re: decode the dialog table -- 70/70 records, DLG_SELECT_DIFFICULTY is id 2000

Chasing the reach I recorded this morning found the binding it said was missing.
Every DLG_ string in the image is pointed at by one aligned word at a 12-byte
stride: {u32 handler, u32 id, u32 name_ptr}, spanning 0x820A0A2C..0x820A0D68 with
three distinct handlers. Complete -- 70 names, 70 records, none unmatched -- and the
ids are banded and monotonic with a single gap at 24. Read from the image directly.

Refutation attempt on the shared reach, which both agents had recorded: 'another
four-button dialog with the same rows would be indistinguishable'. Scanned every
build in every pak for four buttons within 6 px of 259/329/399/469. Control found
both incumbents; zero rivals exist anywhere on the disc. So the geometric
identification is unique disc-wide, which is stronger than what either of us
claimed.

Still unbound: id 2000 to a pak entry. The tie remains uniqueness plus the oracle
capture, not a pointer.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Wuu56cE8vJGTBtn1ppsk8v

crates/sylpheed-formats/examples/four_button_row_rivals.rs

@@ -0,0 +1,55 @@
+//! Is `GP_DIALOG` 2/3 the ONLY build on the disc with four buttons at 259/329/399/469?
+//!
+//! Both agents recorded the same reach on the DIFFICULTY identification: entries
+//! 2/3 are picked out by button count and geometry, not by a binding from
+//! `DLG_SELECT_DIFFICULTY` to a pak entry, so "another four-button dialog with the
+//! same rows would be indistinguishable". This tests whether such a rival exists.
+//!
+//! CONTROL: the scan must find GP_DIALOG 2 and 3 themselves. A rival-search that
+//! cannot find the incumbent proves nothing.
+//!
+//!     cargo run -p sylpheed-formats --example four_button_row_rivals
+use sylpheed_formats::{pak::PakArchive, ui_layout};
+use std::path::PathBuf;
+
+const WANT: [i32; 4] = [259, 329, 399, 469];
+
+fn main() {
+    let root = PathBuf::from(std::env::var("SYLPHEED_DISC").expect("SYLPHEED_DISC"));
+    let mut paks: Vec<_> = std::fs::read_dir(root.join("dat"))
+        .expect("dat")
+        .filter_map(|e| e.ok().map(|e| e.path()))
+        .filter(|p| p.extension().map(|x| x == "pak").unwrap_or(false))
+        .collect();
+    paks.sort();
+    let (mut incumbent, mut rivals) = (0, 0);
+    for p in &paks {
+        let Ok(ar) = PakArchive::open(p) else { continue };
+        let pname = p.file_name().unwrap().to_string_lossy().to_string();
+        for (i, e) in ar.entries().iter().enumerate() {
+            let Ok(by) = ar.read(e) else { continue };
+            let Some(b) = ui_layout::parse_build(&by) else { continue };
+            let mut ys: Vec<i32> = b
+                .elements
+                .iter()
+                .filter(|el| el.name.contains("btn") && !el.name.contains('f'))
+                .map(|el| el.rest().map(|k| k.y).unwrap_or(el.pivot_y as i32))
+                .collect();
+            ys.sort();
+            ys.dedup();
+            if ys.len() != 4 {
+                continue;
+            }
+            let close = ys.iter().zip(WANT.iter()).all(|(a, b)| (a - b).abs() <= 6);
+            if close {
+                let is_inc = pname == "GP_DIALOG.pak" && (i == 2 || i == 3);
+                if is_inc { incumbent += 1 } else { rivals += 1 }
+                println!("  {}{pname:24} entry {i:4}  rows {ys:?}",
+                         if is_inc { "INCUMBENT " } else { "RIVAL     " });
+            }
+        }
+    }
+    println!("\ncontrol: found {incumbent} incumbent build(s) (want 2) — {}",
+             if incumbent == 2 { "PASSED" } else { "FAILED" });
+    println!("{rivals} rival build(s) elsewhere on the disc");
+}

docs/re/data/difficulty-is-a-dialog.txt

@@ -64,3 +64,51 @@
 # the DLG_* names live in a string list, and what maps a name to an archive entry
 # is not decoded. Another 4-button dialog with the same rows would be
 # indistinguishable by this evidence.
+
+################################################################################
+# ✅ DECODED 2026-08-31 (later): THE DIALOG TABLE. `DLG_SELECT_DIFFICULTY` = id 2000.
+#
+# Chasing the reach above -- "no binding from the DLG_ name to anything" -- found
+# one. Every DLG_ string in the image is pointed at by exactly one aligned word,
+# at a regular 12-byte stride. The table is:
+#
+#     struct { u32 handler; u32 id; u32 name_ptr; }   // 12 bytes, big-endian
+#
+# spanning 0x820A0A2C .. 0x820A0D68, three distinct handler values
+# (0x821CFD80, 0x821D05D8, 0x821D0808).
+#
+# ✅ COMPLETE: 70 DLG_ names in the image, 70 records, ZERO names without one.
+# ids are banded and monotonic, with a single gap at 24:
+#     0..23   sign-in, storage, save/load/replay, GO_TITLE, GO_NEXT
+#     25..43  autosave, training, weapon develop, custom keys, menu jumps
+#     98, 99  DLG_MESSAGE_BOX_YES_NO, DLG_MESSAGE_BOX
+#     1000    DLG_MISSION_OBJECTIVE
+#     2000    **DLG_SELECT_DIFFICULTY**
+#     2001..3 SYSTEM_PAUSE, LEADERBOARD_MENU_JUMP, LEADERBOARD_MENU_NEXT
+#     2100..15 DLG_STAGE_TITLE01..16
+#     8000    DLG_RATING_ESRB
+#     9000..2 trial dialogs
+#
+# Read from /image/sylpheed.pe directly -- these are the bytes the console ran,
+# not a database row.
+#
+################################################################################
+# ✅ REFUTATION ATTEMPT ON THE SHARED REACH -- IT SURVIVES, AND IS NOW BOUNDED.
+#
+# Both agents recorded: "another four-button dialog with the same rows would be
+# indistinguishable by this evidence". Tested by scanning EVERY build in EVERY pak
+# for four buttons within 6 px of rows 259/329/399/469
+# (examples/four_button_row_rivals.rs):
+#
+#   INCUMBENT  GP_DIALOG.pak entry 2   rows [259, 329, 399, 469]
+#   INCUMBENT  GP_DIALOG.pak entry 3   rows [259, 329, 399, 469]
+#   control: 2 incumbents found (want 2) -- PASSED
+#   RIVALS ELSEWHERE ON THE DISC: **0**
+#
+# So the hypothetical rival does not exist here. The geometric identification is
+# UNIQUE DISC-WIDE, which is a stronger statement than the one either of us
+# recorded, and it was cheap to get.
+#
+# ⚠️ WHAT IS STILL NOT BOUND: id 2000 -> a pak entry. The table gives name -> id
+# and the disc gives a unique four-button build; nothing found so far connects the
+# two. The tie is uniqueness of geometry plus the oracle capture, not a pointer.