Community · E2 · artifact verified
Triage a store inbox into four lanes
A quiet clerk for a small online store's inbox: a decision model reads every customer email first and sorts it into four lanes - needs a person now, can wait, template could answer, sales pitch - packaged as an n8n workflow with a shadow mode.
01 · Role in the system
What Jev does here
Every customer email becomes typed questions answered with calibrated probabilities, and code maps the answers into four lanes: needs-human-now, can-wait, template-answerable, and sales pitch. The deployment ships as an n8n workflow (generated from a build script) so a store owner plugs it into their existing automation, and a shadow mode records what Jev would have done alongside what the human actually did - so the owner can audit the clerk before trusting it. Triage.py is a small readable core; the lanes and thresholds are config, not magic.
02 · Control boundary
Where Jev sits
Jev as the first reader of a store inbox: typed questions per email, calibrated probabilities mapped by config into four action lanes, deployed as an n8n workflow with a shadow mode that logs judge-versus-human agreement.
Code owns the loop, permissions, thresholds, validation, and side effects. Jev owns only the bounded judgments described above.
03 · Known limits
What this evidence does not prove
- Lanes and thresholds are the author's config for a small-store volume.
- 0 stars; shadow-mode agreement numbers are not published yet.
- n8n deployment assumes an existing n8n instance.
04 · Attribution
Public sources
This is a Community record: the project was published by a third-party community author.
- pentasir ↗Community · github · public · checked 2026-10-08