Beta Update

Earlier beta notes, preserved as originally published.

This is the archive of earlier beta notes. New notes are on the main updates page.

Update 2.61

6/9/2026

Native menu STEP 3: native-list row source reverse-engineered + `list-read` (probe) + live selection read. Continuing the native in-game menu lane (Yojimbo→Jarvis-BAHAMUT), step-3 gap #1 ("where do the native list's rows come from?") is closed via IDA: the list INPUT (`FFX_Menu_List_UpdateInput 0x8B4460`) is 100% generic — pure field math over the 152-byte object (`+40` state, `+48` count, `+50` top, `+52` target, `+58` page, `+66` slots, `+69` result, `+70` scroll, `+72` selected, `+28` validator), touches no global, and is reused by 10 distinct builders — but every native draw is item-shaped: the scene3 list (`FFX_Menu_CreateScene3ScrollableList 0x8B4FB0` → draw `0x8B4A00` → row `0x8B4B20`) reads rows from `unk_159EC30[64i]` (64-byte structs) + `byte_1866250[]` (type), with the item name resolved by `0x86C3C0(id→table)` and an optional direct label at `rowPtr+32` (type 1); it loads scene3/33/34 and depends on Customize globals. There is no "free-text list" draw in the binary — so a native list with OUR text (the arena / Aurora Photo-Mode shell) requires a DLL draw callback (step 5, the path foreseen in tick-bible §4), with the hand-rolled recipe documented (generic input `0x8B4460` + a DLL draw using `sub_8F5F70` window / `sub_9016B0` string / `sub_8C0640` cursor, reading `+69`/`+72`). Shipped via probe TODAY: new verb `ffxprobectl list-read [handleHex]` (READ-only) that scans the menu-object POOL (`g_FFX_MenuObjPool 0x18408C0`, 32 slots × 152B) — finds ANY active native list/popup and identifies it by its callbacks — plus the shortcut "current list" globals (`dword_1866214`/`186A5DC`/`23CC120`). ✅ RT2 PASS (live, 2026-06-09): with the game in the Customize equipment menu, the pool scan caught the navigated list (slot[3] `@0x017A0A88`, `input=0x8D57E0`, a thin wrapper of the generic `0x8B4460`) and `+72`/`SELECTED` tracked the cursor in real time (35→42) as the operator moved it → step-3 "read the selection of a native navigable list" PROVEN without writing anything. (Bonus: the reset object's `+62 = group 0x101` matches `FFX_MenuObj_Reset` (IDA) — offsets confirmed against memory.) No camera RAM / Aurora / `dllmain.cpp` / save touched; IDA read-only (`_claude_ida` copy, rename-queue in the doc). `ffxprobectl` build 0 errors. Step 5 (blueprint): paste-ready skeleton of the native shell at `RuntimeTools/NativeMenuShell/NativeMenuShell.h` (+README) — object hand-roll + draw callback (window+rows+cursor) + FFX font encoder + decoupled bridge to Aurora's Photo Mode actions; ABIs/offsets confirmed (IDA verify workflow) and adversarially reviewed (2 lenses FFX-vs-IDA + C++ → no crash/compile blocker); x86 guard + anti-OOB clamp + load-bearing invariants annotated; not wired, does not touch `dllmain.cpp`. Step 5.1 (wire blueprint, NOT applied): plan + patch draft (`docs/ai/HANDOFF_BAHAMUT_NATIVE_MENU_WIRE_BLUEPRINT_2026-06-09.md` + `docs/patches/BAHAMUT_NATIVE_MENU_WIRE_DRAFT_2026-06-09.patch`) wiring the shell into `ffx-hooks.dll` via a detour on the menu pump `FFX_Menu_PerFramePump 0x8A9C50` (`int __cdecl(uint)`, IDA-verified, main-thread) — OFF by default (`FFXHOOKS_ENABLE_NATIVE_MENU` + F7 hotkey), bridge only calls `PhotoMode::`; adversarially reviewed (2 lenses) no blocker (compile reviewer applied it to a scratch tree, `git apply --recount --check` exit 0); `dllmain.cpp` NOT touched/committed. Doc: `docs/reverse/FFX_NATIVE_MENU_LIST_ROW_SOURCE_2026-06-09.md`.

Read notes

SeymourBETA

Update 2.60

6/9/2026

updatemonster-ai

Reusable read-only Atlas evidence badges, applied to the Monster AI Editor. First layer of `HANDOFF_ATLAS_READONLY_MODULE_BADGES`: a reusable evidence-badge component that shows, per entry, where the data came from and how far it can be trusted — with no writer, no apply, no SIN. New `Modules/Common/AtlasEvidenceBadgeStrip.axaml(.cs)` (drop-in UserControl with an `Evidence` property) + `Modules/Common/AtlasEvidenceBadge.cs` (`AtlasEvidenceBadge` colored by severity + `AtlasEvidenceInfo` carrying badges/origin/"where it appears", factories `ForBibleEntry`/`ForDetail`). Key decision (after the adversarial review): ONE engine — `AiBibleEvidence` was parameterized into `Derive(...)` and the strip reuses the entry's canonical badges via `entry.EvidenceBadges`, so the Monster AI inline strip and the full BIBLE window can never disagree; `IDA` stays a separate structural badge and never collapses into a green `proved` (a bug the review caught and fixed). Added the mission's `partial` token. The "where it appears in game" text became `AiBibleWhereAppears` (one DRY source shared by BIBLE and the modules). Applied to the `Monster AI Editor` inline BIBLE panel: badges + origin + a "where it appears" expander for the selected entry; the existing 📖 Abrir guia completo button opens the focused BIBLE. No runtime, SIN, save-path, or `SpiraDataAtlasCatalog` provider changes. Build 0 errors / 361 warnings; `--spiraatlas-rt0` and `--aicmdmeta-rt0` PASS; `AiScriptLab --ai2` PASS (validator clean 346/346).

Read notes

SeymourBETA

Update 2.59

6/9/2026

updatemonster-ai

BIBLE OF SPIRA / Atlas UI 2.0: multi-badge evidence navigation + filters + provenance panel (read-only). The `BIBLE OF SPIRA` window gained an honest evidence-health navigation layer with no new writer. `AiBibleEntry` now carries structured Atlas provenance (`SourcePath`, `WriterPolicy`, `DetailKind`) plus a single source of evidence tokens (`AiBibleEvidence`) derived only from fields the entry already has — it never invents proof. Every entry shows multi-badges (`blocked`, `RT2`, `IDA`, `proved`, `RT0`, `parser-corpus`, `presence-index`, `metadata-only`, `semantic-candidate`, `RT2-pending`, `read-only`) colored by severity with honesty-preserving tooltips (`RT0` = Atlas parse/catalog integrity, not runtime authority; `parser-corpus` proves structure/presence, not effect). A new Evidence filter (`Todas`, `Proved/RT0`, `Parser corpus`, `Metadata-only`, `Blocked`, `RT2 pending`) cascades cleanly with the Kind/Domain filters and search, with counts. Atlas-derived entries get an Atlas: origin & policy panel exposing `Domain`, `DetailKind`, `WriterPolicy`, and `SourcePath` (without dumping raw CSV). The "where it appears in game" text was rewritten per domain with honest wording (`monster-presence` = formation/slot presence, not behaviour; `sin-eligibility` = a queue, not authorization; `gear-name-model` = the `w_name.bin` row, not the final reward name; `blitzball-prize` = metadata-only; Sphere `Unknown6` = raw/unknown). Search now accepts `metadata-only`/`blocked`/`RT2-pending`/`domain:` via tokens added to `SearchBlob`. Temp build 0 errors / 361 warnings; `--spiraatlas-rt0` and `--aicmdmeta-rt0` PASS (`979/867/112`). No runtime, SIN writer, or `SpiraDataAtlasCatalog` provider changes: read-only / UX only.

Read notes

SeymourBETA

Update 2.58

6/9/2026

updateruntime-hook

Aurora Overlay Lab in the editor: persistent config for `ffx-hooks.dll`. New `Live Tools -> Aurora Overlay Lab` module to control the runtime overlay through `modules\config\aurora_overlay.ini`, while syncing the legacy flag files (`aurora_overlay.flag`, `aurora_overlay_d3d11.flag`, `aurora_w2s_sniff.flag`) and exposing GDI/D3D11 mode, W2S scan, D3D sniff, budget/cooldown, refresh, and `Auto-pause hits` knobs. The DLL reads the `.ini` at boot, with env vars still taking priority; `d3d_sniff_autopause_hits=3` pauses heavy constant-buffer inspection while the D3D fallback matrix is still fresh. The panel normalizes game root / `FFX.exe` / `modules` input, shows DLL/config/log state, opens `%TEMP%\ffx-hooks.log`, launches through Steam appid `359870`, warns when the live `FFX.exe` root differs from the selected target, and catches config I/O failures instead of crashing the UI. Prior runtime proof: D3D11 in-frame labels in live FFX; this pass validated editor build `0 errors` / `361 warnings` and DLL deploy.

Read notes

SeymourBETA

Update 2.57

6/9/2026

updatemonster-ai

Custom Boss Creator: Advanced Loot light for Drop1/Steal/Bribe. The stats panel now has an Advanced Loot card with optional overrides for common/rare `Drop1`, common/rare `Steal`, and `Bribe`: item IDs, counts, and raw chances where those fields exist. Blank or invalid fields inherit from the selected `Loot used`; Preview / Diff resolves item names through `Item_Dictionary` and shows changes such as `Potion x1 → Dark Matter x2`, while listing invalid fields as inherited. The cloner now applies the copied/inherited whole `LootFile` and then stamps `Gil/AP` plus Drop1/Steal/Bribe light overrides on top; `.bossrecipe.json` records those overrides when filled. Honesty: `Drop2`, Overkill drops, and Gear/ability pools remain in Monster Editor / a later cut. Editor build: 0 errors / 361 warnings.

Read notes

SeymourBETA

Update 2.56

6/9/2026

updatemonster-ai

Monster AI Editor: field/stat planter + final Trinity/BIBLE polish. The MONSTER AI TRINITY header is now centered as a proper first-read callout for `AEON observes · BIBLE teaches · SIN creates`. The Workbench gained a Plant field/stat card in the fine-tuning area: it searches any `btlActorProperty` from `AiChrPropertyNames`, accepts decimal values or `0xNN`, and saves into the real `onTurn` hook with backup using the correct `PUSHII <field> · PUSHII <value> · CALLPOPA 70AB` shape (`setStatField(field,value)`, no `actorRef`, not confused with `writeChrProperty`). The default ordering promotes `OverdriveMax`, `OverdriveCurrent`, `showOverdriveBar`, and hit-reaction fields for Dark Shiva/overdrive-style edits. The inline BIBLE panel also got cleaner Spira/FFX colors for selection and detail panes instead of the clashing terminal block. Editor build: 0 errors / 361 warnings.

Read notes

SeymourBETA

Update 2.55

6/9/2026

updatemonster-ai

Custom Boss Creator: Boss Recipe import/export/round-trip in the UI. The left column now has a Recipe card with `Import` and `Export` buttons for `.bossrecipe.json`. Export writes the current draft recipe without creating an `m###.bin`; import validates schema `ffx.bossrecipe.v1`/version 1 and reapplies source monster, target ID, display name, AI mode/source, stats, and simple rewards (`RewardGil`, `RewardAp`, `RewardApOverkill`) to the screen fields. If the target already exists, import warns in the status panel and lets the user choose another ID before clicking `Create Monster`. Honesty: the recipe is offline editor metadata; materializing the boss still uses the create button, and placement/RT2 remain Formation Editor/Battle Sandbox/in-game work. Editor build: 0 errors / 361 warnings.

Read notes

SeymourBETA

Update 2.54

6/9/2026

updatearena

Battle Sandbox v2: up to 8 enemies, location/route picker, and spread/camera planner. The Battle Sandbox tab in `LiveBattleLab` is no longer a vertical A/B/C stack: it now has a compact 8-enemy slot layout with per-slot HP and quick AI. Routes can be selected from the decoded `btl.bin` list (`map/battleId/field/group/formation`) or edited manually; Prepare route writes the resolved `btl_` formation slots and `MonsterLive` positions with `.sandbox.btl.bak`, reusing `BattleArenaAuthor`, `Battle_File.WriteWithFormationSlots()`, and `BattleArenaPositionWriter`. The planner offers spreads (`2 compact rows`, wide, boss+adds, corridor, ring) and computes a camera framing recommendation for all active monsters; automatic camera writing remains honestly deferred to Aurora/RT2. `Force Battle` still requires the connected DINPUT8 probe. Solution build: 0 errors / 361 warnings; local screenshot at `work/screenshots/battle-sandbox-tab.png`.

Read notes