docs: query-spec honesty-fix for live search field set (#168 Direction B)

[rdhyee]

Summary

Direction B of #168, decided mechanically by the locked latency thresholds against measured #167 baseline data. Doc-only PR — no code changes.

The current Interactive Explorer searches `label` + `place_name` against `samples_map_lite.parquet`. `query-spec.qmd:225` previously claimed `label` + `description` + `place_name` — a known mismatch. This PR corrects the spec to match the live behavior and forward-points to #169 / SEARCH_INDEX_V1.md as the substrate work that lifts the gap.

What the data said

I implemented Direction A (swap to `sample_facets_v2.parquet`) on a side branch, ran the perf-smoke against three forms (baseline, naive LEFT JOIN, CTE-then-keyed-join), and posted the comparison on #168 (comment). Headlines:

metric	baseline	A (CTE)	threshold (#168)	passes?
cold `pottery`	8.7 s	12.0 s	≤ 5 s	NO
cold multi-term	5.1 s	14.8 s	≤ 6 s	NO
cold composed-source	5.0 s	5.7 s	≤ 8 s	YES

Recall improvement is real (`pottery Cyprus` flips 0 → 50 results, `Çatalhöyük` flips 0 → 50, `100%` flips 0 → 50), but latency cost exceeds the locked threshold for the bare-text and multi-term cases. Native DuckDB benchmark showed CTE-then-join is 8× faster than naive LEFT JOIN; in-browser, cold-cache HTTP range fetches dominate cost, so the optimization evaporates. Direction A is structurally too slow on the current parquets.

Per the #168 decision rule (latency thresholds drive direction): land Direction B.

What this PR changes

• `query-spec.qmd:225`: corrects the live-Explorer field-set claim from `label + description + place_name` to `label + place_name` only against `samples_map_lite.parquet`.
• Adds a forward pointer to Explorer FTS Track 2: search_index_v1 contract doc #169 / SEARCH_INDEX_V1.md as the path that lifts both the recall gap (description + concept labels in v1 minimum) and the latency gap (substrate-backed FTS).

What this PR explicitly does not do

• Does not swap the search backend. The recall gap remains until the substrate work in Explorer FTS Track 2: search_index_v1 contract doc #169-Explorer FTS Track 5: GO/NO-GO decision gate #172 lands.
• Does not add UI hint. A "search currently covers labels + place names only" inline UI hint near the search box was considered (#168 B+ option). Reasonable follow-up; not included here to keep this PR doc-only.

Test plan

• Read `query-spec.qmd:225` after merge; verify the wording is honest about current behavior and points forward correctly.

Closes #168. Refs #165, #167, #169, PR #173.

🤖 Generated with Claude Code

[rdhyee]

Reviewed the query-spec honesty fix. The wording change itself looks correct and matches the live label + place_name behavior.

One scope note: query-spec.qmd still says substrates that cannot index the full field set MUST surface the limitation in UI, while this PR explicitly remains doc-only and does not add the inline UI hint. I am treating that as a follow-up rather than a blocker for this doc honesty fix.

[rdhyee]

Review result: the query-spec honesty fix itself looks correct and matches the live label + place_name behavior.

Scope note: query-spec.qmd still says substrates that cannot index the full field set MUST surface the limitation in UI, while this PR explicitly remains doc-only and does not add the inline UI hint. I am treating that as a follow-up rather than a blocker for this doc honesty fix.

[rdhyee]

Added UI hint (commit `006d335`)

Codex review correctly flagged that `query-spec.qmd` §3.2's MUST clause ("surface the limitation in UI") wasn't closed by the doc-only fix. Adding the inline hint here so the spec normative requirement lands in the same PR.

`explorer.qmd` changes (+5 lines, inline style only — no styles.css touch):

• New `.search-help` line under the search bar: "Searches sample labels and place names only — descriptions are not yet indexed."
• Link to Explorer FTS Track 2: search_index_v1 contract doc #169 (substrate FTS tracking) so users see the limitation is tracked, not abandoned.
• Replaces the placeholder example `pottery Cyprus` (which returns 0 results per the Explorer FTS Track 1a: Browser perf-smoke baseline #167 baseline — the search bar was suggesting a broken query!) with `basalt California` which actually matches.

Diff: +5/-1.