docs(audit): correct postcode finding — full postcodes do trigger proximity search

Recheck with 'B91 3DL' via the search box shows the geocoded radius
search works (distances, nearest-first, radius selector, map toggle).
Original P0.3 tested only a partial postcode, which silently falls back
to text search — refiled as P2.0 (silent mode switch on partial input).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

docs/superpowers/specs/2026-07-02-ux-audit-report.md

@@ -13,6 +13,7 @@ Full evidence trail: `docs/superpowers/specs/2026-07-02-ux-audit-notes/` (six no
 These are strengths the fixes below must not regress:
 
 - **Name search is genuinely excellent.** "Welland Primary" → the correct school as sole top result, in 3 taps with instant loads — comfortably under the 15s target. Result cards are information-rich (Ofsted grade+year, RWM % with trend arrow, "+12 pts vs national", pupil count, LA) and adapt correctly to phase (Attainment 8 for secondary, RWM for primary).
+- **Full-postcode proximity search works, and works well** *(verified on 2026-07-02 recheck)* — typing a full postcode (e.g. `B91 3DL`) into the home search box yields "13 schools within 1.0 miles of B91 3DL", distance on every card, nearest-first ordering, a 0.5–5 mile radius selector, and a List/Map toggle; the "Schools near me" geolocation button offers the same without typing. Only partial input degrades (see P2.0).
 - **Search is the unmistakable primary action** — above the fold on both viewports, 70px tap target, plain-English placeholder, content order puts the task first.
 - **The primary (KS2) school page explains its data in plain English** — "End-of-primary-school tests taken by Year 6 pupils", the in-place "Why is combined lower?" explainer, legends on subject charts. This is the model the secondary template should copy.
 - **Everything is anchored to the national average** ("+14 pts", "National avg 39.1") — a non-specialist can judge good/bad without leaving the page.
@@ -43,13 +44,6 @@ These are strengths the fixes below must not regress:
   - Recommendation: restack the mobile hero so map, H1, and compare button occupy non-overlapping space (e.g. compare button below the identity block or docked, map height capped above the H1). One template, both school-detail components (see P2.6).
   - Uplift: **school-page bounce-back-to-Google — decrease, moderate** (fixes the first-glance orientation on every SEO landing; the page content behind it is already strong, so retention should follow).
 
-- **P0.3 — Postcode search returns a distance-less, unsorted, mixed-phase list** *(J1-F2)*
-  - Evidence: `/?search=B91+3` → "8 schools found", sort "Relevance", **no distance on any card**, top result a Secondary school, phases interleaved (secondary/independent/primary/sixth-form) — behaviour consistent with a text match on the postcode column rather than a geographic radius search. Screenshot `j1-postcode-results-mobile.png`. The geolocation "Schools near me" button proves proximity machinery exists but the typed-postcode path isn't wired to it.
-  - Criterion: Nielsen #2 (match between system and the real world) + task-flow obstruction on the highest-traffic path. The placeholder explicitly invites "School name **or postcode**".
-  - Argument: a parent typing their postcode expects nearest-first with distances. What they get looks broken — no way to judge which school is closest, led by the wrong phase. Home is 63% of entries and 46% of exits; this is the largest observed task-level leak feeding that exit rate, on the discovery path used by exactly the parents who *don't* already know a school name.
-  - Recommendation: detect postcode-shaped queries, geocode (postcodes.io is already integrated), sort by distance, and show "X.X miles" on each card — i.e. route the postcode text path through the same logic as "Schools near me". Default the phase filter sensibly or surface it prominently on postcode results.
-  - Uplift: **home 46% exit rate — decrease, moderate-to-large** (repairs the primary discovery task for the know-my-area-not-my-school segment; large band justified because the failure currently reads as "site is broken" rather than "site lacks a feature").
-
 ### P1 — High
 
 - **P1.1 — Filter and Sort selects on home results have no accessible name** *(J1-F3)*
@@ -103,6 +97,13 @@ These are strengths the fixes below must not regress:
 
 ### P2 — Medium
 
+- **P2.0 — Partial or malformed postcode input silently falls back to text search with no distances** *(J1-F2, corrected on 2026-07-02 recheck — the original P0 claim that postcode search lacks proximity was wrong: a full postcode works, see "What works today")*
+  - Evidence: the search submit handler routes input through a strict full-postcode regex (`FilterBar.tsx` `isValidPostcode()`). `B91 3DL` → `/?postcode=…&radius=1`: "13 schools within 1.0 miles", distance on every card, nearest-first, radius selector, map toggle (`j1-fullpostcode-results-mobile.png`). But `B91 3` (an outcode+sector a parent plausibly types) → `/?search=B91+3`: text-matched, relevance-ordered, distance-less, mixed-phase results led by a secondary school (`j1-postcode-results-mobile.png`) — with no notice that a different, better mode exists.
+  - Criterion: Nielsen #1 (visibility of system status — silent mode switch) and #9 (help users recognise and recover); error tolerance for the "or postcode" promise in the placeholder.
+  - Argument: the two result pages look near-identical, so a parent who types "B91" or fat-fingers a character never learns that proximity search exists; they just get an apparently arbitrary list on the highest-traffic path.
+  - Recommendation: detect postcode-*shaped* input that fails full validation (outcode/sector patterns, near-miss typos) and either geocode it anyway (postcodes.io supports outcodes) or show an inline nudge — "Enter a full postcode like B91 3DL to see schools by distance."
+  - Uplift: **home 46% exit rate — decrease, small-to-moderate** (recovers the postcode-search subset who type partial input; the full-postcode path already works, which caps the band).
+
 - **P2.1 — Mobile data tables hide the decision-critical columns behind unaffordanced horizontal scroll** *(merges J3-F1, J4-F3)*
   - Evidence: compare's "Detailed Comparison" table is 626px inside a 324px wrapper — only Year + the first school's column visible, no scroll cue (`j3-compare-two-schools-mobile.png`); rankings shows RANK/SCHOOL/AREA but the metric % (the ranking's justification) sits off-screen (scrollWidth 551 > 324) (`j4-rankings-mobile.png`).
   - Criterion: mobile-usability standard (primary content hidden off-viewport); Nielsen #6 — the user must memorise school A's numbers while scrolling to school B.
@@ -201,8 +202,8 @@ P0.1 contrast token fixes; P1.1 label the two selects; P1.4 chart `aria-label`s
 **Batch 2 — School-template mobile fold (small effort, direct SEO-retention uplift).**
 P0.2 un-overlap the H1/map/compare button; P2.5 scrollable section nav; P3.8 trend-chart legend + badge overlap; P3.3 add-confirmation live region. One template, all SEO landings.
 
-**Batch 3 — Location search done right (moderate effort, biggest home-funnel lever).**
-P0.3 postcode → geocoded, distance-sorted, distance-labelled results (reuse the "Schools near me" machinery); P1.6 autocomplete on the home search (reuse the compare modal's live search); P1.7 mobile value-prop line. Together these target the 46% home exit rate from three directions.
+**Batch 3 — Home search assistance (small-moderate effort, home-funnel lever).**
+P1.6 autocomplete on the home search (reuse the compare modal's live search); P2.0 graceful handling of partial/malformed postcodes (geocode outcodes or nudge toward a full postcode — the full-postcode proximity search already works); P1.7 mobile value-prop line. Together these target the 46% home exit rate from three directions.
 
 **Batch 4 — Discovery and sharing bridges (moderate effort, compare-funnel uplift).**
 P1.2 "Schools nearby" module on school pages (feeds compare); P1.3 URL-over-localStorage precedence for shared links; P2.2 mobile +Compare on rankings rows; P3.1/P3.2 compare state-clarity nits.