feat(simulation): T-1170 A2/A3 + T-1168 A5 — course inventor, coast termination, riparian signal

A2 (river_course.rs): Stage A rung-independent valley-seeking control
path (chord/8 stations, k=5 bilinear-scored candidates + continuity
penalty); Stage B rung-indexed perpendicular warp on GLOBAL arc-length
(window-independence, Ruling 1e), band chord/2 down to min_wavelength_m
hard-truncate, sine taper to zero at anchors, amplitude min(8% chord,
half-cell) slope/class-scaled. SeedDomain::RiverCourse=17, distinct
salt. Wire: RiverCourse{edge_id,class,points,terminus} on
DistrictWindowLayer.courses (serde-default); bbox-culled, window-
cropped +1 station. TerrainAnalysisCache retains Layer1Output (the
gen_queue:626 discard, Ruling 4b). A3: mouth termination walks
stations sampling the window's OWN rung-consistent morphology verdict,
6-iteration bisect; land-at-anchor probes one segment then None;
EdgeDrain never probes. A5: near_perennial_water point-to-segment
predicate (D-239 §8 governed bands 1-3m class-scaled) threaded through
both batch and window paths; Region always false; never touches
moisture_q. Discipline closed: dormant zoom-ladder bench run + numbers
recorded in the design doc (District 3.009/1.785, Quarter 1.823
us/cell, Region window 0.617ms); course-cost bench CAUGHT a real
+12-36% per-cell riparian scan regression -> precomputed bbox O(1)
reject (60ns->2.2ns/call), final delta +3.4-6.9% at budget; three
determinism tests (overlapping-window byte-identity, cross-rung
amplitude bound, warp-stream cross-correlation r<0.3); goldens: window
sweep gained a verified course-bearing position (pure append), new
river_course golden at both rungs, believability verified unchanged.
Revert-verification discovered the pole-row branch is structurally
unreachable (flow_direction bounds-check) — the real edge-drain path
is k<0 flat-plateau; test fixture rewritten to exercise reality.
scale.rs stale comment fixed. Full cargo test green.

Tickets: T-1170, T-1168

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

docs/architecture/atlas-zoom-ladder-t1143.md

@@ -156,6 +156,20 @@ Release build, this session, body GJ338Bd, seed "yolo", 16 cores / Rayon pool 14
 | Octave-cutoff derive (`min_wavelength_m`-bearing) | — | not measured | **UNBUILT, UNMEASURED** — every planetary-rung and canvas-sampling latency claim resting on "cheaper because fewer octaves" is an unmeasured extrapolation on top of code that doesn't exist yet |
 | Full-canvas planetary sample, 1920×1080 (~2.07M cells) | — | ~2.6–3.0 s single-thread; ~0.2–0.3 s parallel (claimed) | **ESTIMATED, UNVERIFIED** — depends on both the unbuilt cutoff and the unbenchmarked parallel throughput above; treat as directional only |
 
+> **UNMEASURED gap closed (2026-07-23, T-1170 A2 discipline item 1 — Dudley).** `zoom_ladder_bench.rs`'s three `#[ignore]`d benches (dormant since T-1149/T-1152) were run `--release --ignored` on this session's hardware (16 cores, same class of machine as the §7 table above) before the T-1170 course inventor landed, per the batch's own discipline requirement. Numbers below are **MEASURED**, not extrapolated — this table row is retired as UNBUILT/UNMEASURED for the octave-cutoff case specifically (the cutoff plumbing has existed since T-1162; what was missing was ever actually running the bench):
+>
+> | Sweep (64×64 = 4,096 cells, release) | Total | ns/cell | µs/cell |
+> |---|---|---|---|
+> | District spacing (2,048 m), cutoff=0 (uncut) | 12.33 ms | 3,009.4 | 3.009 |
+> | District spacing (2,048 m), cutoff=2,048 m | 7.31 ms | 1,784.7 | 1.785 |
+> | Quarter spacing (512 m), cutoff=512 m | 7.47 ms | 1,822.9 | 1.823 |
+> | Orbital (`derive_orbital_at_metres`), Region spacing (204.8 km) | 5.96 ms | 1,454.2 | 1.454 |
+> | Full `derive_at_metres` (invention included) at Region spacing, for comparison | 11.18 ms | 2,729.4 | 2.729 |
+> | Orbital speedup vs. full derive at the same (Region) spacing | — | — | **1.88×** |
+> | Served Region window at `DISTRICT_WINDOW_MAX_N_REGION` (n=6,400 → 64×64=4,096 cells, `WIRE_CAP_CELLS` capped, real production `build_district_window_layer` path, row-chunked `par_iter`, 16 Rayon threads) | 12.35 ms / 20 calls | — | **0.617 ms/call** |
+>
+> Extrapolated full-canvas figures (1600×900 @ 1/2 px-per-cell) from the orbital per-cell rate: ~2,094 ms single-thread @1px/cell, ~524 ms @2px/cell — still **EXTRAPOLATED**, not independently measured at full canvas size (that remains a separate, not-yet-run measurement; the per-cell rate itself is now real). The octave cutoff itself is confirmed cheaper (cutoff=2,048 m runs at ~59% of the uncut cost at District spacing) rather than assumed cheaper. This closes the "not load-bearing for T-1170's design, but the gap closes now" item from the T-1170 ruling's Discipline §1 — the course inventor's Stage A/B split does not depend on these numbers (it deliberately avoids per-candidate `derive_at_metres` calls, Ruling 3b), but the gap this table flagged is no longer open.
+
 **Interactive-latency verdict:** district and quarter rungs (capped) are comfortably interactive on a warm `TerrainAnalysis` — 1–5 ms per response at current cap sizes. The **only** rung with a real latency question is planetary/canvas-fill, and its numbers are two extrapolation-hops from anything actually measured this session. **No implementation should proceed on the planetary rung's cost story without first (a) prototyping the octave cutoff and re-measuring, and (b) chunked-`par_iter`-ing the window loop and re-measuring** — both are cheap to do (the probe binary already exists) and should happen before, not during, the follow-up ticket.
 
 ---