The difference
Why not just ask a chatbot?
A chatbot makes up a profile on the spot. Ours is crafted from the published record, reduced to the core elements that matter, the same way for every bottle, any bottle. New release or dusty. So your read gets measured against something that holds still. Notes you’re handed, you forget. A pour you figured out, you keep.
The real problem is confidence
Ask a general chatbot for a bottle’s profile twice. The gist matches; the details shift. It answers from memory, blurs batches and releases together, and sounds exactly as sure whether it knows a great deal or almost nothing. For a bourbon fan who wants to check their own palate against something, that is the wrong kind of answer. You can only measure a read against a thing that stays put.
We use AI too. We put it on a leash.
Socrates is an AI guide. We would rather say how it is controlled than pretend it is not there. Four controls sit between the model and what you see, and each one is enforced in code, not asked for politely in a prompt.
For a named pour, Socrates retrieves what has actually been published: the producer’s own notes for that bottling first, then the independent reviews that agree with them. It does not answer from memory. How the record is built, in plain terms.
Every record is reduced to the same core elements, designed by John McKillop from 400+ documented blind tastings: colour, nose, proof, and the eight palate axes of the Bourbon Brothers™ Method. The same shape for every bottle means one read can be compared with the next.
A pour with a published record is researched once and the record is kept. The next sipper who names that bottle reads the same record you did. Different spellings, acronyms and year formats resolve to one record, not several.
When nothing bottle-specific has been published, Socrates tells you, reads the pour by its type, and speaks in ranges. When the record could be a different release than the one you named, it asks which bottle is in the glass instead of guessing. It never invents a proof, an age or a distillery.
The score is math. No AI touches it.
Once your read is committed and the record is on the table, the comparison is arithmetic over things that can be counted. No language model sits anywhere in the scoring path. Run the same read against the same record tomorrow and you get the same numbers.
The palate axes where your none / some / big call matched the record’s, as a share of the axes you read.
Your rating minus the record’s public score. Signed, so over time it shows whether you run hot or cold against the consensus.
Global Palate Similarity: the two blended into one position, with a tolerance band for the record’s own noise.
The GPS is a position, not a grade. A high number means your palate sits close to the published consensus; a low one means it sits somewhere else, which can be the more interesting palate. Nobody is ranked by it and there is nothing to “improve.” What the Bourbon GPS is, and how to find yours.
Search and answer, together
You always get the full answer. What changes is the order. You make your read first, stage by stage, and only then does the record open. Then you see the distance between the two as coordinates. Nothing is withheld; the answer simply arrives after your own, so it has something to land on. Any bot can describe it. You get to earn it.
The questions are the method, and the method is scrutinised
The guide runs the Bourbon Brothers™ Method, John McKillop’s human-authored tasting method from 400+ documented blind tastings over 15 years with the DGC bourbon club, and the world of the novel Bourbon Brothers. John serves as Whiskey Historian for the Medicinal Whiskey Charity. The method is not a tone of voice the model puts on. It is a set of rules the code checks every turn:
- Ask first. The guide never names a descriptor you have not reported. It confirms each stage against the record only after you commit your read, and the stages ahead stay sealed.
- Eight axes, every session. The order of the palate pairs is set by the server, not the model. A turn that tries to skip an axis or reveal early is caught and rewritten to the Method’s own ask. Every session grades all eight.
- Era is arithmetic. A bottling year minus an age statement is a distillation year, and the dusty houses’ eras are hard dates. That is computed before the search runs, so a 1998 twelve-year is never read as a modern shelf bottle.
- Every reveal names the near-miss. The runner-up the pour could have been, the one tell that separates them, and a rule you can carry to the next glass. A reveal that only describes tonight’s pour is treated as a failure.
Where the guide sits and why it exists is the origin story.
When the record is thin
Single barrels and store picks are the honest hard case. Two barrels from the same warehouse can read differently, and most picks have no published profile at all. Socrates says so. It reads the pour at the level of its mashbill and proof, the reveal grades against what is typical for that type, and the Profile Confidence plate on the session banner tells you how settled the record is before you commit a single read. What the Confidence Score counts.
It lasts
What you figure out, you remember, and the Palate Journal keeps the receipt. Every read is kept against its record, axis by axis, with your alignments over time, your top pours and distilleries, and a printable Tasting Book. See a sample journal.
Notes you’re handed, you forget.A pour you figured out, you keep.
Three sessions free.No app store hassle ever.