From d159d4c0f288c7891116fb51923a461ddb09ac26 Mon Sep 17 00:00:00 2001
From: wassname <1103714+wassname@users.noreply.github.com>
Date: Sat, 6 Jun 2026 03:35:33 +0000
Subject: [PATCH] route2: fail loud if real v_grad band collapses (extraction
 broken)

Fresh-eyes review flagged that nothing asserted upper>lower for the REAL
v_grad: a broken extraction (hack pairs aligning no more than clean) would
silently degenerate into the random-control sign gate via the max(.,1e-6)
floor. Assert mean band width > 0 on non-Haar runs; the Haar control is still
allowed to collapse. No correctness change to the gate math (review found
conservation, per-rollout recovery, cosine masking, closure capture all OK).

Co-Authored-By: Claudypoo <288921227+claudypoo@users.noreply.github.com>
---
 docs/spec/20260606_pair_routing_design.md |  7 +++++++
 src/vgrout/train.py                       | 11 +++++++++--
 2 files changed, 16 insertions(+), 2 deletions(-)

diff --git a/docs/spec/20260606_pair_routing_design.md b/docs/spec/20260606_pair_routing_design.md
index 07b0d90..3270da4 100644
--- a/docs/spec/20260606_pair_routing_design.md
+++ b/docs/spec/20260606_pair_routing_design.md
@@ -182,6 +182,13 @@ as real). Defence (a) is mandatory; (b) only if (a) shows a mass gap.
 
 ## Implementation plan (src/vgrout/train.py)
 
+STATUS 2026-06-06 (commit 485839d): route rewrite DONE and smoke-verified. `route_band_edges`
+builds the band at extract + on refresh; `_route2_grad_filter` is the banded ramp gate;
+`build_route2_anchors`, the EMA `tau` state, `--gate-anchor-teacher-only`, and
+`scripts/verify_gate_anchor.py` are gone. Smoke: band width +0.289 real vs -0.014 Haar-random;
+`||delta_S_hack||>0`, R3 span assert green, resid~0. DEFERRED: the held-out-pair separation
+gauge (needs a second forward over the `n_val` pairs; diagnostic only, not load-bearing).
+
 Rollback tag `pre-routing-refactor`. erase already works; the code below is the route rewrite.
 
 1. **DELETE `build_route2_anchors`** (~line 337) and its call site. No anchors from teacher
diff --git a/src/vgrout/train.py b/src/vgrout/train.py
index 6098325..b2ed2ad 100644
--- a/src/vgrout/train.py
+++ b/src/vgrout/train.py
@@ -494,10 +494,17 @@ def main(cfg: Config) -> int:
             # Routing band from the pairs (against the FINAL v_grad, so a Haar override
             # collapses the band -- the real-vs-random discriminator).
             route_band = route_band_edges(raw_grads, v_grad, device)
-            _bw = [hi - lo for lo, hi in route_band.values()]
+            _mean_bw = sum(hi - lo for lo, hi in route_band.values()) / len(route_band)
             logger.info(f"route2 band: edges from {len(route_band)} modules, "
-                        f"mean width(upper-lower)={sum(_bw)/len(_bw):+.3f} "
+                        f"mean width(upper-lower)={_mean_bw:+.3f} "
                         f"(>0 = pairs separate; ~0 = random/degenerate)")
+            # On a REAL v_grad the band must open (hack pairs align more than clean).
+            # A collapsed/inverted real band = broken extraction silently mimicking the
+            # random control -> fail loud. The Haar control is allowed to collapse.
+            if cfg.route2_random_v_seed is None:
+                assert _mean_bw > 0, (
+                    f"real v_grad gave non-positive mean band width {_mean_bw:+.3f}: "
+                    "hack pairs do not separate from clean -> extraction broken")
             model.train()
     else:
         # v_hack path resolution, most-specific first. The pairset (personas) is