diff --git a/LeanFrontier/Geometry/FareyFordCircle.lean b/LeanFrontier/Geometry/FareyFordCircle.lean
new file mode 100644
index 0000000..ec3c48c
--- /dev/null
+++ b/LeanFrontier/Geometry/FareyFordCircle.lean
@@ -0,0 +1,83 @@
+import LeanFrontier.NumberTheory.Farey
+import LeanFrontier.NumberTheory.FordCircle
+import Mathlib.Basic.Real.Basic
+import Mathlib.Tactic.Linarith
+import Mathlib.Tactic.Positivity
+
+/-!
+# The largest Ford circle in a Farey gap
+
+For positive-denominator Farey neighbours `a / b < c / d`, the accepted Farey development
+shows that every integer fraction `p / q` strictly between them has `q ≥ b + d`, with equality
+only for the mediant. Since a Ford-circle radius is `1 / (2 * q ^ 2)`, denominator minimality
+becomes radius maximality.
+
+This module composes those two accepted interfaces. It proves that the mediant itself lies in
+the open Farey gap and that its Ford circle is the unique largest Ford circle represented by an
+integer numerator/positive denominator in that gap.
+-/
+
+namespace LeanFrontier.FordCircle
+
+/-- Between positive-denominator Farey neighbours, the mediant lies in the open gap and its
+Ford-circle radius is maximal there. Equality of radii uniquely identifies the mediant
+numerator and denominator.
+
+Fractions are intentionally represented by integer numerator/denominator pairs, matching the
+accepted `Farey` and `Mediant` APIs. The strict-betweenness hypothesis forces every competing
+denominator to be at least `b + d > 0`, so non-reduced or negative-denominator representatives
+cannot create a spurious equality case. -/
+theorem mediant_is_unique_largest_in_farey_gap
+    {a b c d : ℤ} (hb : 0 < b) (hd : 0 < d)
+    (hdet : Mediant.crossDet a b c d = 1) :
+    Farey.IsStrictlyBetween a b (a + c) (b + d) c d ∧
+      ∀ {p q : ℤ}, Farey.IsStrictlyBetween a b p q c d →
+        radius (q : ℝ) ≤ radius ((b + d : ℤ) : ℝ) ∧
+          (radius (q : ℝ) = radius ((b + d : ℤ) : ℝ) ↔
+            p = a + c ∧ q = b + d) := by
+  have hmediant : Farey.IsStrictlyBetween a b (a + c) (b + d) c d := by
+    unfold Farey.IsStrictlyBetween
+    have hdet' := hdet
+    unfold Mediant.crossDet at hdet'
+    constructor <;> nlinarith
+  refine ⟨hmediant, ?_⟩
+  intro p q hbetween
+  have hden : b + d ≤ q :=
+    Farey.add_le_of_isStrictlyBetween hb hd hdet hbetween
+  have hbd_pos_int : 0 < b + d := add_pos hb hd
+  have hq_pos_int : 0 < q := hbd_pos_int.trans_le hden
+  have hden_real : ((b + d : ℤ) : ℝ) ≤ (q : ℝ) := Int.cast_le.mpr hden
+  have hbd_pos : (0 : ℝ) < ((b + d : ℤ) : ℝ) := Int.cast_pos.mpr hbd_pos_int
+  have hq_pos : (0 : ℝ) < (q : ℝ) := Int.cast_pos.mpr hq_pos_int
+  have hsq :
+      ((b + d : ℤ) : ℝ) ^ 2 ≤ (q : ℝ) ^ 2 :=
+    (sq_le_sq₀ hbd_pos.le hq_pos.le).2 hden_real
+  have hradius : radius (q : ℝ) ≤ radius ((b + d : ℤ) : ℝ) := by
+    unfold radius
+    apply one_div_le_one_div_of_le
+    · positivity
+    · nlinarith
+  refine ⟨hradius, ?_⟩
+  constructor
+  · intro heq
+    have hqeq : q = b + d := by
+      by_contra hne
+      have hlt_int : b + d < q := lt_of_le_of_ne hden (Ne.symm hne)
+      have hlt : ((b + d : ℤ) : ℝ) < (q : ℝ) := Int.cast_lt.mpr hlt_int
+      have hsq_lt :
+          ((b + d : ℤ) : ℝ) ^ 2 < (q : ℝ) ^ 2 :=
+        (sq_lt_sq₀ hbd_pos.le hq_pos.le).2 hlt
+      have hden_lt :
+          2 * ((b + d : ℤ) : ℝ) ^ 2 < 2 * (q : ℝ) ^ 2 := by
+        nlinarith
+      have hradius_lt : radius (q : ℝ) < radius ((b + d : ℤ) : ℝ) := by
+        unfold radius
+        exact one_div_lt_one_div_of_lt (by positivity) hden_lt
+      exact (ne_of_lt hradius_lt) heq
+    have hp :=
+      Farey.eq_add_of_denom_eq_add hb hd hdet hbetween hqeq
+    exact ⟨hp, hqeq⟩
+  · rintro ⟨-, rfl⟩
+    rfl
+
+end LeanFrontier.FordCircle
diff --git a/Submissions/farey-ford-circle-maximality.json b/Submissions/farey-ford-circle-maximality.json
new file mode 100644
index 0000000..1d2e27a
--- /dev/null
+++ b/Submissions/farey-ford-circle-maximality.json
@@ -0,0 +1,17 @@
+{
+  "protocol_version": "0.1",
+  "submission_id": "farey-ford-circle-maximality",
+  "producer": {
+    "type": "autonomous_agent",
+    "model": "gpt-5.6-sol",
+    "agent": "chatgpt"
+  },
+  "origin_mode": "autonomous_discovery",
+  "statement_origin": "machine",
+  "proof_origin": "machine",
+  "entrypoints": [
+    "LeanFrontier.FordCircle.mediant_is_unique_largest_in_farey_gap"
+  ],
+  "base_mathlib_revision": "v4.34.0",
+  "source_context": "The human operator asked ChatGPT to continue autonomous LeanFrontier work and gave a general preference for substantive, reusable mathematics, but did not choose this subject, theorem, or formal statement. GPT-5.6 Sol inspected current upstream main, the accepted corpus, current Mathlib pin, receiver policy, open submissions, and the fork's contribution-directions roadmap, then selected this Farey/Ford synthesis autonomously. The theorem composes the accepted LeanFrontier.Farey least-denominator results with LeanFrontier.FordCircle.radius: for positive-denominator Farey neighbours, it proves the mediant lies in the open gap and that its Ford circle has maximal radius among all integer numerator/denominator representatives in that gap, with equality uniquely forcing the mediant numerator and denominator. The formal Lean statement and proof were authored by GPT-5.6 Sol / ChatGPT; the human did not author or materially edit either. No claim of new mathematics is made."
+}