diff --git a/crates/ml/src/trainers/dqn/fused_training.rs b/crates/ml/src/trainers/dqn/fused_training.rs
index 87cbdcf16..bce6d42c9 100644
--- a/crates/ml/src/trainers/dqn/fused_training.rs
+++ b/crates/ml/src/trainers/dqn/fused_training.rs
@@ -3370,9 +3370,10 @@ impl FusedTrainingCtx {
     /// budgets) feed the full-buf trunk/value scaling call — preserving SP5 Layer B
     /// behaviour for shared parameters (D3 decision in SP6 spec).
     ///
-    /// Cold-start floor: ISV slots read 0 before first observation. Floors (0.05 C51,
-    /// 0.05 IQN, 0.02 CQL, 0.02 Ens) are Invariant 1 carve-outs (numerical-stability,
-    /// not tuned constants).
+    /// Cold-start bootstrap: ISV slots read 0 (sentinel) before first observation.
+    /// IQN keeps a structural BASE_IQN=0.11 floor (reference budget, Invariant 1
+    /// carve-out). C51/CQL/ENS use sentinel-aware bootstrap: 0.05/0.02/0.02
+    /// on cold start, controller value verbatim otherwise (SP7 Task 7).
     pub(crate) fn compute_adaptive_budgets(
         &mut self,
     ) -> ([f32; 4], [f32; 4], [f32; 4], [f32; 4], f32, f32, f32, f32) {
@@ -3446,9 +3447,11 @@ impl FusedTrainingCtx {
     /// docstring claimed a "1−iqn−cql−ens" formula that was never
     /// implemented; per `feedback_trust_code_not_docs`, code wins over docs.
     pub(crate) fn last_c51_budget_eff(&self) -> f32 { self.trainer.last_c51_budget_eff }
-    /// SP5 Pearl 2 (flatness-modulated): last per-step ensemble budget (mean
-    /// of 4 per-branch values). Bootstrap floor ENS_BOOTSTRAP_BUDGET=0.02
-    /// applied sentinel-aware in `compute_adaptive_budgets`.
+    /// SP7 (2026-05-03): last per-step ensemble budget (mean of 4 per-branch
+    /// values). ENS has no active controller driver in SP7 (Pearl 2 no longer
+    /// writes BUDGET_ENS_BASE; loss-balance controller covers CQL/C51 only).
+    /// Sentinel-aware bootstrap ENS_BOOTSTRAP_BUDGET=0.02 applied in
+    /// `compute_adaptive_budgets` on cold start / fold boundary.
     pub(crate) fn last_ens_budget_eff(&self) -> f32 { self.trainer.last_ens_budget_eff }
 
     /// F3: Spectral gap — sigma_1 / sigma_2 via rolling Gram + power iteration.
diff --git a/docs/dqn-wire-up-audit.md b/docs/dqn-wire-up-audit.md
index 211815a42..9b79e7f65 100644
--- a/docs/dqn-wire-up-audit.md
+++ b/docs/dqn-wire-up-audit.md
@@ -3921,4 +3921,7 @@ extended. No producer kernel yet — that arrives in the next commit.
   on last_cql_budget_eff, last_c51_budget_eff, last_iqn_budget_eff, and
   last_ens_budget_eff replaced with accurate SP7/SP5 owner references. Layer
   A complete: producer + consumer + observability all wired and consistent.
+  T7 polish landed (commit ⟨pending⟩) — 2 stale docstrings around
+  compute_adaptive_budgets (header floors + last_ens_budget_eff Pearl-2
+  attribution) corrected to reflect the SP7 sentinel-aware bootstrap contract.
 - T9–T10: smoke + 50-epoch verification.