@@ -11,9 +11,9 @@ Updated 2026-11-07 late morning: housekeeping committed and pushed (1d69021 spec
 084b739 handover, 049f6a1 owner rules/steering, 4835b62 spec status, 103e1aa MIT,
 f11179e frozen prototype plan). The owner approved prototype execution; S2 ran (R28):
 artifact admission receipts collected and Rust digest parity with the Python recorders
-proven on the frozen E4 traces. Attempt-1 U7 charge is 774 canonical lines against the
-frozen 600, so S10 is paused for an owner revision decision (raise, descope, stop);
-S9 may proceed without new charged lines. T4 regenerator turn bytes mismatch their
+proven on the frozen E4 traces. Attempt-1 U7 charge is 774 canonical lines; the owner
+raised the U7 ceiling from 600 to 1,400 on 2026-11-07 (all other targets unchanged),
+so S9 and S10 proceed under the amended budget. T4 regenerator turn bytes mismatch their
 committed sidecars and are inadmissible as bound; E4 turns verify.
 
 - Native-core phase: requirements-generated; requirements/design/tasks unapproved.
 
@@ -88,7 +88,7 @@ and derived ceilings before execution. Never lower targets post hoc.
 | U4 | E4 uses frozen shared core and descriptor semantics | Shared code frozen after E3; only declared E4 scope support changes; diffs/hashes |
 | U5 | Useful complete replacement | $C_{\text{app}} \le 0.80 \cdot B_{\text{app}}$, $B_{\text{app}} - C_{\text{app}} \ge 200$, $C_{\text{total}} \le B_{\text{total}} - 200$, each scope's support smaller, zero seed/observation loss; 3-stratum physical/separated-form reports |
 | U6 | Compatibility and truthful protection | Existing prepare/compare executable/arguments/results/exits checked; differences/residue faithfully reported; pure turn-input transformation checked for accepted and failed turns against frozen protocol examples, including exact permitted feedback and retained failed attempts |
-| U7 | Bounded effort and reproducible evidence | <=600 added/modified canonical maintained lines of shared + adapter-wiring code cumulatively, <=2 attempts; receipts and independent review |
+| U7 | Bounded effort and reproducible evidence | <=1,400 (amended 2026-11-07 from <=600 by owner decision after R28; all other targets unchanged) added/modified canonical maintained lines of shared + adapter-wiring code cumulatively, <=2 attempts; receipts and independent review |
 
 ## U7 effort-charging rules (fail-closed, no brownfield deadlock)
 1. **New and modified apparatus/wiring files charged in full**: Charge the union of all
 
@@ -112,9 +112,15 @@ Report physical, canonical and separated-form statements together; count
 shared code once on each side; tests/fixtures separate; relocated logic is not
 deletion. These are preregistered targets, not measured results.
 
-## 5. U7 effort budget and attempt rules (frozen)
+## 5. U7 effort budget and attempt rules (frozen; amended 2026-11-07)
 
-- `<= 600` added/modified maintained canonical lines of shared + adapter-wiring
+Amendment 2026-11-07 (owner decision after R28): the cumulative U7 ceiling is raised
+from `<= 600` to `<= 1,400` added/modified maintained canonical lines. Reason: S2
+proved byte-parity requires an order/token-preserving JSON model (774 charged lines
+against the 600 estimate, which assumed a thin `serde_json` formatter). All other
+targets — `<= 2` attempts, strata ceilings, U1-U6 — are unchanged and never lowered.
+
+- `<= 1,400` added/modified maintained canonical lines of shared + adapter-wiring
   code cumulatively across `<= 2` attempts; restarts consume attempts;
   deletions do not refund budget.
 - Charge new/modified shared Rust and Python runner/shim files in full at each
 
@@ -1,7 +1,7 @@
 {
   "feature_name": "deconvolution-native-core",
   "created_at": "2026-11-06T19:30:00+01:00",
-  "updated_at": "2026-11-07T10:50:00+00:00",
+  "updated_at": "2026-11-07T11:05:00+00:00",
   "language": "en",
   "phase": "requirements-generated",
   "approvals": {
@@ -27,7 +27,7 @@
     "proposal": "slice-proposal.md",
     "approved_plan": "prototype-plan.md",
     "prototype_execution_approved": true,
-    "prototype_execution_approval": "Owner decision 2026-11-07 (opencode chat): freeze the bounded S9/S2/S10 plan in-repo and execute under U1-U7 without further pause; selective housekeeping commits first (done: 1d69021, 084b739, 049f6a1, 4835b62, 103e1aa).",
+    "prototype_execution_approval": "Owner decision 2026-11-07 (opencode chat): freeze the bounded S9/S2/S10 plan in-repo and execute under U1-U7 without further pause; selective housekeeping commits first (done: 1d69021, 084b739, 049f6a1, 4835b62, 103e1aa). U7 ceiling raised 600->1,400 on 2026-11-07 by owner decision after R28 (774 charged at S2); all other targets unchanged.",
     "mandatory_criteria": [
       "U1",
       "U2",
 
@@ -76,15 +76,22 @@ are reported separately per requirement 6.4 and are not charged to U7.
    "Python-compatible digest emission in Rust" S2 admission prerequisite.
 3. Stop after S9 evidence.
 
+## Amendment 1: U7 revision (owner decision, 2026-11-07)
+The owner raised the U7 cumulative ceiling from 600 to 1,400 canonical lines
+(attempt limit and all other U1-U6 targets unchanged), accepting the recorded
+reason: byte parity requires an order/token-preserving JSON model. S10 may
+proceed after S9 under the amended budget; the U7 ledger continues from 774.
+
 ## Limits and interpretation
 Parity is proven for the Python JSON shapes the apparatus uses (three writer modes and
 their escapes/floats), on the frozen E4 artifacts and generated fixtures; it is not a
 general JSON library claim. Known-scope evidence only.
 
 ## Tool/spec consequence and next gate
-S9 may proceed (existing apparatus, zero new charged lines). S10 implementation stays
-paused until the owner resolves the U7 revision; `spec.json.solution_unlock.status`
-remains `not-run` with U7 forecast failed on attempt 1 unless amended.
+S9 may proceed (existing apparatus, zero new charged lines). Amendment 2026-11-07:
+the owner raised the U7 ceiling to `<=1,400`; S10 proceeds under the amended budget
+with the ledger continuing from 774. `spec.json.solution_unlock.status` remains
+`not-run` until U1-U7 pass with independent evidence review.
 
 ## COBESY check
 Target behavior: agents surface budget overruns immediately with the ledger, instead of
 
@@ -7,8 +7,8 @@ sidecars; the six T4 turn files mismatch/missing (inadmissible as bound). Rust
 `pyjson`/`admission` modules reproduce the Python recorders' byte shapes exactly —
 5.6 MB golden and held-out traces re-emitted byte-identically (nextest 76023a7d, 109/109;
 independent PS + Rust collectors agree on all 21 bindings). U7 charged lines 774 > 600
-frozen before S10; owner revision requested (raise, descope, or stop). S9 may proceed
-(zero new charged lines); S10 paused pending the U7 decision.
+frozen before S10; owner raised the U7 ceiling to 1,400 (2026-11-07, all other targets
+unchanged). S9 next (zero new charged lines); S10 proceeds under the amended budget.
 
 ## Proposal: solution-first native slice (2026-11-07)
 [R27](2026-11-07/01-solution-first-native-slice.md): owner-requested spec amendment,