How Cophi decides what to recommend.
Cophi's recommendations come from a published, deterministic rule set — structured filters and fixed scoring. There is no machine learning here, and the system does not learn from you between sessions.
The seven steps
- 01
Quick Finder collects structured preferences
Brewing method, flavour families, roast level, caffeine and an AUD budget are captured as discrete values — not free text — so every later step works on the same shape of data.
- 02
Ask Φ translates a sentence into the same structure
Natural-language input is parsed into that identical preference object. Ask Φ is a different way in, not a different engine, so the two paths cannot disagree.
- 03
Hard eligibility filters are applied
Coffees that fail a stated requirement are removed outright: over budget, wrong caffeine type, unsuitable for the brewing method, or not currently available.
- 04
Eligible coffees are scored deterministically
Each remaining coffee earns points for matching the stated signals. The rules are fixed and published — the same preferences always produce the same score.
- 05
Results are ranked
Coffees are ordered by score, with a deterministic tie-break so a repeated search never shuffles.
- 06
Match reasons are generated
Each result carries short, plain-language reasons drawn from the record's own attributes — the brewing method it suits, the flavour family it shares, the price against your ceiling.
- 07
The percentage is compatibility, not quality
A match percentage measures how well a coffee fits the preferences you entered in this session. It is never a rating, a review score or an expert verdict.
Signals we use
- Brewing method
- Whether the coffee is listed as suitable for your method — espresso, V60, AeroPress, plunger, batch brew, stovetop or cold brew.
- Flavour families
- Fruity, citrus, berry, floral, chocolate, caramel, nutty or balanced, taken from the record's own profile.
- Roast level
- Light through medium-dark; an exact match scores highest, an adjacent level scores partially.
- Process
- Washed, natural, honey or experimental, used to reinforce flavour expectations.
- Origin
- Country and region, used for relatedness rather than as a hard filter unless you ask for it.
- Budget
- A hard ceiling in AUD per bag. Nothing above your maximum is ever shown as a match.
- Caffeine
- Regular or decaf, treated as a hard filter.
- Availability
- Sold-out records are excluded from recommendations.
- Shipping
- Where a roaster's shipping regions are recorded, they are used to keep results deliverable.
Missing data
When a record does not state something, Cophi neither rewards nor punishes it for the gap: the signal simply contributes nothing to the score. A missing value is never treated as a match, and it is never invented to fill the sentence in a match reason.
Deterministic tie-breaking
Where two coffees score the same, the order is decided by fixed rules — the stronger brewing-method fit first, then the lower price, then the name in alphabetical order. No randomisation, no rotation and no paid placement enters the ranking at any point.
Current limitations
- The catalogue behind these recommendations is fictional demo data, so a match today demonstrates the method rather than a coffee you can buy.
- Preferences are held for the current session only. Nothing is stored and nothing is personalised across visits.
- Scoring reflects stated attributes, which cannot capture everything about how a coffee tastes in your kitchen with your water and your grinder.
- Compatibility is not quality. A 92% match is a close fit to your inputs, not a score out of one hundred for the coffee.
Getting better results
- Name your brewing method precisely — espresso and V60 pull different coffees.
- Pick two flavour families rather than five; a narrow brief scores more sharply.
- Set a realistic budget ceiling instead of leaving it open.
- Say if you need decaf, so it is applied as a filter rather than a preference.
- Read the match reasons — they tell you which input drove the result.
Try it now with the Quick Finder or Ask Φ. You can also browse the coffee catalogue directly, or read more about Cophi.