# Chapter 3 · Tara web search and result navigation · brief · 11 September 2026

The third chapter of the _Tara · find a practice_ track, rewritten in the Eve
edition on 11 September 2026. It follows the plan in `EVE-V1-REDESIGN-TODOS.md`
and the track brief in
[tara-find-a-practice-track.md](tara-find-a-practice-track.md). The receipt is
under `verification/`.

## The job

This is the member's other way in. Chapter two built a catalog from a
recommendation set; this screen does not use it. The Tara search component holds
its own inventory of twenty-eight example records inline, matches a query
against them locally, groups what it finds into four fixed sections, and
remembers the queries a member acted on. No search request leaves the page.

The failure it teaches against is reading a result row as the front of a journey
the screen does not own any of.

**Running example.** The query `calm`: seven local results, a synonym indicator,
a row that matched without containing the word, and a click that records history
and opens nothing.

## Three learning promises (the cover)

1. **Everything here is local** — twenty-eight inline records, a local timer, no
   request to any service.
2. **Order is two decisions, not one** — a global score sorts, then fixed type
   sections decide what is read first.
3. **A click is not a launch** — history is written before the route decides,
   and meditations have no route at all.

## Three things to remember (the close)

1. **A spinner is a timer**, and an empty result is a finished search.
2. **The reading order is not the ranking.**
3. **A click is not a launch**; most rows have no destination.

**Bridge.** Discovery, saved items and launch — chapter four.

## Sections

Four sections of four, opened by a `section-divider` that names all four.

| Section                            | Slides                                                        |
| ---------------------------------- | ------------------------------------------------------------- |
| What this screen actually searches | the capture, the inventory, the timer, the three states       |
| How a match is made                | expansion, the weighted score, the two orders, the highlight  |
| What the screen remembers          | spelling, the four events, the exact-string list, durability  |
| Where a result goes                | the route map, the selection boundary, interaction, the close |

## Treatment decisions

- **The two product views became real captures with numbered pins.** Both
  already cited pinned images from the search harness; they now carry a legend
  and a reading order instead of two untethered notes.
- **The empty-result comparison became a state machine.** Its three rows were
  waiting, results and no results — which is not a comparison, it is a machine,
  and drawing it as one also makes visible that there is no error state, because
  there is no request that could fail.
- **The inline inventory became a stat panel, not a graph.** The plan asked for
  a containment view, and the first attempt was one: four kinds hanging off the
  array. It was four rows tall and ran fifty pixels past the page. The counts
  are the slide's whole content — twelve, six, six and four — so four exact
  numbers with their units say it better and fit.
- **Two comparisons became compare panels.** The grouping slide reads one
  executed query across the global sort and the rendered order; the durability
  slide reads three storage cases across what the screen shows and what the key
  holds. Both have a verdict that is the slide's real claim.
- **The history events map became a graph diagram with the list in focus**, and
  the fourth event — typing — is deliberately absent from it, which is the
  point.
- **Two gated flows became decision trees**, keeping two outcomes each: the
  debounce has settled and waiting, and the selection boundary has a pushed
  route and a stay. Neither has a third branch in the source.

## What the chapter names

Nothing here is new to the code. What the rewrite makes explicit:

- Expansion runs one pass over the words the member typed, so `stress` adds
  `calm` and stops, while typing `calm` reaches both groups.
- Popularity boosts are gated behind a non-zero text match, so a highly rated
  record with no textual match scores zero rather than a small number.
- The sorted ranking is discarded at render time in favour of a fixed type
  order, so the page cannot be read downwards as a relevance list.
- Highlighting receives the original words, not the expanded ones, so a row can
  be present for a reason the page never shows.
- History is written before the route is consulted, and the route returns
  nothing for meditations — which are the largest kind in the inventory and the
  first section on the page.
