Repository navigation
feat(slop): free Slop Score engine — 50-page content/code/design sweep - #114
Merged
Merged
Conversation
Adds a free, deterministic scan engine that sweeps up to 50 same-origin
pages and scores how careless a site looks: 0 is pristine, 100 is maximum
slop. Output is the headline score, per-page fix lists, and systemic
rollups for defects that live in a shared template.
Discovery is sitemap.xml first (following sitemap indexes), then
breadth-first from the entry page, bounded by a 3-minute deadline so it
stays inside the worker's stuck-audit cutoff. A few same-origin
stylesheets are fetched so the design checks can see palette,
typography, and !important sprawl, which is invisible from HTML alone.
Analyzers (lib/audit/checks/slop.ts) cover three dimensions:
content — filler-phrase density, no first-party evidence, thin pages,
near-duplicate bodies (5-word shingle Jaccard), boilerplate
intros, placeholders, stale copyright, misspellings
code — leaked template variables, dev/staging hosts, console.* and
TODO leftovers, duplicate/missing metadata, dead links,
deprecated tags
design — missing viewport, unsized images, placeholder alt text,
stock-only imagery, inline-style density, style sprawl
Two design rules are load-bearing:
It reports observable defects, never "this was written by AI". An
AI-probability score is unfalsifiable, the classifiers are unreliable
(they systematically flag non-native English writers), and it would
accuse paying customers. Every finding is something an owner can verify
in ten seconds and fix.
Ambiguous markers only count in unambiguous positions. Dogfooding on our
own blog and on Hacker News turned up four false positives, each now
covered by a regression test: "a feature is coming soon" and "your brand
name in an AI answer" in running prose, "[product]"/"[your site]" as
editorial shorthand, and .netlify.app/.vercel.app appearing as scan-result
text or as an outbound link rather than a leaked URL. Hence
standalone-only phrase matching and attribute-only host scanning, with
preview hosts counted only in resource positions. Palette sprawl also
ignores custom-property definitions, so having design tokens is no longer
penalised.
Wiring: engine registered at cost 0 (lib/credits.ts), worker dispatch,
engines panel, report view with a SlopMeter hero stat, and an
audits_engine_check migration — without it every 'slop' insert would
violate the constraint, the same trap the fugu and zai migrations
document. Anonymous visitors can now pick between the AEO audit and the
Slop Score on the hero form; both are free, self-hosted, and
page-budgeted, and the per-IP anonymous limit still applies. Runs no LLM,
so it is immune to the shared-provider-quota outages that stall Autoblog.
Verified with 38 unit tests plus live sweeps of crawlproof.com (24,
Clean), ugig.net (29) and news.ycombinator.com (58, Sloppy). Found a real
duplicate-meta-description bug on three of our pages and unsized images
across 38.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
vu1nz Security Review0 finding(s) in PR #? No security issues found. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Adds Slop Score — a free, deterministic scan engine that sweeps up to 50 same-origin pages and scores how careless a site looks. 0 is pristine, 100 is maximum slop.
Output is the headline score, a per-page fix list, and systemic rollups — when 38 pages share a defect it leads with "this lives in your template, fix it once" instead of 38 separate to-dos.
What it checks
{{template}}/undefined/[object Object], dev-staging hosts,console.*and TODO leftovers, duplicate + missing titles/descriptions, dead links, deprecated tags!importantsprawl read from fetched stylesheetsCrawl discovery is
sitemap.xmlfirst (following sitemap indexes), then breadth-first from the entry page, bounded by a 3-minute deadline so it stays inside the worker's stuck-audit cutoff.Two design rules that are load-bearing
1. It reports observable defects, never "this was written by AI." An AI-probability score is unfalsifiable, the classifiers are unreliable (they systematically flag non-native English writers), and it would accuse paying customers. Every finding is something an owner can verify in ten seconds and fix.
2. Ambiguous markers only count in unambiguous positions. Dogfooding turned up four false positives, each now covered by a regression test:
[insert …]onlydevdrafts.netlify.app,aetherfall-ten.vercel.app/recent; outbound HN linkPalette sprawl also now ignores custom-property definitions, so having design tokens (a Tailwind theme ramp) is no longer penalised as "107 distinct colors".
Wiring
cost: 0(lib/credits.ts), worker dispatch, engines panel, report view with aSlopMeterhero stataudits_engine_checkmigration — without it everyslopinsert violates the constraint, the same trap the fugu and zai migrations document. Needs applying to prod before slop scans will insert.rule). Both are free, self-hosted and page-budgeted; the per-IP anonymous daily limit still appliesVerification
38 new unit tests, full suite green (472 passing), clean typecheck, successful production build, plus live sweeps:
<center>, no viewport,/≈/newsat 88%)It found a real duplicate-meta-description bug on three of our own pages and unsized images across 38.
Branched from
masterrather than the current working branch so the diff is self-contained; the DeepSeek V3→V4 label change stays with its own PR.🤖 Generated with Claude Code