1. Exact variant
We record model number, US electrical region, published nominal diameter, editorial machine family, product variant, offer availability, reviewer, and check date. A family label is never treated as proof of mounting geometry.
Methodology
A checkmark must explain itself.
Caffeware uses exact US variants, claim-level evidence types, and conservative conflict handling. A shared diameter is never promoted to verified fit by itself.
Methodology details
2. Evidence before inference
Exact manufacturer statements come first. Ecosystem statements and attribute matches remain Probable or Unknown when the exact model is not named.
3. Conservative conflict
Missing dimensions become Unknown. Conflicting evidence keeps the safer result and is marked for editorial review.
Status definitions
- Verified compatible
- An exact manufacturer statement or exact model-family record supports an attaching fit.
- Machine-independent tool
- A separate workflow classification: the item does not attach to the machine. It does not claim exact physical fit or counter clearance.
- Adapter / limitations
- The fit or procedure is conditional, and every condition is shown.
- Probably compatible
- Strong ecosystem or attribute evidence exists, but exact confirmation is pending.
- Not compatible
- A known required interface, dimension, or procedure conflicts.
- Unknown
- The current source does not name the exact model or a critical measurement is missing.
Offers
Prices are snapshots for a named variant. Availability is shown only when explicitly observed; unknown availability is never emitted as InStock structured data.
Troubleshooting
Advice cites manuals or official support records, identifies the relevant manual sections, distinguishes general patterns from model-specific checks, and stops before electrical, thermal, pressure, or disassembly risk.
Product illustrations
Every catalog visual is an original AI-assisted Caffeware editorial illustration, clearly labeled as an illustration rather than a product photograph. These visuals communicate product category only; dimensions, fit, finishes, and compatibility always come from the cited evidence records.
Setup Builder scoring is deterministic
For each machine, the builder adds one experience score, one drink score, and one space score from the table below. It removes machines without a currently in-stock verified offer, then keeps options within budget and selects the highest total; price breaks ties. This is a curated ruleset, not predictive AI or statistical personalization.
- Bambino
- Experience B/I/A: 8/6/5 · drinks milk/mixed/straight: 6/7/6 · space compact/standard/open: 9/6/5
- Bambino Plus
- Experience B/I/A: 9/6/3 · drinks milk/mixed/straight: 10/8/4 · space compact/standard/open: 7/6/5
- Barista Express
- Experience B/I/A: 7/6/4 · drinks milk/mixed/straight: 6/7/5 · space compact/standard/open: 2/6/5
- Dedica Arte
- Experience B/I/A: 6/4/2 · drinks milk/mixed/straight: 5/5/3 · space compact/standard/open: 9/5/5. Currently excluded automatically because no machine-verifiable offer is installed.
- Classic Pro E24
- Experience B/I/A: 1/8/10 · drinks milk/mixed/straight: 3/5/10 · space compact/standard/open: 3/6/5
B/I/A means beginner/intermediate/advanced. Existing equipment is retained and excluded from price. Advanced or straight-espresso paths prefer Opus 2 only while its exact variant is in stock; otherwise the first in-stock espresso-capable seeded grinder is used. Accessories are added only when the relationship is Verified, Machine-independent, Adapter, or Limitations, the exact offer is in stock, and the total remains within budget.
Independent autonomous review
A researcher agent may propose a claim but cannot approve it. Every source bundle needs two current attestations from different registered principals, independence groups, and executions. Every compatibility decision and every troubleshooting flow receive their own two-agent quorum after the underlying sources pass. Each attestation is bound to the candidate content hash and to downloaded evidence receipts containing requested and resolved URLs, HTTP status, capture time, response hash, and normalized evidence hash.
This is automated multi-agent review, not human barista or technician review. Any disagreement, missing evidence, stale receipt, reused identity, unsafe instruction, or failed test blocks the authority claim. Humans are reserved for unresolved legal, image-rights, or severe safety conflicts.
Continuous verification
The daily Git-backed verifier is intentionally narrow: it refreshes only exact, unambiguous USD price and availability fields. A separate scheduled Codex editorial cycle audits or expands content with independent agents, evidence capture, quorum, and fail-closed release gates. Every accepted change must pass catalog validation, authority audit, types, lint, unit tests, a production build, browser tests, accessibility checks, sitemap checks, candidate health, public release identity, and rollback protection.