validateTownCache now warns when medianShopSpacing(cache) > MAX_MEDIAN_SPACING_M (150 m). Metric +
threshold exported from the barrel so no lane re-derives them. WARN, NEVER ERROR: a flagged cache still
boots a valid district (toowoomba shipped green through the entire v4.0 pack) — E curates, A flags.
selfcheck ALL GREEN 161,372/161,372 (+72); synthetic 0x3fa36874 frozen; NO golden moved (only E's
toowoomba re-bbox may move one this round, per the round's law).
THE THRESHOLD IS MEASURED, NOT PICKED. The brief offered darwin 27 m / toowoomba 388 m as "plenty of
daylight for a line" — but that's 2 points. Measuring all 23 towns shows a near-continuum, and the line
is FORCED into (119.3, 254.9]: braddon at 119.3 m is ALIVE by D's own hub test (7 shops within 160 m,
identical to darwin's 7), ballarat at 254.9 m is not. 150 m sits in the pack's widest gap (x2.14), with
1.26x margin below and 1.70x above. Flags 2 of 23.
TWO TOWNS FLAG, NOT ONE. The brief names only toowoomba (D spot-checked the two 12-shop towns). The
measurement adds BALLARAT at 255 m — nobody has looked at it, and it's independently one of the three
towns my R22 graded cluster fallback fired on (no venue candidate at the >=3-shops/160 m floor). Two
signals agree. Filed for E as a curation call, counted either way.
THE BRIEF'S ANCHORS WERE IN THE WRONG SPACE, AND IT MATTERS. D's 27 m is PLAN space (post-lift metres);
validateTownCache is CACHE space (raw lat/lon — the only thing that exists at ingest, and the layer where
a bad bbox must be caught). Darwin is 68.3 m in cache space, not 27. Interpolating a threshold between
D's figures would have silently mixed spaces. I verified the hypothesis before trusting it: computing the
metric plan-side reproduces D's numbers exactly (darwin 27.2 vs 27, toowoomba 386.8 vs 388, hubs 7 and 3).
Cache space is correct HERE and provably safe: the lift marches shops at a regularised cadence, so it
COMPRESSES dense towns (darwin plan/cache 0.40) but not sparse ones (toowoomba 0.98, ballarat 1.01) --
the error lives at the dense end, the warn only fires at the sparse end. Verified all 23 towns in both
spaces: exactly ONE disagrees, newtown_godverse (94.4 cache passes / 158.3 plan would warn). Filed for
G/Fable, NOT acted on: no cache-space line catches it without false-positiving braddon, and godverse is a
census subset (18 thriftgod shops over the footprint that carries 72 OSM shops) so its sparsity is
coverage, not a bad bbox — re-bboxing, the warn's only remedy, cannot fix it. Worth knowing since R23
stocks the real crate in that town and the beta asks "fitzroy_godverse if the census is dense there".
Also fixed en route: the R17 silent-fallback trap bit again in my own scratch harness — an unregistered
town key resolves to the melbourne fixture rather than erroring, so five towns silently measured
identical. Registered the caches; noting it because it has now cost two rounds.
Coverage: sparse cache warns + still boots (structural + district suites on a flagged town), compact
cache does NOT warn (no false positive), metric edge cases (< 2 shops -> null, coincident -> 0, no throw
on empty), and an exact-value check (a 100 m pair measures 100 m, projection sane).
Cost: ~1 ms/town at registration (25.6 ms for all 23 incl. fitzroy's 160 shops) vs a <100 ms plan budget.
E -> use the warn as your acceptance test for #4: re-bbox, re-run, warn goes silent = the town has a high
street. If your fix moves toowoomba's or ballarat's cache, ping me — goldens are mine to re-pin, and only
what you move.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
292b07aa91