# Fenris TUI information-architecture PROTOTYPE (throwaway)

**This is throwaway code answering [ticket #3](https://git.bongbetic.com/xavierk/Fenris/issues/3).** It is not the redesign, reads nothing real, and never ships. Branch: `prototype/tui-information-architecture`.

## Question

What screen hierarchy, navigation, and action model makes Fenris's projection contract (ADR 0002 §13), the four separate service facts (ADR 0003 §8), warming-up, unexplained gaps, and changing habits understandable in a keyboard-first terminal?

## Run (one command)

```sh
./run
```

(creates `.venv` and installs `textual` on first use)

## What to flip through

**Variants (← / →)** — three structurally different answers, not restylings:

| Key | Variant | Idea |
|-----|---------|------|
| A | **Panes** | everything on one dense screen, btop-style; no navigation, panes are zones |
| B | **Pages** | persistent three-fact header (lifespan · confidence · freshness) + pages 1–5 |
| C | **Ledger** | one scrolling document in reading order, headline sentence first |

**States (s)** — same variants, six shapes of the contract:

1. steady · Supported (with one unexplained 3-hour gap)
2. warming up · Limited (11 of 14 days)
3. habit changed · Limited (regime 6 days old, scenario spread visible)
4. stale · Supported→Limited (last collect FAILED, 61 h old)
5. no baseline · Unavailable (wear too coarse to imply endurance)
6. paused · Limited (period closed by deliberate disable)

**Actions** — `p` pause (asks confirmation) · `r` resume (doesn't) · `c` collect now (synchronous outcome). Each suspends the TUI and runs `polkit_stub.py` on the real terminal: this validates the tty passthrough ADR 0003 requires for the polkit prompt. Results land in the tty log (variant B service page; every variant's log is the same list).

## What to react to

- Which variant's hierarchy matches how you think about the drive? (Mixing — "header from B, density of A" — is a valid answer and the point.)
- Are the four service facts separable at a glance?
- Do confidence states + contributing facts read as evidence, not as a percentage?
- Is the pause confirmation the right amount of friction?
- Did the polkit tty stub actually prompt in your terminal? (That's the mechanism check.)

`screenshots/` holds headless captures (`smoke_test.py`) of each variant at 80×24 and 140×40.