diff --git a/src/xrpld/app/consensus/ConsensusExtensions.cpp b/src/xrpld/app/consensus/ConsensusExtensions.cpp
index 6405b3942..717ab20cb 100644
--- a/src/xrpld/app/consensus/ConsensusExtensions.cpp
+++ b/src/xrpld/app/consensus/ConsensusExtensions.cpp
@@ -565,9 +565,10 @@ ConsensusExtensions::shouldSendExplicitFinalProposal() const
     // We only enable explicit-final when operators intentionally opt in via
     // runtime config/env for measurement/diagnostics.
     //
-    // TBD (2026-03-03): Keep collecting tx-bearing network data before
-    // revisiting whether explicit-final can be safely promoted beyond
-    // experimental use.
+    // TODO: remove the explicit-final proposal path. The implicit accept-time
+    // injection path is the consensus path; explicit-final never found a robust
+    // timing model and still carries separate experimental-only alignment
+    // hazards. Do not promote this by just widening the gates.
     auto const cfg = app_.getRuntimeConfig().getConfig("*");
     if (cfg && cfg->explicitFinalProposal.has_value())
         return *cfg->explicitFinalProposal;
@@ -590,6 +591,9 @@ ConsensusExtensions::buildExplicitFinalProposalTxSet(
     // the agreed set at timeout boundaries. Routing explicit-final
     // (experimental, default-off) through it keeps this path byte-identical to
     // the implicit one. txns.id() is the BASE tx set hash for the fallback.
+    //
+    // TODO: delete this with the explicit-final proposal path; keep this helper
+    // only while the runtime-config experiment still exists.
     auto const selection = selectEntropy(txns.id(), seq);
     uint256 const finalEntropy = selection.digest;
     std::uint8_t const entropyTier = selection.tier;
diff --git a/src/xrpld/app/consensus/ConsensusExtensionsDesign.md b/src/xrpld/app/consensus/ConsensusExtensionsDesign.md
index 807436348..bf6e311f3 100644
--- a/src/xrpld/app/consensus/ConsensusExtensionsDesign.md
+++ b/src/xrpld/app/consensus/ConsensusExtensionsDesign.md
@@ -212,8 +212,9 @@ is the safe degradation path, not a consensus failure.
 
 > Known exception: the experimental, default-off explicit-final proposal path
 > counts alignment over the unfiltered proposer set (not the active view). It is
-> flagged in-code as an outstanding F1 gap and must apply the same active-view
-> filter before it is ever enabled.
+> flagged in-code as an outstanding F1 gap, but the intended disposition is
+> removal rather than repair/promotion; it must not be enabled as a production
+> consensus path.
 
 Examples with six active validators on a UNLReport-anchored view (validator_quorum
 threshold five, participant_aligned threshold four; six is the smallest view with
diff --git a/src/xrpld/consensus/ConsensusExtensionsTick.h b/src/xrpld/consensus/ConsensusExtensionsTick.h
index b479ee44c..0a9faefb6 100644
--- a/src/xrpld/consensus/ConsensusExtensionsTick.h
+++ b/src/xrpld/consensus/ConsensusExtensionsTick.h
@@ -938,10 +938,10 @@ extensionsTick(Ext& ext, Ctx const& ctx)
             // the
             //   normal implicit path (accept-time pseudo-tx injection).
             //
-            // TBD (2026-03-03): We did not find a robust timing model that
-            // folds this into a guaranteed-safe explicit final proposal
-            // across lossy/reordered links without increasing churn. Keep
-            // this path as opt-in for future evaluation.
+            // TODO: remove this explicit-final proposal path. We did not find
+            // a robust timing model that folds it into a guaranteed-safe
+            // explicit final proposal across lossy/reordered links without
+            // increasing churn. It remains opt-in only until deleted.
             {
                 bool fullParticipantCoverage = false;
                 bool entropyAligned = false;
@@ -966,10 +966,9 @@ extensionsTick(Ext& ext, Ctx const& ctx)
                     // main gate. Before this path is EVER enabled it must
                     // filter peers through activeValidatorView()->containsNode
                     // and gate the local +1 on local active-view membership
-                    // (mirror inspectTxConvergedSidecarPeers); until then it
-                    // stays default-off. Per the TBD above, no safe timing
-                    // model was found for this path regardless, so it may never
-                    // ship.
+                    // (mirror inspectTxConvergedSidecarPeers) if this code
+                    // survives long enough to be touched. Preferred disposition
+                    // is deletion, not promotion.
                     auto const participants = ctx.peerPositions.size() + 1;
                     auto const expectedParticipants = ctx.prevProposers + 1;
                     fullParticipantCoverage =
@@ -1297,10 +1296,10 @@ extensionsTick(Ext& ext, Ctx const& ctx)
                 if (exportState.conflict && quorumAligned())
                 {
                     // Export sidecar roots are signed through ExtendedPosition
-                    // whenever featureExport is active. A quorum-aligned hash is
-                    // therefore enough to proceed; requiring every tx-converged
-                    // active peer to publish an exportSigSetHash would let a
-                    // missing minority sidecar force retry/expiry.
+                    // whenever featureExport is active. A quorum-aligned hash
+                    // is therefore enough to proceed; requiring every
+                    // tx-converged active peer to publish an exportSigSetHash
+                    // would let a missing minority sidecar force retry/expiry.
                     JLOG(ext.j_.info())
                         << "Export: exportSigSetHash conflict ignored"
                         << " reason=quorum-aligned"