Rider feedback triage
Report
Present
Load batch
Model prompt
Coding rules
Print report
Fast read
Static demo on GitHub Pages: mock labels, read-only. Run it locally (python3 server.py) to change labels, edit costs, load a batch or use live mode.
Repo
Needs attention today
What's working
Work the queue
← Back
Next →
← → slides · F full screen · Esc back to report
Full screen
Exit
What the model sees
Rider ids are never sent. In mock mode no call is made; this is the exact prompt live mode sends.
Step 1: label each message
Step 2: what waiting costs
Coding rules
Each message gets one primary code (the issue needing action soonest), up to three secondary codes and any modifiers. Messages are grouped by the primary code.
Modifiers
Context, not issues. Solid border: chosen by the model (or a human). Dotted border: set by code from the batch.
Event codes
Built like a procedure code (ICD-10-PCS): 7 positions, one character each, every position one axis, together describing exactly what happened. Each category code gets one, documented by the same quote. Z means not stated or not applicable: the tool never guesses. Positions 2 to 7 never use I or O. Code sets position 5 from the message time and the single, repeat and corroborated qualifiers from the batch.
Validity tables (which values each position allows, given the positions before it)
Load a new batch
Same columns as the export: id, created_at, channel, route, rider_id, star_rating, message.
Triage this batch
Back to the original batch