SeymourBETA

Update 2.14

6/6/2026

v2.14.0BETA

Seymour

JP text decoder reads the 2-byte KANJI (payoff of today's crack):

the FfxEncoding lossless codec now recognizes the 2-byte font-bank leads (0x06/0x26-0x2F, proven in docs/reverse/FFX_EVENT_TEXT_ENCODING_CRACKED_2026-06-06.md) and surfaces readable glyph refs — <FTCX:n> (this event's kanji), <K:n> (base.ftc kanji), <F2/F3/F5:n> (other banks) — instead of the old garbled <C..>+<MISS>. Fully symmetric encoder (token↔bytes exact). New gate --jptext-rt0 PROVES on the event corpus (367 files, 18,935 JP scripts, 15,467 with 2-byte glyphs): glyph drift = 0 + codec idempotent (pre-existing single-byte char aliases canonicalize once then stay fixed = safe; no-edit already returns the original bytes). Directly benefits BattleTextExplorer (uses the lossless codec). offline_ci = 23 gates PASS. csproj bump (user-visible capability)

v2.14.1BETA

Seymour

MonMagic Save is now byte-safe (the last drifting writer, killed):

Ability_Command.WriteList got a preserve-only path: it captures the original text-pool offsets at read and, on a save WITHOUT a text edit, re-emits the original pool verbatim (placing each script at its original offset, so shared strings — e.g. monmagic's many empty fields all at offset 0 — stay SHARED instead of being duplicated). Read-back verification: any length/content-changing text edit falls back to the usual append. This kills the monmagic drift without the dedup that broke command/item (previously refuted). AbilityCommandLab gate: command/item RT0 4/4 + monmagic RT0 4/4 (was 0/4 drift) + 1-flag edit localized for all; monmagic PROMOTED to CERTIFIED (a drift now FAILS the gate). offline_ci PASS. csproj bump (the kernel editor's monmagic Save is now byte-faithful)

v2.14.2BETA

Seymour

BattleTextExplorer gains the JP source (real kanji, no more `<MISS>`) + 2 new gates:

BattleTextExplorer_DataModel.ReloadSources now loads jppc/battle/kernel/btl_txt.bin through the JpDecoder — the visible payoff of the 2-byte crack (v2.14.0). Proven headless by the new --btltext-jp-decode gate (Tools/BtlTextJpDecodeRt0): JP btl_txt.bin = 128 entries / 451 words → 44 kanji surface as <K:n> (base.ftc bank), 0 <MISS>, with a lossless round-trip 451/451 (decode→encode==bytes). Reality refined the guess: battle text uses only the common <K:n> bank (no <FTCX>/F2/F3/F5 — those are per-event). Phantom-flag bug fixed: the --btltexttable-rt0 flag was advertised in BtlTextTableRt0.cs's header but never wired into Program.cs — running it in the background fell through to the GUI (WinExe, 0 stdout) and "died with no output". Now wired (JP default, language fallback): JP PASS (3982/3982) + US PASS (1910/1910). --wave1-rt0 still 12/12 PASS; --btltext-jp-decode wired into offline_ci.ps1 → 25 gates PASS (was 24). (The 2 gates would be doc-only; the csproj bump is for the JP source in the explorer UI.)

v2.14.3BETA

Seymour

Monster Commands (monmagic) open-crash KILLED + workspace persistence + Aurora names:

Crash: opening "Monster Commands 1/2" killed the editor (NPE in Ability_Command.WriteList→WriteBytesIntoTextFile, fired in the ctor's BuildFile()). Root cause: the model was renamed UnusedText1/2*→JapaneseOnlyText1/2* and gained Original*Offset (v2.14.1 preserve-only), but the UI wrappers (KernelCommands_Wrapper, MonsterStatSheet_Wrapper) kept the OLD names and lacked the offsets → PropertyUtil.CopyProperties dropped those fields on Wrap/Unwrap → null → preserve-only failed, fell to the append rebuild, and NPE'd on monmagic's empty JP-only fields. Fix: full round-trip in both wrappers (rename + Original*Offset) + null-safe FfxEncoding.WriteBytesIntoTextFile. New gate --kernelcmd-roundtrip (Tools/KernelCommandRoundtripRt0) proves the EXACT GUI path (ReadList→Wrap→Unwrap→WriteList) byte-identical 4/4 (command/item/monmagic1/2) — which the AbilityCommandLab (direct path) structurally couldn't see. Wired into offline_ci → 26 gates PASS.
Workspace persistence: the loaded master folder is saved to %LocalAppData%\FFXProjectEditor\last-project.txt and auto-restored on startup (priority: CLI arg → last loaded → defaults). Load once, remembered forever; no more defaulting to the Steam-mod folder.
Global crash logger (%LocalAppData%\FFXProjectEditor\crash.log, via Utils/CrashLog) + defensive load in KernelCommands (explanatory red error instead of killing the app).
Aurora Chamber: monster anchors now show the NAME (🔴 <name>) instead of "monster (live)".
Doc-only ride-along: docs/governance/FFX_EDITOR_UI_CONVENTIONS.md (panel standard = Shop Explorer, not a 50-column DataGrid) + DINPUT8 rule revoked (probe is free) + multi-language/MagicViewer notes in the checklist. csproj bump (UI features).

Back to updates

SeymourBETA

Update 2.14

6/6/2026

v2.14.0BETA

Seymour

JP text decoder reads the 2-byte KANJI (payoff of today's crack):

the FfxEncoding lossless codec now recognizes the 2-byte font-bank leads (0x06/0x26-0x2F, proven in docs/reverse/FFX_EVENT_TEXT_ENCODING_CRACKED_2026-06-06.md) and surfaces readable glyph refs — <FTCX:n> (this event's kanji), <K:n> (base.ftc kanji), <F2/F3/F5:n> (other banks) — instead of the old garbled <C..>+<MISS>. Fully symmetric encoder (token↔bytes exact). New gate --jptext-rt0 PROVES on the event corpus (367 files, 18,935 JP scripts, 15,467 with 2-byte glyphs): glyph drift = 0 + codec idempotent (pre-existing single-byte char aliases canonicalize once then stay fixed = safe; no-edit already returns the original bytes). Directly benefits BattleTextExplorer (uses the lossless codec). offline_ci = 23 gates PASS. csproj bump (user-visible capability)

v2.14.1BETA

Seymour

MonMagic Save is now byte-safe (the last drifting writer, killed):

Ability_Command.WriteList got a preserve-only path: it captures the original text-pool offsets at read and, on a save WITHOUT a text edit, re-emits the original pool verbatim (placing each script at its original offset, so shared strings — e.g. monmagic's many empty fields all at offset 0 — stay SHARED instead of being duplicated). Read-back verification: any length/content-changing text edit falls back to the usual append. This kills the monmagic drift without the dedup that broke command/item (previously refuted). AbilityCommandLab gate: command/item RT0 4/4 + monmagic RT0 4/4 (was 0/4 drift) + 1-flag edit localized for all; monmagic PROMOTED to CERTIFIED (a drift now FAILS the gate). offline_ci PASS. csproj bump (the kernel editor's monmagic Save is now byte-faithful)

v2.14.2BETA

Seymour

BattleTextExplorer gains the JP source (real kanji, no more `<MISS>`) + 2 new gates:

BattleTextExplorer_DataModel.ReloadSources now loads jppc/battle/kernel/btl_txt.bin through the JpDecoder — the visible payoff of the 2-byte crack (v2.14.0). Proven headless by the new --btltext-jp-decode gate (Tools/BtlTextJpDecodeRt0): JP btl_txt.bin = 128 entries / 451 words → 44 kanji surface as <K:n> (base.ftc bank), 0 <MISS>, with a lossless round-trip 451/451 (decode→encode==bytes). Reality refined the guess: battle text uses only the common <K:n> bank (no <FTCX>/F2/F3/F5 — those are per-event). Phantom-flag bug fixed: the --btltexttable-rt0 flag was advertised in BtlTextTableRt0.cs's header but never wired into Program.cs — running it in the background fell through to the GUI (WinExe, 0 stdout) and "died with no output". Now wired (JP default, language fallback): JP PASS (3982/3982) + US PASS (1910/1910). --wave1-rt0 still 12/12 PASS; --btltext-jp-decode wired into offline_ci.ps1 → 25 gates PASS (was 24). (The 2 gates would be doc-only; the csproj bump is for the JP source in the explorer UI.)

v2.14.3BETA

Seymour

Monster Commands (monmagic) open-crash KILLED + workspace persistence + Aurora names:

Crash: opening "Monster Commands 1/2" killed the editor (NPE in Ability_Command.WriteList→WriteBytesIntoTextFile, fired in the ctor's BuildFile()). Root cause: the model was renamed UnusedText1/2*→JapaneseOnlyText1/2* and gained Original*Offset (v2.14.1 preserve-only), but the UI wrappers (KernelCommands_Wrapper, MonsterStatSheet_Wrapper) kept the OLD names and lacked the offsets → PropertyUtil.CopyProperties dropped those fields on Wrap/Unwrap → null → preserve-only failed, fell to the append rebuild, and NPE'd on monmagic's empty JP-only fields. Fix: full round-trip in both wrappers (rename + Original*Offset) + null-safe FfxEncoding.WriteBytesIntoTextFile. New gate --kernelcmd-roundtrip (Tools/KernelCommandRoundtripRt0) proves the EXACT GUI path (ReadList→Wrap→Unwrap→WriteList) byte-identical 4/4 (command/item/monmagic1/2) — which the AbilityCommandLab (direct path) structurally couldn't see. Wired into offline_ci → 26 gates PASS.
Workspace persistence: the loaded master folder is saved to %LocalAppData%\FFXProjectEditor\last-project.txt and auto-restored on startup (priority: CLI arg → last loaded → defaults). Load once, remembered forever; no more defaulting to the Steam-mod folder.
Global crash logger (%LocalAppData%\FFXProjectEditor\crash.log, via Utils/CrashLog) + defensive load in KernelCommands (explanatory red error instead of killing the app).
Aurora Chamber: monster anchors now show the NAME (🔴 <name>) instead of "monster (live)".
Doc-only ride-along: docs/governance/FFX_EDITOR_UI_CONVENTIONS.md (panel standard = Shop Explorer, not a 50-column DataGrid) + DINPUT8 rule revoked (probe is free) + multi-language/MagicViewer notes in the checklist. csproj bump (UI features).