diff --git a/docs/BACKLOG.md b/docs/BACKLOG.md
index d21a2c8d..51d1e5b4 100644
--- a/docs/BACKLOG.md
+++ b/docs/BACKLOG.md
@@ -7981,6 +7981,7 @@ gate is the wrong shape, validation of the walk is the right one.
 > 🔢 **Filed 2026-08-11 -- found by Session C while building #1051; measured and documented rather than fixed, because a floor changes the accepted-configuration set.** Value **4/10** -- Difficulty **2/10** -- _fill-in_. Two halves. (1) **`0` or a negative value loads clean and dead-letters on the FIRST failure** -- the check is `attempts >= max_attempts` against a post-increment count, so `0` means "give up immediately" while reading like "no limit". (2) **There is no TOML or env spelling for retry-forever**: `""`, `none` and `null` all raise `ValidationError`, so that posture is reachable **per-outbound in code-first configuration only**.
 
 > Both are conditional on a first deployment: an operator who wrote `retry_max_attempts = 0` intending "unlimited" would get single-attempt dead-lettering, silently. Nothing is misconfigured today; there are zero deployments. Note the interaction with **#1051**, which set the finite default to 100 -- a floor should be decided alongside whether retry-forever needs a config spelling at all.
+> **AMENDED 2026-08-14 -- HALF 1 IS SHIPPED; THE ITEM STAYS OPEN ON HALF 2.** PR #383 floors the operator-facing `retry_max_attempts` at `ge=1`, so `0` and negative values are now REFUSED AT LOAD rather than accepted and silently dead-lettering on the first failure. Refused, not clamped: a silently-corrected setting is one the operator never learns was wrong. **The floor is on the OPERATOR-FACING setting only** -- `RetryPolicy(max_attempts=0)` remains a deliberate internal idiom for a permanent no-retry failure (store `mark_failed`), asserted by five live tests, and constraining it would have deleted a used mechanism while claiming to add a guard. `None` stays legal: retry-forever is a documented posture. **HALF 2 -- whether retry-forever needs a TOML or env spelling -- IS UNTOUCHED AND IS A PRODUCT QUESTION**, which this item's own text says should be decided alongside the floor. Banner authored by the Lander under the sanctioned interim, because a builder may not author ledger content (owner ruling 2026-08-13).
 
 > ⚠️ **AMENDED 2026-08-13 (dispatcher) -- HALF 1 BUILT on PR #383, HALF 2 OUTSTANDING, ITEM STAYS OPEN. Do not close this on #383.** Banner authored by the dispatcher rather than the builder, per the owner's 2026-08-13 ruling that a builder may not author ledger content.
 >
diff --git a/messagefoundry/config/settings.py b/messagefoundry/config/settings.py
index 73b66374..0acf4766 100644
--- a/messagefoundry/config/settings.py
+++ b/messagefoundry/config/settings.py
@@ -1056,7 +1056,13 @@ class DeliverySettings(_Section):
     # 100 default (#1051) — a test guards the sync, and the two MUST move together: leaving this at
     # None would restore retry-forever for every outbound that declares no retry= of its own, which
     # is the overwhelmingly common shape.
-    retry_max_attempts: int | None = 100
+    # `ge=1` because a configured 0 loaded clean and dead-lettered on the FIRST failure: the delivery
+    # check is `item.attempts >= max_attempts` against a POST-increment count, so 0 means give up
+    # immediately while READING like "no limit". `None` is the documented retry-forever posture and
+    # stays legal. The floor is on the OPERATOR-FACING setting only -- `RetryPolicy(max_attempts=0)`
+    # remains a deliberate internal idiom for a permanent, no-retry failure (store `mark_failed`), and
+    # constraining that instead would delete a used mechanism while claiming to add a guard.
+    retry_max_attempts: int | None = Field(default=100, ge=1)
     retry_backoff_seconds: float = 5.0
     retry_backoff_multiplier: float = 2.0
     retry_max_backoff_seconds: float = 300.0
diff --git a/tests/test_settings.py b/tests/test_settings.py
index cda62b85..631f7767 100644
--- a/tests/test_settings.py
+++ b/tests/test_settings.py
@@ -13,6 +13,7 @@
 from messagefoundry.config.settings import (
     ApiSettings,
     AuthSettings,
+    DeliverySettings,
     DrSettings,
     ServiceSettings,
     SqlAuth,
@@ -1273,3 +1274,32 @@ def test_an_ip_literal_origin_is_still_fine_with_no_tls_posture_declared() -> No
     developer running on 127.0.0.1 is locked out."""
     s = ApiSettings(public_origin="http://127.0.0.1:8765", serve_ui=True)
     assert s.public_origin == "http://127.0.0.1:8765"
+
+
+@pytest.mark.parametrize("bad", [0, -1, -100])
+def test_retry_max_attempts_below_one_is_refused_at_load(bad: int) -> None:  # #1217
+    """`0` loaded clean and dead-lettered on the FIRST failure, while READING like "no limit".
+
+    The delivery check is ``item.attempts >= max_attempts`` against a POST-increment count, so a
+    configured `0` means give up immediately -- the opposite of what an operator writing `0` in a
+    retry field almost certainly intends, and the opposite of what `None` means two lines above it.
+    A negative value was equally accepted and equally silent.
+
+    Refused at LOAD rather than clamped, because a silently-corrected setting is a setting the
+    operator never learns was wrong.
+    """
+    with pytest.raises(ValidationError):
+        DeliverySettings(retry_max_attempts=bad)
+
+
+def test_retry_forever_and_the_finite_default_both_survive_the_floor() -> None:  # #1217
+    """The negative control, and it guards the two values the floor must NOT touch.
+
+    `None` is retry-forever and is a legitimate, documented posture -- a floor that refused it would
+    remove a capability while claiming to add a guard. `100` is the finite default #1051 set, and the
+    settings model mirrors `RetryPolicy`'s; a test already guards that sync and the two must move
+    together.
+    """
+    assert DeliverySettings(retry_max_attempts=None).retry_max_attempts is None
+    assert DeliverySettings().retry_max_attempts == 100
+    assert DeliverySettings(retry_max_attempts=1).retry_max_attempts == 1