Consistency checks, with real output: missing attributes and title/description contradictions
Part 3 of 6 in CatalogSentry's launch series.
Part 2 listed what CatalogSentry checks today.
This post shows two of those checks running, with real findings — not paraphrased, not
invented. Every evidence string below was produced by running CatalogSentry's actual check
code against its own demo fixture (actor_input/demo--input.json, shipped in the repo), with
this call:
from utils.feed_loader import products_from_inline from utils.checks import CHECKS import json data = json.load(open("actor_input/demo--input.json")) products = products_from_inline(data["products"]) findings = CHECKS["missing_attributes"](products, required_attributes=["brand", "category", "price"]) If you have the repo, you can run this yourself and get the identical output.
missing_attributes: does every product have what it needs?
This check takes a list of required attribute keys — by default brand, category, and
price, or whatever you configure in requiredAttributes — and flags any product where that
field is missing or empty (including a string that's just whitespace). It checks both
top-level fields (brand, category, price, and others) and arbitrary custom attributes.
The demo fixture includes a "Cotton Towel Set" listing with no brand field set. Running the
check against it produces:
{ "type": "missing_attribute", "severity": "medium", "sku": "TOWEL-SET", "entity": "Cotton Towel Set", "evidence": "Attribute 'brand' is missing or empty.", "explanation": "'brand' is required but has no value for this product.", "recommendation": "Populate 'brand' from the source system or remove the product from the feed.", "confidence": "high" } No AI judgment call here — it's a direct check of whether a required field is present and
non-blank, for every product, every run. That's deliberate: for something this binary,
deterministic is strictly better than a model guessing whether a field "seems" populated.
title_description_contradiction: do these two fields describe the same product?
This is the check that catches what no format validator can: two fields, both well-formed,
both present, asserting different facts about the same product. Today it covers two
extraction classes — storage/memory capacity (e.g. "256GB," "2TB") and pack/quantity counts
(e.g. "3-pack," "12 ct") — pulled independently from the title and the description, then
compared.
The demo fixture's phone listing says "Samsung Galaxy S24 256GB" in the title, and "with
128GB storage" in the description. Running the check:
{ "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" } (The evidence is in MB because that's the common unit the check normalizes both sides to
before comparing — 262144 MB is 256GB, 131072 MB is 128GB. Same contradiction, just shown in
a consistent unit so a 1GB-vs-1024MB phrasing difference never produces a false positive.)
This is exactly the example from part 1 —
a buyer reading the title expects 256GB, the description promises 128GB, and a format
validator sees two valid strings and moves on.
What this check is not
Worth being precise here, because it's easy to overstate: this is explicit value extraction
and numeric comparison, not semantic understanding of the whole product description. It
catches a specific, high-value class of contradiction (capacity and pack-count mismatches) —
it does not currently catch every kind of factual disagreement a title and description could
have. That's a real limitation, not a hidden one; extending the extraction classes this check
covers is on the roadmap, not shipped today.
Next: Part 4 — duplicate detection and variant consistency
Want to run these checks yourself before the next post? Try
the interactive demo, or run it directly on
Apify Store — it's live today.