Methodology · corrections · change log

How Cousure verifies what it publishes.

Cousure is a source-backed reference. Every claim ties to an official source URL with a literal quote and a retrieved-at date. This page describes the discipline. The change-log and correction history appear here as the index grows.

The pipeline

snapshot → extract → validate → review → publish

1 · Snapshot

We fetch the brand's official page and save the raw HTML plus cleaned text to disk, with the fetch timestamp and hash. If the brand later changes the page, we still know what our claim was based on.

2 · Draft, never publish

A model reads the saved snapshot and proposes a structured record with a verbatim quote per field. The model output is a draft. It never enters the live site directly.

3 · Validate + review

A validator confirms every non-null field has a sourceQuote that appears verbatim in the snapshot. A human reviewer then approves, edits, or flags the draft. Only reviewed records become public.

What we never do

Invent a fact

If the brand's page doesn't say where a bag is made, we say "not disclosed" — not a guess. Even when we know.

Sort by affiliate

The order brands appear in our index and comparisons does not depend on whether we earn a commission from them.

Use unlicensed photos

We use abstract leather silhouettes in v1. As brand cooperation grows, we'll display licensed images or our own photography only.

Data depth model

How deep a brand profile can go

Tier 1 · official sources

Homepage, about/story, product pages, material pages, repair policy, return policy, shipping, care guide, FAQ, size guide. This is the core trust layer — every public claim must trace here first.

Tier 2 · price feeds (later)

Current price, stock status, product URLs from affiliate or licensed feeds. Useful for freshness. Never used for editorial ranking. Not yet active.

Tier 3 · original testing (future)

Real weight, what fits, hardware feel, lining, strap comfort, 30/90-day wear. This is how Cousure becomes hard to copy. Requires physical access to the product.

Internal depth labels

Stub — brand listed, no sources reviewed. Basic — at least one source fetched. Reviewed — at least one product source reviewed and imported. Deep — homepage + multiple products + policy pages reviewed. Tested — original Cousure physical testing data exists.

Source slots

For each brand we track required source slots: homepage, about/story, craftsmanship, care, repair, shipping, returns, FAQ, size guide, and each product page. A slot that's been fetched but not reviewed shows as a gap on the brand profile.

Missing disclosure is a feature

When a brand's page does not disclose where a bag is made, the weight, the hardware material, or the lining — we say so explicitly. “Not disclosed” is evidence. It is never replaced with a guess.

Source-slot kinds

Every source slot has an explicit kind that controls how it can be imported. brand_home and brand_about contribute brand-level facts. product_detail is the only kind that can become a product record. catalog_list and policy_page are evidence-only.

Catalog vs product pages

A catalog or category page shows many bags in a grid with prices. We use these for coverage (what families a brand publishes, what materials, what price range) and price discovery. A catalog page is never imported as a single product record. Product-level claims like dimensions, leather origin, made-in, and weight come only from individual product pages.

Insufficient evidence

When a reviewer opens a snapshot and finds no new factual content beyond what's already covered by another source, the slot is marked insufficient evidence rather than approved. The slot stays in the registry as “reviewed” — we visited the page, the page had nothing unique to add.

How Leather Guide sources are verified

Same pipeline as product claims

Education source → snapshot.

Cousure registers an education source (Leather Naturally, Leather Working Group, IULTCS, tannery technical references) and fetches its public page. The raw HTML and cleaned text are saved with a SHA-256 hash and retrieval date.

Extract → validate quote.

A claim is drafted with a verbatim quote from the snapshot. The validator confirms that quote appears in the cleaned text exactly. Forbidden marketing phrases (“always better”, “Italian leather means made in Italy”, “LWG means sustainable”) are flagged.

Human review → verified.

A reviewer reads the snapshot, decides whether the quote actually supports the claim, and either approves, edits, or rejects. Only approved claims with passing quote-validation are imported as “verified”. The rest stay visibly as “placeholder” on Leather Guide pages until a real source supports them.

How education connects to comparison

Leather Guide →

Vocabulary first.

The Leather Guide teaches the words you encounter on a brand's product page: full-grain, top-grain, Italian leather, LWG certified, handcrafted in Europe. Each phrase is paired with what it proves and what it does not.

Sourced facts second.

Product pages show only what the brand actually published, with the verbatim quote. The Source Ledger is the contract: every cell has a source, a confidence, a review status. Unknown fields stay visible as unknown.

Comparison last.

When you compare two bags, you are comparing source-backed cells using the vocabulary the Leather Guide taught you. Editorial assessments, ratings, and tier labels stay out of the table — only what the brand said, with the words decoded.

How Cousure keeps claims current

Open the change log →

Sources are re-checked.

Every source URL behind a public claim is periodically re-fetched. The cleaned-text hash and each stored quote are compared to the new content. Drift is flagged for review, never auto-deleted.

Prices append; they never overwrite.

When a product page is re-fetched, the current price is compared to the latest price-history row. If it changed, a new row is appended with the new check date. The old row stays — history compounds.

Corrections are public.

When Cousure makes a data-model fix, finds a bad URL, or flags a quote as semantically suspicious, the change is logged on /methodology/corrections with what changed, what surface was affected, and why.

If you represent a brand

Send a source URL that contradicts a Cousure claim, with the verbatim text and the date it was published. We review within 7 days and log every accepted change.

If you're a shopper

Same form, same response time. The most useful corrections are made-in disclosures, dimensions, and price changes you can cite to an official page.

Change log

Every accepted correction appears in the change log with the date, the field, the old value, the new value, and the source. Coming soon.