# Paddling Below

> A service blueprint for one everyday service you went through - collecting a parcel, a haircut, a
> repair, a renewal at a desk. List three to twelve steps in order and put each in one of four lanes:
> you, in front of you, out of sight, and what it runs on. Each step carries its minutes and marks for
> waiting, going wrong and - below the waterline - your guesses. A free in-browser engine draws it as a
> pond in cross-section with a swan (the waterline is the line of visibility) and works out the total
> minutes, the waiting share, the waterline crossings, the longest stretch out of sight, the snags and a
> verdict: snagged below, snagged in view, long paddle or smooth glide. A metered blueprint reading
> names the longest stretch out of sight, what went wrong, the longest wait, one fix for whoever runs
> the service and one question for the person to ask next time. Derived from @owl-listener/service-blueprint.

https://paddling-below.skillsafe.ai/

## What it does

- The four lanes come from service blueprinting: customer actions, frontstage (in view), backstage (out of sight) and support processes (what it runs on). The waterline between in front of you and out of sight is the line of visibility.
- The verdict, checked in this order: a step out of sight or in what it runs on went wrong is snagged below; else a step of yours or in front of you went wrong is snagged in view; else waiting of half the minutes or more is a long paddle; else a smooth glide.
- The longest stretch out of sight is the run of consecutive steps below the waterline with the most minutes (a tie goes to the earlier one). The fix goes to the first snag, else the longest wait, else where that stretch begins; the question is about the first guessed step.
- Free downloads: the pond drawing (SVG, PNG), the steps (CSV), and everything with the reading (Markdown, JSON).
- Saved blueprints (opt-in) keep the steps and the latest reading, and show how the verdict moved between runs.

## The paid task

One task, `blueprint`, on gpt-terra, metered per run. The input is four string fields: `task`, `service` (name and note), `steps` (one a line with its lane, minutes and marks) and `facts` (the page's numbers). The reply is one JSON object: verdict (snag-below | snag-in-view | long-paddle | smooth-glide, always the page's), headline, out_of_sight, snag_note, wait_note, fix_for_them and ask_next_time. The page re-checks the reply: the verdict and the waiting share must be the page's, each note must name the right step, and nothing may set a day or time the steps did not give.

Guesses about what happened out of sight are labelled as guesses. A thinking aid for everyday services, not professional advice.

- API tutorial: https://paddling-below.skillsafe.ai/api.html
- Source skill: https://skillsafe.ai/skill/@owl-listener/service-blueprint (owl-listener/designer-skills, MIT)