diff --git a/.github/workflows/tests-rs-wallet.yml b/.github/workflows/tests-rs-wallet.yml
index 64cc36997d9..ed8bc622717 100644
--- a/.github/workflows/tests-rs-wallet.yml
+++ b/.github/workflows/tests-rs-wallet.yml
@@ -201,10 +201,22 @@ jobs:
             --locked \
             -- --no-deps -D warnings
 
-      # Mirrors the workspace job's non-shielded step for the wallet crates:
-      # same package subset, same `not test(~shield)` filter (shielded wallet
-      # tests are not run there either).
-      - name: Run wallet tests (non-shielded)
+      # No test filter: this job runs everything in the three wallet packages,
+      # shielded suite included. There is nothing here worth excluding —
+      # `wallet::shielded::` is 145 tests in ~6 s wall (~97 s of CPU) under
+      # nextest, and `--all-features` compiles those binaries whether or not
+      # they are selected, so skipping them saved execution time on artifacts
+      # already paid for and thrown away.
+      #
+      # This job and tests-rs-workspace.yml are mutually exclusive paths for a
+      # given PR, so the shielded suite has to run in BOTH or a wallet-scoped
+      # PR silently loses it. The workspace job carves the wallet packages out
+      # of its own shield exclusion for the same reason.
+      #
+      # `--no-tests fail` is the zero-match guard: a selection that matches
+      # nothing must fail the step, not pass green. Pinned rather than left to
+      # nextest's default, which is a default and not a promise.
+      - name: Run wallet tests
         run: |
           cargo nextest run \
             --package platform-wallet \
@@ -212,7 +224,7 @@ jobs:
             --package platform-wallet-ffi \
             --all-features \
             --locked \
-            -E 'not test(~shield)'
+            --no-tests fail
         env:
           RUST_MIN_STACK: 4194304
           CARGO_PROFILE_DEV_DEBUG: "0"
diff --git a/.github/workflows/tests-rs-workspace.yml b/.github/workflows/tests-rs-workspace.yml
index 1c159c2c706..f9891e528bd 100644
--- a/.github/workflows/tests-rs-workspace.yml
+++ b/.github/workflows/tests-rs-workspace.yml
@@ -324,6 +324,20 @@ jobs:
       # caught minutes after merge — the same safety-net pattern as the
       # shielded phase. Keeping this phase's filter identical across events
       # keeps the `rust` codecov flag comparable between PR and push runs.
+      #
+      # `test(~shield)` is a substring match on the FULL test path, so it does
+      # not mean "the shielded suites" — it means "anything with the word in
+      # its name". The shielded phase below compensates for dpp / drive /
+      # drive-abci / dash-sdk, which genuinely need it: they are Orchard
+      # proving tests, and that phase runs them under `cargo test` in a shared
+      # process so one verifying key is built instead of one per test.
+      #
+      # The three wallet packages have no such compensator and no such need —
+      # their whole `wallet::shielded::` suite is 145 tests in ~6 s under
+      # nextest's process-per-test model — so excluding them dropped 145
+      # shielded tests plus ~47 more that merely contain the word (input
+      # selection, FFI error codes, memo encoding, SQLite viewing-key rows)
+      # into a gap no job covered. They are carved out of the exclusion here.
       - name: Run non-shielded tests with nextest (parallel, compiles all packages)
         if: steps.coverage-cache.outputs.reuse == 'false'
         run: |
@@ -352,7 +366,8 @@ jobs:
             --package keyword-search-contract \
             --all-features \
             --locked \
-            -E 'not test(~shield) and (not binary_id(=drive-abci::strategy_tests) or test(~comprehensive_mixed_operations))'
+            --no-tests fail \
+            -E '(not test(~shield) or package(platform-wallet) or package(platform-wallet-storage) or package(platform-wallet-ffi)) and (not binary_id(=drive-abci::strategy_tests) or test(~comprehensive_mixed_operations))'
         env:
           RUST_MIN_STACK: 4194304
           CARGO_PROFILE_DEV_DEBUG: "0"
@@ -370,6 +385,9 @@ jobs:
       # The chain simulations excluded from the phase above. PR runs skip
       # them (and upload no rust-strategy coverage — codecov carries the base
       # commit's forward); push, nightly, and dispatch runs execute them all.
+      # The `~shield` exclusion here IS compensated — drive-abci is in the
+      # shielded phase below — so it stays; `--no-tests fail` is the guard
+      # against the whole expression silently selecting nothing.
       - name: Run full chain-simulation suite (push and nightly only)
         id: strategy-tests
         if: >-
@@ -380,6 +398,7 @@ jobs:
             --package drive-abci \
             --all-features \
             --locked \
+            --no-tests fail \
             -E 'binary_id(=drive-abci::strategy_tests) and not test(~comprehensive_mixed_operations) and not test(~shield)'
         env:
           RUST_MIN_STACK: 4194304
diff --git a/packages/rs-platform-wallet/src/wallet/platform_wallet.rs b/packages/rs-platform-wallet/src/wallet/platform_wallet.rs
index c1ae06ba180..2b316334032 100644
--- a/packages/rs-platform-wallet/src/wallet/platform_wallet.rs
+++ b/packages/rs-platform-wallet/src/wallet/platform_wallet.rs
@@ -2303,23 +2303,24 @@ mod shield_input_selection_tests {
 
     #[test]
     fn regression_reports_max_from_usable_suffix_not_total_account_balance() {
-        // Real account snapshot: the leading address is below the reserve, so
+        // Account snapshot whose leading address cannot pay the fee, so
         // capacity must come from the usable suffix, not the account total.
-        assert!(
-            297_264_780 <= reserve(),
-            "regression shape requires the leading address to stay below the reserve; \
-             re-seed the balances if the versioned reserve drops under 297_264_780"
-        );
+        // The leading balance is derived from the reserve — one credit below
+        // the strict `> reserve` viability threshold, the largest balance that
+        // must still be rejected as input 0 — so the shape holds whatever the
+        // versioned fee schedule does next.
+        let dust = reserve() - 1;
+        let usable = 3_623_849_220;
         let candidates = vec![
-            (addr(1), 297_264_780),
+            (addr(1), dust),
             (addr(2), 2_000_000_000),
             (addr(3), 1_623_849_220),
         ];
         let plan = plan(candidates).unwrap();
-        let expected_max = 3_623_849_220 - reserve();
+        let expected_max = usable - reserve();
 
-        assert_eq!(plan.preflight.account_balance_credits, 3_921_114_000);
-        assert_eq!(plan.preflight.usable_balance_credits, 3_623_849_220);
+        assert_eq!(plan.preflight.account_balance_credits, dust + usable);
+        assert_eq!(plan.preflight.usable_balance_credits, usable);
         assert_eq!(plan.preflight.fee_reserve_credits, reserve());
         assert_eq!(plan.preflight.max_shieldable_credits, expected_max);
         assert!(plan.preflight.can_shield);
@@ -2329,11 +2330,13 @@ mod shield_input_selection_tests {
         assert!(!chosen.contains_key(&addr(1)));
         assert_eq!(chosen.values().sum::<u64>(), expected_max);
 
+        // `available` reports the usable suffix, never the account total —
+        // the whole point of the regression.
         let err = plan.select_inputs(expected_max + 1).unwrap_err();
         assert!(matches!(
             err,
             PlatformWalletError::PlatformShieldCapacityExceeded { available, required }
-                if available == 3_623_849_220 && required == 3_623_849_221
+                if available == usable && required == usable + 1
         ));
     }