## [4.7.0] - 2026-07-11

Service Areas rebuilt as exact isochrones with pedshed analysis; every
tool's help now teaches how to interpret its results.

### Changed
- **Service Areas (Isochrones) rebuilt** - the old implementation kept
or dropped WHOLE street segments (an edge with one endpoint under the
break was included entirely, overshooting by up to a full block) and
offered only buffer polygons. Reach is now computed exactly: streets
are trimmed at the precise point where the cost budget runs out,
including meet-in-the-middle coverage from both ends and mid-edge
facility entries. Facilities enter the network at the nearest point of
the nearest street (not the nearest junction), and with length costs
the straight-line approach distance is deducted from the budget.
- **Every algorithm's help (64/64) now ends with a "How to read the
results" section**: per-field meaning in planning terms, reference
values where the literature has them (NACH benchmarks, LTS audiences,
pedshed and LCRPGR ranges, dB(A) and headway thresholds...), and
practical next steps. Existing descriptions kept verbatim.

### Added
- Service Areas: **polygon methods** - street buffer (hugs the trimmed
network), concave hull (isochrone blob, detail parameter, convex
fallback on older GEOS) and convex hull; **per-facility catchments**
(optional facility label field) alongside the merged nearest-facility
scope; **rings mode** (band differences) for clean cartography.
- Service Areas: **straight-line catchment circles** per facility and
break - the regulation "desired reach" radius (e.g. a 500 m school
catchment) drawn next to the real network catchment.
- Service Areas: **catchment summary table with the pedshed ratio**
(network catchment area / circle area), circle and network areas, and
the trimmed reached street length per scope and break.
- Reached-street output now carries facility, band, cost_from and
len_m per trimmed piece - pieces are split at the break boundaries for
exact band styling.
- New pure-NumPy `engine/isochrone.py`: vectorized partial-edge reach
fractions, interval algebra (merge/subtract/length), polyline cutting
by arc-length fraction and mid-edge entry budgets - all hand-fixture
unit-tested.

### Fixed
- README test counts were stale (387 unit / 317 e2e); now 425 / 363
and released numbers will track the suites.

### Tested
- 425 engine unit checks (19 new isochrone fixtures: trimming,
meet-in-the-middle, interval algebra, polyline cuts, entry budgets)
and 363 real-QGIS e2e assertions on QGIS 3 LTR and QGIS 4, including
hand-derived exactness: 1000/3400 m trimmed street length on the grid
fixture, an 80,000 m2 convex-hull diamond with pedshed = 2/pi on a
cross network, and a 60 m entry piece after a 10 m snap. Dashboard
harness 24/24 on both QGIS.

None