What CatalogSentry actually checks today
Part 2 of 6 in CatalogSentry's launch series.
Part 1 made the case for why this matters.
This post is the honest inventory: what CatalogSentry is, what it checks, and — just as
importantly — what it doesn't do yet, so nothing here gets oversold.
What it is
CatalogSentry is an Apify Actor that scans a product feed — CSV, JSON, or Google Merchant
XML — and finds the issues that field-format validators miss. It runs on a feed URL for a
scheduled, recurring audit, or against a small inline product list for quick testing. Every
finding is evidence-backed: it names the exact SKU, quotes the conflicting values, and
recommends a fix. Never a vague "this looks off."
The four checks this series walks through
These four are fully deterministic — no inference calls, no LLM judgment calls, no per-token
cost. Just explicit logic comparing fields against each other and against your requirements.
They're also CatalogSentry's original four, so they're the clearest introduction to how a
finding gets built — parts 3 and 4 walk through real output from each.
-
missing_attributes— flags products missing a required attribute (brand, category, price, or any attribute key you configure) with no value, or only whitespace. -
title_description_contradiction— extracts factual claims from title and description (storage/memory capacity like "256GB," and pack/quantity counts like "3-pack") and flags when the two fields assert different values for the same product. -
duplicate_detection— flags exact duplicates (two SKUs sharing one GTIN) and near-duplicates (two SKUs with a matching normalized title + brand). -
variant_consistency— for products sharing anitem_group_id, flags variants that claim different brands within the same group, and flags "variants" whose color/size/ material/style attributes are actually identical to a sibling's — meaning they aren't distinguishable variants at all.
We'll walk through real output from each of these, run against CatalogSentry's own demo
fixture, in parts 3 and 4 of this series.
CatalogSentry has grown since this series was first drafted: there are now 8 deterministic
checks on by default (these four, plus image hygiene, GTIN/EAN checksum validation, thin-content
detection, and catalog-internal price-outlier detection) and 7 opt-in checks using this Actor's
own built-in AI-assisted detection — see the README for the current, complete list.
The opt-in checks
Seven checks are off by default. Six of them — semantic_duplicate_detection,
image_description_alignment, seo_keyword_gap, visual_brand_consistency,
category_voice_consistency, and translation_consistency — use this Actor's own built-in
AI-assisted detection; no API key or endpoint to configure, the AI backend is operator-hosted,
so enabling the check in checks is the only step. The seventh, semantic_cannibalization,
is deterministic by default (a word-overlap pass) with an optional AI-assisted embedding pass
layered on top. One of the AI-assisted ones, semantic_duplicate_detection, flags likely
duplicate listings by semantic similarity rather than exact GTIN or title match — useful for
catching duplicates described in very different words. We'll cover the rest of the opt-in
checks, with real output, in parts 5 and 6.
Input and output, briefly
Point feedUrl at your feed (or pass a small products array inline), pick which checks to
run (default: the 8 deterministic ones), optionally set requiredAttributes to match your
own taxonomy's mandatory fields, and run the Actor. Findings land as dataset rows you can
export as JSON, CSV, HTML, or Excel; a summary lands in the key-value store. Each finding
looks like this:
{
"type": "title_description_contradiction",
"severity": "high",
"sku": "SGS24-256",
"entity": "Samsung Galaxy S24 256GB",
"evidence": "title mentions capacity [262144.0] MB; description mentions capacity [131072.0] MB.",
"explanation": "Storage/memory capacity differs between fields describing the same product.",
"recommendation": "Verify the authoritative capacity and correct the inconsistent field.",
"confidence": "high"
}
That's a real finding — not a mockup. It's the actual output of running
title_description_contradiction against CatalogSentry's own actor_input/demo--input.json
fixture. We'll show the rest of that fixture's findings, and explain how each check arrived
at them, next.
What this is not (yet)
To be direct about where the line is today: CatalogSentry does not generate or rewrite content,
check multilingual coverage, or compare your prices against an external market signal (only
against the rest of your own catalog). Those are real ideas on our roadmap — but roadmap ideas
aren't shipped checks, and we're not going to describe them as if they already ran in your feed.
(Image hygiene and SEO keyword analysis, which this post originally listed as roadmap items,
have since shipped as image_technical_qa and seo_keyword_gap — see the README for the
current list.)
Update: this has since been run end-to-end against a real, unmodified 995-product public catalog
(Flipkart's e-commerce sample feed) with every check enabled — 8383 findings, no crashes, real
images and real prices, not the 5-product demo fixture. 8 checks are deterministic by default;
7 more are opt-in and clearly gated.
Next: Part 3 — consistency checks, with real output
Want to see this running before you read the next two posts? Try
the interactive demo, or run it directly on
Apify Store — it's live today.
Top comments (0)