# Between Hours

> A field reading for liminal, in-between places. Describe a real place you pass through - a stairwell, a corridor, a waiting room, a car park, an empty pool - switch on the cues you notice, and get an urban planner's reading of why it does or does not feel in-between, the signs in your own words, the hour it is most in-between and a courtesy note.

URL: https://between-hours.skillsafe.ai/
API guide: https://between-hours.skillsafe.ai/api.html
Built on the SkillSafe skill @rysweet/urban-planner-analyst: https://skillsafe.ai/skill/@rysweet/urban-planner-analyst

## Free in the browser

- A description box (up to about 700 characters are sent; longer text keeps its start and end), a kind of place (stairwell, corridor, waiting room, concourse, car park, mall, school or office after hours, hotel corridor, indoor pool, playground, somewhere else) and an hour (early morning, middle of the day, evening, late at night, off-season or closed).
- Eight switches, drawn as the ceiling lights of a corridor: built for passing through or waiting; built for a crowd that is not there; nobody else around; flat or buzzing artificial light; repeating doors, tiles or rows; no windows or no clock; feels familiar from childhood; a hum, an echo or a hush.
- A word match over the description that outlines the lamps your words suggest - it never switches one on for you.
- A fixed verdict rule, checked in order: just a place unless built for passing through or for a crowd; deeply in-between when also nobody is around and at least one more cue is on; otherwise in-between at the right hour.
- Exports: Markdown, a CSV of the eight cues, JSON.

## The paid task (metered per run, gpt-terra)

One task, `read`. Input fields (all strings): task, place, kind, when, cues, page_facts, optional clip_note and retry_note. The reply is one JSON object: title, cues_counted, verdict (deeply_liminal, at_the_right_hour or just_a_place), verdict_reason, reading, signs (cue and seen), best_hours (when and why), tip_more and visit_note. The page re-checks every reply: all switched-on cues counted, extra cues backed by the words, the verdict rule, one sign per counted cue, word limits, and nothing brought in that the description does not say - no added lights, sounds, objects or weather, no names, nothing spooky, and no suggestion to go anywhere closed or private.

## Saved places

Saving is opt-in. Saved places go to a declared `readings` collection in the person's SkillSafe account (a local copy is kept in the browser), so the loop is: read a place, go back at the suggested hour, flip the switches and read it again. A second reading of the same place shows what changed since the last one (verdict, counted cues, most in-between hour), also in reopened saves; the saved list marks a place whose verdict changed. The reading also notes whether your clock is at the hour it names, and "Copy caption" gives one line of at most 280 characters (title, verdict, reason, hour) to paste under a photo.