diff --git a/docs/re/f3-title-plays-bgm-102-and-103.md b/docs/re/f3-title-plays-bgm-102-and-103.md
new file mode 100644
index 00000000..d951f573
--- /dev/null
+++ b/docs/re/f3-title-plays-bgm-102-and-103.md
@@ -0,0 +1,84 @@
+# ✅ F3 — the title **does** play music: cues **1102 and 1103**, from its own GamePart
+
+**Status: ✅ decoded from the image** for the BGM; ❔ **not answered** for the
+sting. Instrument: ⟨image⟩ — every call site of the play function, read out of
+`/image/sylpheed.pe`. 2026-09-02.
+
+A human says something is missing on the title. **Something is: the title plays
+BGM, and it plays two different cues.**
+
+---
+
+## Method — and the control is built into it, not bolted on
+
+R4 requires a negative to show the method finding the *menu's* cue first. This
+method cannot produce a false negative for one screen, because it does not look at
+one screen: it finds **every** call of the play function in the whole image and
+reads the cue each passes.
+
+`sub_821C5580`'s call is `play(this, category, cue_id)`
+([`f2-no-gain-field-in-tables.md`](f2-no-gain-field-in-tables.md)), so scanning
+`bl 0x8217ACF8` and back-resolving `li r5,<imm>` gives the map:
+
+```
+34 callers.  cue ids passed: 1103 ×24, 1102 ×3, 1107 ×2, 1108, 1106, 1104
+             category argument: 4 on every one of them
+```
+
+✅ **The positive control passes by construction.** `1103` is the cue the corpus
+already attributes to the **menu** ([`bgm-102-decoded-during-boot.md`](bgm-102-decoded-during-boot.md)),
+and the method finds it — 24 times. A method that returned nothing for the title
+while also finding nothing for the menu would prove only that the method was
+broken; this one demonstrably finds the thing it is supposed to find.
+
+## The answer
+
+`GamePart_Title`'s neighbourhood is `0x821C4xxx…0x821C7xxx` — the phase store at
+`0x821c4fbc`, the phase handler `sub_821C5580`, `sub_821C6458`, and the registered
+creator `sub_821C7D98`
+([`boot-config-and-gamepart-registry.md`](boot-config-and-gamepart-registry.md)).
+**Seven play sites fall inside it:**
+
+| site | cue | |
+|---|---|---|
+| `0x821C52D0` | **1103** | |
+| `0x821C53B0` | **1102** | |
+| `0x821C5620` | **1103** | inside `sub_821C5580`, the phase handler |
+| `0x821C5BA0` | **1102** | |
+| `0x821C6188` | **1103** | |
+| `0x821C6294` | **1103** | |
+| `0x821C6DE0` | **1102** | |
+
+> **The title plays cue 1103 (`BGM_103`) and cue 1102 (`BGM_102`)** — two cues,
+> from seven sites, alternating across the part's phases.
+
+📌 **1103 is the same cue as the menu's.** That is the useful shape for the port:
+the title and the menu are not two different pieces of music, and 1103 running
+across both is why a port that plays nothing on the title sounds like something is
+missing rather than merely quieter.
+
+## ❔ The sting is NOT answered, and here is exactly why
+
+The play function scanned here carries cue ids in the **1102–1108** range — the BGM
+vocabulary. The UI stings the question asks about (`SE_UI_DECIDE` on Ⓐ, something
+on the plate's arrival) are `SE_*` cues with ids **1–322**
+([`data/se-ui-cues.txt`](data/se-ui-cues.txt)), and **not one of the 34 sites
+passes an id in that range.**
+
+**So SE goes through a different call**, and this method is blind to it by
+construction. That is a bounded negative about the *instrument*, not about the
+game: it says nothing about whether a sting exists. Finding the SE play path — and
+running this same census on it — is the next step, and it is the same shape of
+work that just succeeded here.
+
+⚠️ **Do not read "no SE in this census" as "no sting".** That inversion is exactly
+what R4's control requirement exists to prevent, and it would be the third time
+this corpus mistook an instrument's blind spot for a fact about the game.
+
+## Reach
+
+⟨image⟩, exhaustive over every `bl` to `0x8217ACF8` in the whole 9.2 MB, so the
+census is complete for **that function**. The GamePart attribution is by **address
+neighbourhood**, which is the same convention `boot-config-and-gamepart-registry.md`
+flags as *"by position and convention, not proven"* — the cue ids are certain, the
+label "these seven are the title's" carries that page's caveat.