Worked example · Taxon 2.2 · sample file

Amplitude export: 52 raw names → 15 clusters

Sample file (not a client export): the bundled Taxon Amplitude demo CSV. 52 event-name strings (50 after whitespace and casing cleanup) fall into 15 clusters. Taxon reads Mixpanel, Heap, PostHog, GA4, Segment, and warehouse exports the same way. Below: one Accept, one Keep separate, the pack, the next dump, and a quick look at Values.

Open Taxon → Try this sample →

The mess

Same actions under different spellings. A few from the file:

CartAdd to Cart · AddToCart · add_to_cart · add to cart
ButtonButton Clicked · button clicked · btn_clicked
CTActa_clicked · Call To Action Clicked · call_to_action_clicked

Property keys drift too: button_id / Button ID, error_code / errorCode, user_plan / User Plan, sku / SKU.

Cluster → Accept

Four “add to cart” spellings are one action. Accept one canonical, Cart Item Added, and the rest become aliases. Why? shows the shared tokens and properties behind the match.

Taxon accepting one canonical name, Cart Item Added, for four Add to Cart variants — Accept · four cart spellings → `Cart Item Added`.

Cluster → Keep separate

Not every near-match should merge. Button, CTA, and btn variants can be different product intents, so keep them apart when the meaning differs. Keep separate stays sticky when you re-normalize. The same Accept / Keep separate pass works on the Properties tab for near-duplicate keys.

Taxon keep-separate on Button Clicked variants — Keep separate · sticky when you re-normalize.

Pack

Preview and edit the pack in-tab: notes under each canonical, plus template chips. Then export taxon-pack.md for the tracking-plan doc, taxon-pack.json, and taxon-map.json, the reusable map for next dump. A CSV map is optional for spreadsheets. Packs can carry value decisions when Values is used.

Next dump

Open the following month’s file against the pack: new · on pack · drifted, with sticky counts that jump to the first new or first drifted name. A drifted name is an old spelling coming back, or a known name with a new suggested canonical.

Values

Names are half the taxonomy. On the Values tab, Taxon picks eligible keys and skips IDs, emails, URLs, timestamps, and high-cardinality keys, each with a reason. Near-duplicate values cluster per key (US · U.S. · us). Accept or keep separate, then export taxon-values.csv or copy a dbt case snippet for accepted clusters. On the next dump, the banner shows new values · on last pack · drifted · stale keys.

The Values walkthrough uses the bundled Values sample: Values sample →

Taxon Values tab on the bundled sample: keys rail with 4 eligible and 9 skipped (user_id, device_id, timestamp, email skipped with reasons), and the country cluster where US, USA, United States, us, U.S. and usa resolve to US with Accept canonical and Keep separate — Values · SAMPLE · `4 eligible · 9 skipped`, `country` → `US`, with Export values CSV and dbt case.

Files stay in the tab

This pass ran client-side on a sample CSV, and nothing was uploaded for cleaning. Maps, packs, value decisions, and next-dump history live in this browser until you export them.

Rename with an effective date How do you rename an event without rewriting history? Pack aliases + dated note + next-dump drift.
Required fields and versioned quarantine How do you require fields without breaking replays? Contract, quarantine, Values drift.
FAQ: questions these teardowns answer Basic pass · rename with a date · required fields / quarantine.

Try this sample

Open the same Amplitude demo CSV (52 → 15), or drop your own export. Questions: hello@coldindex.net.

Try this sample → Values sample → Open Taxon Product page