image (
fetchpriority="high" and a preload), decoding="async" to images and lazy loading below the fold, and speculation rules for prefetch and prerender with a cart, checkout and add-to-cart exclusion that cannot be switched off. Safe Mode disables all of it in one click, from the admin bar as well, and a recovery drill lets you test the escape hatches on a day when nothing is wrong. Velocity Forge is not a page cache: it recognises seventeen caching and optimisation plugins and stands aside where their work overlaps.
Free
- Core Web Vitals collection from real visitors (LCP, INP, CLS, TTFB, FCP), 30 days of history by default, up to 365 days if you ask for it in Settings
- Element, file and plugin behind each slow measurement
- Trend chart of the 75th percentile over 30 days, with the same numbers in a table
- Regression monitor with an asset difference after updates, and an e-mail to the site owner
- Performance budgets for LCP, INP, CLS and TTFB
- LCP hints and lazy loading, speculation rules
- Page profiler: audit single pages or batches of up to 50 URLs of your own site
- Asset inventory: every script and stylesheet handle with its owner
- Database health: autoloaded options, expired transients, overdue cron, and the heaviest autoloaded options with the plugin that wrote them
- Cache plugin coexistence
- Safe Mode, rollback manager and recovery drill
- The Optimiser: asset rules (defer, async, delay until interaction, dequeue, preload, preconnect), critical CSS, WebP and AVIF image conversion, autoload and transient cleanup
- Rules that take themselves back when real visitors measure a regression
- Regression alerts to Telegram, Slack or your own webhook
- CSV export of field data
- AI Performance Advisor, which reads your field data and returns an ordered plan without applying anything
- Scheduled lab audits through your own PageSpeed Insights key, which also check the Total Blocking Time budget The free plugin's own requests all go to your own site. Details are under External services.
安装:
- Upload the
velocityforgefolder to/wp-content/plugins/, or install through Plugins → Add New. - Activate VelocityForge through the Plugins menu.
- Open VelocityForge → Dashboard to run your first page audit.
- Visit VelocityForge → Settings to tune thresholds and enable modules.
常见问题:
Can I run it alongside my caching plugin?
Yes. Velocity Forge is not a page cache and does not want to be one. It recognises seventeen caching and optimisation plugins, and its overlapping work stands down while one of them is active. The notice names every feature that stepped back, not only the obvious overlap.
An update slowed the site down. What does Velocity Forge tell me?
Which assets changed. Before an update replaces any file, the plugin fetches your real front page and reads its script and stylesheet queue. Afterwards the regression row reads the difference and names what was added, what disappeared and what grew. "LCP is up 400 ms" is a symptom; "this update added three render-blocking stylesheets" is something you can take to its author.
What happens if I break something?
Every change has a rollback path. Revert the single rule, press Safe Mode in the admin bar on whichever screen you noticed the problem, use the secret panic URL shown on the Rollback screen, or add define('VELOCITYFORGE_PANIC', true); to wp-config.php. The wp-config.php line is checked before any database read, because on the day you need it the database is often the problem.
How do I know the escape hatches work before I need them?
Run the recovery drill on the Rollback screen. It presses the real hatches, including a genuine loopback request to your secret panic route, and puts everything back exactly as it found it. It writes no files.
Does it send my data anywhere?
No measurement leaves your site. The free plugin makes requests of its own, and every one goes to your own site: the page profiler fetches the pages you audit, the regression monitor reads your front page after an update, the asset inventory sends a HEAD request to each script and stylesheet your pages load from this site, and the recovery drill calls your own panic route when you press its button. The free plugin sends no telemetry.
Do I need an account or API key for the free version?
No. Measuring and diagnosing happen entirely on your own server. See External services for exactly what this download does and does not contact.
Can Velocity Forge undo a rule on its own?
Only with the Pro add-on, only for a rule you have just created, and only after refusing six ways: no licence, the beacon switched off, fewer measurements than your sample floor, Safe Mode on, a protected handle, or a rule you pinned. When it acts, it reverts that one rule, records the reason on the rollback event and e-mails you the numbers and how to put it back. Below the sample floor the verdict is "inconclusive" and the rule stays as it was.
更新日志:
1.4.2
- AI advisor: with the VelocityForge Pro add-on (1.4.0 or newer), a request whose answer is lost on the way (the connection drops, or the reply cannot be read) is no longer charged a second time when you try again. The Audit screen says so, and says how many seconds to wait when the first request is still running. If the service refuses a request, the screen shows its reason.
- Uninstalling removes the small records kept for these repeats.
- Background jobs: on a site where WP-Cron is switched off (DISABLE_WP_CRON), queuing a job no longer sends the request that wakes WP-Cron. The site's own cron runner starts the job, as you set it up.
- Privacy: Velocity Forge now remembers a page by its web address only. Everything after a question mark or a hash in a link (an e-mail address, a password-reset code, a campaign tag, a signed image link) is no longer stored, and the measuring snippet no longer sends it. On sites without pretty permalinks, the few parameters WordPress uses to tell pages apart are kept when they hold a number or a short name, so those pages are shown separately instead of being folded into the home page. An order number that is part of the address itself is still stored, and anything named like a key, token, password, e-mail or session is never kept.
- Addresses collected by earlier versions are cleaned up in the background, a little at a time, so the site does not slow down. If your host has switched WordPress's scheduled tasks off, the clean-up runs while an administrator is working in the WordPress admin, at most once every few minutes. An image detail that cannot be read and may contain an address is emptied. The pages on your Important pages list are tidied once into the same plain form: a stray tail pasted after an address is removed and two ways of writing the same page become one entry. Everything saved for those pages keeps working, so nothing has to be set up again.
- Field measurements: visitors behind Cloudflare are now counted one by one (their address is read from the CF-Connecting-IP header, but only on connections from Cloudflare's published ranges) instead of sharing one quota and losing measurements. An IPv4 address written in IPv6 form (::ffff:a.b.c.d) counts as that IPv4 address, so a forged header cannot get round the per-visitor limit. One request no longer uses up two units of that hourly limit, a request body over 16 KB is refused with a 413 answer, and pages that show the admin bar are not measured, so staff visits do not count towards the field figures. The visitor address is taken from the connection itself unless your site names a trusted proxy header.
- Regression checks: a page that looks slower right after an update is measured once more straight away, and a regression is recorded and mailed only if the second measurement is slower too, so the cold cache an update leaves behind no longer raises a false alarm. When a cold and a warm scan fall in the same second, the newer one counts as the latest. The re-check after an update covers the ten pages visitors opened most recently (the old query could leave a recent page out), counts a page once however its address is written, and gets the details of the update from the scheduled task (it used to receive only the word "plugin_update"). The regression e-mail is now sent through the shared Forge mail template; before, it could fail with an error and no mail went out.
- Reset to last known good now reads the history per metric: it goes back to before the oldest metric that is over budget began to slip, so a good reading of one metric (INP, say) that is newer than a bad reading of another (LCP) can no longer leave the rule that caused the regression switched on. A measurement that carries none of the budgeted metrics is not taken as a good moment. When there is no good moment at all, no rule is switched off, Safe Mode is turned on instead and the screen says so. Using the panic URL now writes a row in the Rollback journal.
- Page caches: rolling back, switching Safe Mode, using the panic URL, changing a rule, restoring a version or changing critical CSS now empties the cache of the detected cache plugins, so the site no longer keeps serving pages cached before the change. Saving a setting that no page prints (alert address, retention, usage statistics) no longer empties the caches, and the panic URL is retired before they are emptied. W3 Total Cache 2.x is now detected. When a page cache and an optimiser are installed together, the optimiser is recognised as the owner of the page's assets (the first match used to win), and image hints and speculative loading step back for it, as the notice already said. Saved preload and preconnect rules keep printing next to a page-only cache.
- Images: the hero image preload carries the same srcset and sizes as the image, at the size your theme prints, so the browser chooses from the same files for both instead of being pointed at the full-size file. The site logo (class custom-logo) is no longer taken as the main (LCP) image or made lazy, and an image keeps the loading and fetchpriority attributes its caller wrote; a lazy or low-priority picture is not taken as the main image either. Deleting an image from the Media Library also deletes its WebP and AVIF copies, with or without the Pro add-on and inside the uploads folder only. Uninstalling removes the pending image-conversion events.
- Assets screen: a page that redirects to the login page is no longer read as that page, and an asset from your own site's host outside the usual folders is listed as unknown instead of external. Saving a rule that removes a script or style that other assets depend on is held back with a warning until you tick "Dequeue anyway". Pages are read through the shared address check of the Forge plugins, which now resolves a host name once, connects to the addresses it checked and follows redirects one checked step at a time; where it refuses, it says why in plain words.
- Autoload screen: restoring an option from its backup now clears WordPress's cached copy of the options, so the restored value is used at once on sites with a persistent object cache. A refusal says what happened (protected option, empty option name, failed backup) instead of blaming the add-on. Uninstalling marks the backups it restored as restored, so a reinstall does not list those options as switched off.
- Clearer messages: a manual scan on the Audit screen says why nothing was scanned (no address entered, or none belongs to this site) and names the pages the server could not read, instead of saying "Audit started". Saving Important pages says "Saved 5 of N" when the list is cut, instead of dropping the rest without a word. Screens that depend on the Pro add-on (Assets, Monitoring, Images, Critical CSS) show a sentence instead of the bare code pro_required, and the Critical CSS settings are not saved without the Pro extractor. The Bulk generate line on the Critical CSS screen says each template is read from a page of its own kind, and the Recovery drill description lists the four checks it runs.
- Multisite: every site of a network now gets its scheduled events, its panic secret and its install stamp on its first request (the activation hook ran on one site only). Deactivating the plugin for the whole network clears the scheduled events on every site, and uninstalling removes the site-setup mark together with the panic secret.
- When AI credits run out, the message no longer promises a monthly reset and links to the page where you buy credits.
- The plugin is listed as "Velocity Forge – Core Web Vitals, Page Speed & Performance" in the Plugins list and on wordpress.org.
- The logo in the admin screens is the new mark.
- With the VelocityForge Pro add-on installed, Settings → Privacy & uninstall has a second box, "Also include sanitised error context", under "Share anonymous usage signals". It cannot be ticked while the usage box is empty.
- Turning the usage box off now also deletes the statistics queued and not yet sent and the stored error-context choice; the second box is empty when usage is ticked again. A save that does not carry the usage box (another setting saved on its own) no longer switches the choice off.
- New: when you have not named any pages to watch, the Monitoring screen offers “Suggested pages”: your home page, then the pages with the most recorded visits, or your shop pages, with WooCommerce, while there are none yet. Nothing is saved until you press “Watch these pages”.
- New: a page that is still short of observations now says how long it may take to get enough, in the readiness checklist and beside “Insufficient data” in the baseline table: “≈ 3 days at the current traffic”, “no visits recorded yet”, or “not within the baseline window at the current traffic” when the window cannot hold that many. It is an estimate, never a promise of a verdict, and the last wording is held back during the first two days of recording.
- Reading the busiest pages is limited to the newest 50,000 records, so the Monitoring screen opens quickly on a site with millions of records; a smaller site is read in full.
- The new lines are translated in all 15 shipped languages.
- Deleting the plugin leaves the job queue the Forge plugins share in place while another Forge plugin or add-on is still on the site. The check used to miss add-ons, and Rank Forge installed from wordpress.org, so the queue could be dropped while a neighbour still had jobs in it.
- Requires WordPress 6.6 or newer. The database health check asks WordPress which autoload values it loads on every request, and WordPress answers that only from 6.6. Sites on WordPress 6.5 stay on 1.2.5.
- The recovery drill checks the secret panic URL with a one-off secret instead of the site's own, and verifies TLS like any other request.
- The readme no longer describes the Forge Suite admin layer, which this download does not include.
- Field data is kept for up to a year on every install; the three-month ceiling without the Pro add-on is gone.
- The licence screen, the plan badge in the header and the upgrade card moved into the Pro add-on. This plugin only notes which features the separate add-on provides, with a link to its product page.
- A scan on the Assets screen runs only from the screen's own Scan button or "Create rule" link, which now carry a nonce.
- The readme lists the third-party hosts VelocityForge recognises on your pages and never contacts. Older versions are listed in changelog.txt in the plugin folder.