SeymourBETA

Update 2.65

6/10/2026

v2.65.0BETA

Seymour

PC/Player Growth in the Spira Data Atlas (read-only, `proved-candidate`, byte-grounded, work/-independent)

Plugs the REAL source of PC growth/stats into the provider: new SpiraDataAtlasDetailKind.PlayerGrowthStat + value-guarded accessor SpiraDataAtlasCatalog.TryGetPlayerGrowthStat(source, characterIndex, field, currentRawValue). Source = ply_save.bin (base/current stats) + ply_rom.bin (growth ROM: AP-requirement curve ApReq A/B/C/Max + per-stat auto-growth coefficients *CoefA/B), the SAME tables the PlayerGrowthEditor edits (parser PlayerKernel_File, RT0-proved by --player-rt0). The 20 slots (0–6 PC, 7 guest=Seymour, 8–17 Aeons, 18–19 unmapped) come from a byte-grounded compiled table (FfxLib/SpiraDataAtlas/PlayerGrowthData.cs, generated by tools/ffxdataparser_bridge/build_player_growth_index.ps1 -EmitCSharp from the .bin files), so they exist in ANY build WITHOUT work/ (unlike PcAeonFineStat, which depends on work/pc_save_stats_catalog.csv and is only a snapshot). Labels resolve at runtime via Character_Enum (single source of truth). Honesty (non-obvious): the stat-growth coefficients are Aeon-only — PC/guest slots carry ZERO coefficients (PCs grow stats via the Sphere Grid); the only PC growth knob in this ROM is the AP-requirement curve. This is a DIFFERENT axis from BRASKA's sum_grow.bin (auto-level ≠ customization recipe). REGION-SPECIFIC: ply_save/ply_rom are NOT locale-invariant (the new_uspc data section differs from jppc/inpc; compiled table = new_uspc, the region the editor loads). The in-game AP-per-level / per-coefficient gain formula is RT2-pending (bytes/fields proved, runtime arithmetic not reversed). Value-guard = same value-space the editor edits → editing the field (raw mismatch), or an unknown slot/source/field → accessor false → badge hides (mirror of Aeon/Shop/Mix). Gate --spiraatlas-rt0 21296 → 21316 (+20; all unique) + 18 new asserts (count 20/8/10/2; PC coef zero; Aeon coef non-zero; region uspc; lookup Tidus BaseHp 520 / ply_rom AP max 22000 / Valefor HpCoef 6; value-guard 9999 hides; out-of-range slot/source/field miss; domain search; BIBLE domain player-growth). Read-only: no writer/UI/runtime/SIN; does not touch WriteSave/WriteRom, the PlayerGrowthEditor UI, ANIMA's Common engine (consumed only), or BRASKA/AURON/JECHT/RIKKU/WAKKA/VALEFOR/AURORA/SIN. Build 0 errors / 361 warnings; --player-rt0 + --spiraatlas-rt0 + --aicmdmeta-rt0 PASS. Docs: docs/ai/ATLAS_TIDUS_PC_GROWTH_PROVIDER_RESULT_2026-06-10.md + scout docs/ai/ATLAS_TIDUS_PC_GROWTH_SCOUT_RESULT_2026-06-10.md

v2.64.3

v2.65.1BETAPATCH

Seymour

Atlas evidence badge in the PlayerGrowthEditor + honest Sphere None/Passive labels (read-only, value-guarded, hide-on-edit)

Atlas UI Sweep. (1) PlayerGrowthEditor: new <common:AtlasEvidenceBadgeStrip> in the "Growth Curves" card (ply_rom.bin) consuming the already-built accessor SpiraDataAtlasCatalog.TryGetPlayerGrowthStat(source, characterIndex, field, currentRawValue) (v2.65.0, TIDUS). New AtlasEvidenceInfo? SelectedCharacterEvidence in PlayerGrowthEditor_DataModel, recomputed on slot change and in CharacterRowChanged (edit) → hide-on-edit: the badge anchors on 3 gate-proven growth values spanning both files/cards (ply_save Base HP + ply_rom AP max + HP coef A — the same lookups --spiraatlas-rt0 proves: 520 / 22000 / 6); if any anchor diverges from the byte-grounded corpus the strip self-hides (same anchoring depth as the Mix v2.64.1/Aeon v2.64.3 badges). (2) Sphere labels (honest, partial): sphere.bin ActionValue=0x0000 now reads "None / Passive" and RangeValue=0x00 reads "None" instead of "Unknown" — 0x00 is the third legitimate corpus value (~40% of spheres have no on-touch effect, Jarvis-AURON scout), not unknown; a "None" option is seeded into the dropdown + the fallback label fixed. (3) AiBibleWhereAppears.For: read-only player-growth and mix cases (honest "where it appears" text, no runtime-effect promotion; sphere-grid-node already existed). (REUSE of existing capability: UI consuming the already-built accessor + ANIMA's Common engine + label polish; direct precedent: the Mix v2.64.1/Aeon v2.64.3/Shop v2.61.2 badges were PATCH; tie-breaker doubt→PATCH). No new writer: the SphereGridExplorer badge already shipped (v2.60.1); the player-growth writer/save path and sphere topology are untouched. Does not touch the provider/gate (--spiraatlas-rt0 intact), ANIMA's Common engine (consumed only), runtime/Aurora/native-menu/SIN, or TIDUS/BRASKA/JECHT/RIKKU/WAKKA/VALEFOR/AURORA. Build 0 errors / 361 warnings; --spiraatlas-rt0 + --player-rt0 + --spheregrid-layout-rt0 + --aicmdmeta-rt0 PASS. Doc: docs/ai/ATLAS_LULU_PLAYER_GROWTH_SPHERE_BADGE_UI_RESULT_2026-06-10.md

v2.65.0

v2.65.2BETAPATCH

Seymour

Sphere Grid: empty-node RedundantContent preserved (0xFFFF) + Atlas where-appears polish

Atlas Polish / Evidence Housekeeping (Jarvis-BARTHELLO). (A) Sphere Grid hardening (byte fidelity): in SphereGrid_File.Builder.cs (the v2 LAYOUT writer, gated by --spheregrid-layout-rt0), the AUTHORING/EDIT paths AddNode/NodeWith wrote RedundantContent = (ushort)(contentIndex & 0xFF) → for an empty node (content 0xFF) that produced 0x00FF, diverging from shipped data. The corpus (Jarvis-AURON scout, 3,444/3,444 nodes: low byte == ContentIndex, high byte set only on the 23 empty Expert nodes) proves an empty node = 0xFFFF, a filled node = 0x00<content>. New private helper RedundantContentFor(contentIndex) mirrors the shipped convention (empty→0xFFFF, filled→0x00NN). FromExisting/Clone still copies RedundantContent verbatim → the no-edit round-trip stays byte-identical (untouched, re-proved by the gate). New assert in --spheregrid-edit-rt0 (synthetic + the 3 corpus grids): SetNodeContent(idx, 0xFF) → layout writes 0xFFFF at node+0x06 + contents byte 0xFF + re-reads empty. Gameplay risk low (redundant field; the game reads content from dat09/10/11 + reach-lists — IDA-proved by AURON on both binaries), but it is a fidelity divergence in an already-shipped writer. (B) AiBibleWhereAppears.For polish (read-only): added the aeon-growth case (an Atlas domain closed in v2.64.0 that was falling through to the empty fallback) with honest "where it appears / mapping" text; updated sphere-grid-node (was stale: "Unknown6 stays raw/unknown") to reflect AURON's burn-down — Node.Unknown6 is gameplay-inert (IDA-proved on both binaries), the possible residual = menu layout/visual (metadata-only), with no invented gameplay semantics. (fixes a latent fidelity bug in an already-shipped writer — "a consumer would benefit" → yes, byte fidelity — plus read-only text polish; tie-breaker doubt→PATCH). No new writer, no new UI, no provider/gate-count change (--spiraatlas-rt0 intact), no Common engine (ANIMA), runtime/AURORA/VALEFOR/native-menu/SIN, CHAPPU/save, or real grid topology (create/move/relink node stays a frontier) touched. Build 0 errors / 361 warnings; --spheregrid-edit-rt0 + --spheregrid-layout-rt0 + --spiraatlas-rt0 + --aicmdmeta-rt0 PASS. Doc: docs/ai/ATLAS_BARTHELLO_POLISH_RESULT_2026-06-10.md

v2.65.1

v2.65.3BETAPATCH

Seymour

SEYMOUR audit: the Atlas's 124 ATEL `unknown-call-shape` rows resolved into honest classes (read-only, byte-proven)

ATEL Call-Shape Audit (pre-SIN). The 124 field-writes the v8 semantic pass could not classify (field-write-unknown-call-shape) were audited at the bytecode level over the whole corpus (FFXDataParser target/text, 346 scripts) and decompose entirely into TWO proven classes, 0 remaining. (1) 0x70A8 btlSetMotionData — 116 rows (100%): uniform shape PUSHII actor · PUSHII field · (PUSHII|PUSHF) value · CALLPOPA = (actor, field, value) into the motionProperty namespace (fields 0x00..0x09, all named in AiMotionPropertyNames); the actor is a REAL ref (113× Self 0xFFF3 + Monster#01/#02/#03 0x15/16/17h), across 25 monsters → class set-motion-data-actor-field-value (the explicit-actor sibling of 0x70B2 setMotionField; NOT setStatField). (2) 0x7032 setActorFacingAngle — 8 rows: NOT a field write — it is (actor, angle), a 2-push call with no field-id; they leaked into the field catalog as a regex false-positive (only FLOAT angles render .facingAngle = 90.0 [42B40000h] and the v7 regex matched the .0 [hex] of the decimal; the 63 integer-angle sibling sites = 90 [5Ah] did not leak) → class not-field-write. Changes: the bridge tools/ffxdataparser_bridge/build_atel_semantic_catalog.ps1 (Get-FieldWriteSemanticKind + risk notes) reclassifies both call-ids; v8 regenerates with 0 unknown (116+8 in the new classes); the AtelFieldWriteShape count stays 202 (the relabel preserves the callId|fieldHex|fieldName|semanticKind|fieldCategory key cardinality, proven before/after); the unknown-health pass drops 199 → 75 (the 124 ATEL leave the "blocked-until-audited" bucket); the SpiraDataAtlasCatalog narratives (atlas:field-call-shapes + guardrail + atlas:unknown-health) are updated; the --spiraatlas-rt0 asserts move (detail entries 21316→21192, Unknown health rows 199→75) + 2 new lock-asserts (0 remaining unknown call-shapes + 0x70A8 motion-data present). Honesty: classifying ≠ authorizing a writer — every AtelFieldWriteShape stays read-only;not-writer-authority, 70A8 still has no button/write, nothing is promoted to SIN. Read-only fidelity finding: the EDITOR decoder (AiScript_File.cs) does not gloss 70A8 (absent from FieldArgBack/FieldNameForFunc) and AiStackModel lacks its arity → a display gap handed off to the MAGIC lane (docs/ai/HANDOFF_SEYMOUR_ATEL_DECODER_GLOSS_GAP_2026-06-10.md), with no gameplay effect. PATCH (refines/classifies an existing read-only catalog, resolves a SIN data-blocker; precedent v2.61.3/v2.62.1; tie-breaker doubt→PATCH). No new writer; does not touch runtime/AURORA/VALEFOR/native-menu/SIN-chain-builder/CHAPPU/save. Verified by an adversarial workflow (3 refuters + completeness critic). Build 0 errors / 361 warnings; --spiraatlas-rt0 + --aicmdmeta-rt0 + AiScriptLab --ai2 (346/346) PASS. Doc: docs/ai/ATLAS_SEYMOUR_ATEL_CALL_SHAPE_AUDIT_RESULT_2026-06-10.md

v2.65.2

Back to updates

SeymourBETA

Update 2.65

6/10/2026

v2.65.0BETA

Seymour

PC/Player Growth in the Spira Data Atlas (read-only, `proved-candidate`, byte-grounded, work/-independent)

Plugs the REAL source of PC growth/stats into the provider: new SpiraDataAtlasDetailKind.PlayerGrowthStat + value-guarded accessor SpiraDataAtlasCatalog.TryGetPlayerGrowthStat(source, characterIndex, field, currentRawValue). Source = ply_save.bin (base/current stats) + ply_rom.bin (growth ROM: AP-requirement curve ApReq A/B/C/Max + per-stat auto-growth coefficients *CoefA/B), the SAME tables the PlayerGrowthEditor edits (parser PlayerKernel_File, RT0-proved by --player-rt0). The 20 slots (0–6 PC, 7 guest=Seymour, 8–17 Aeons, 18–19 unmapped) come from a byte-grounded compiled table (FfxLib/SpiraDataAtlas/PlayerGrowthData.cs, generated by tools/ffxdataparser_bridge/build_player_growth_index.ps1 -EmitCSharp from the .bin files), so they exist in ANY build WITHOUT work/ (unlike PcAeonFineStat, which depends on work/pc_save_stats_catalog.csv and is only a snapshot). Labels resolve at runtime via Character_Enum (single source of truth). Honesty (non-obvious): the stat-growth coefficients are Aeon-only — PC/guest slots carry ZERO coefficients (PCs grow stats via the Sphere Grid); the only PC growth knob in this ROM is the AP-requirement curve. This is a DIFFERENT axis from BRASKA's sum_grow.bin (auto-level ≠ customization recipe). REGION-SPECIFIC: ply_save/ply_rom are NOT locale-invariant (the new_uspc data section differs from jppc/inpc; compiled table = new_uspc, the region the editor loads). The in-game AP-per-level / per-coefficient gain formula is RT2-pending (bytes/fields proved, runtime arithmetic not reversed). Value-guard = same value-space the editor edits → editing the field (raw mismatch), or an unknown slot/source/field → accessor false → badge hides (mirror of Aeon/Shop/Mix). Gate --spiraatlas-rt0 21296 → 21316 (+20; all unique) + 18 new asserts (count 20/8/10/2; PC coef zero; Aeon coef non-zero; region uspc; lookup Tidus BaseHp 520 / ply_rom AP max 22000 / Valefor HpCoef 6; value-guard 9999 hides; out-of-range slot/source/field miss; domain search; BIBLE domain player-growth). Read-only: no writer/UI/runtime/SIN; does not touch WriteSave/WriteRom, the PlayerGrowthEditor UI, ANIMA's Common engine (consumed only), or BRASKA/AURON/JECHT/RIKKU/WAKKA/VALEFOR/AURORA/SIN. Build 0 errors / 361 warnings; --player-rt0 + --spiraatlas-rt0 + --aicmdmeta-rt0 PASS. Docs: docs/ai/ATLAS_TIDUS_PC_GROWTH_PROVIDER_RESULT_2026-06-10.md + scout docs/ai/ATLAS_TIDUS_PC_GROWTH_SCOUT_RESULT_2026-06-10.md

v2.64.3

v2.65.1BETAPATCH

Seymour

Atlas evidence badge in the PlayerGrowthEditor + honest Sphere None/Passive labels (read-only, value-guarded, hide-on-edit)

Atlas UI Sweep. (1) PlayerGrowthEditor: new <common:AtlasEvidenceBadgeStrip> in the "Growth Curves" card (ply_rom.bin) consuming the already-built accessor SpiraDataAtlasCatalog.TryGetPlayerGrowthStat(source, characterIndex, field, currentRawValue) (v2.65.0, TIDUS). New AtlasEvidenceInfo? SelectedCharacterEvidence in PlayerGrowthEditor_DataModel, recomputed on slot change and in CharacterRowChanged (edit) → hide-on-edit: the badge anchors on 3 gate-proven growth values spanning both files/cards (ply_save Base HP + ply_rom AP max + HP coef A — the same lookups --spiraatlas-rt0 proves: 520 / 22000 / 6); if any anchor diverges from the byte-grounded corpus the strip self-hides (same anchoring depth as the Mix v2.64.1/Aeon v2.64.3 badges). (2) Sphere labels (honest, partial): sphere.bin ActionValue=0x0000 now reads "None / Passive" and RangeValue=0x00 reads "None" instead of "Unknown" — 0x00 is the third legitimate corpus value (~40% of spheres have no on-touch effect, Jarvis-AURON scout), not unknown; a "None" option is seeded into the dropdown + the fallback label fixed. (3) AiBibleWhereAppears.For: read-only player-growth and mix cases (honest "where it appears" text, no runtime-effect promotion; sphere-grid-node already existed). (REUSE of existing capability: UI consuming the already-built accessor + ANIMA's Common engine + label polish; direct precedent: the Mix v2.64.1/Aeon v2.64.3/Shop v2.61.2 badges were PATCH; tie-breaker doubt→PATCH). No new writer: the SphereGridExplorer badge already shipped (v2.60.1); the player-growth writer/save path and sphere topology are untouched. Does not touch the provider/gate (--spiraatlas-rt0 intact), ANIMA's Common engine (consumed only), runtime/Aurora/native-menu/SIN, or TIDUS/BRASKA/JECHT/RIKKU/WAKKA/VALEFOR/AURORA. Build 0 errors / 361 warnings; --spiraatlas-rt0 + --player-rt0 + --spheregrid-layout-rt0 + --aicmdmeta-rt0 PASS. Doc: docs/ai/ATLAS_LULU_PLAYER_GROWTH_SPHERE_BADGE_UI_RESULT_2026-06-10.md

v2.65.0

v2.65.2BETAPATCH

Seymour

Sphere Grid: empty-node RedundantContent preserved (0xFFFF) + Atlas where-appears polish

Atlas Polish / Evidence Housekeeping (Jarvis-BARTHELLO). (A) Sphere Grid hardening (byte fidelity): in SphereGrid_File.Builder.cs (the v2 LAYOUT writer, gated by --spheregrid-layout-rt0), the AUTHORING/EDIT paths AddNode/NodeWith wrote RedundantContent = (ushort)(contentIndex & 0xFF) → for an empty node (content 0xFF) that produced 0x00FF, diverging from shipped data. The corpus (Jarvis-AURON scout, 3,444/3,444 nodes: low byte == ContentIndex, high byte set only on the 23 empty Expert nodes) proves an empty node = 0xFFFF, a filled node = 0x00<content>. New private helper RedundantContentFor(contentIndex) mirrors the shipped convention (empty→0xFFFF, filled→0x00NN). FromExisting/Clone still copies RedundantContent verbatim → the no-edit round-trip stays byte-identical (untouched, re-proved by the gate). New assert in --spheregrid-edit-rt0 (synthetic + the 3 corpus grids): SetNodeContent(idx, 0xFF) → layout writes 0xFFFF at node+0x06 + contents byte 0xFF + re-reads empty. Gameplay risk low (redundant field; the game reads content from dat09/10/11 + reach-lists — IDA-proved by AURON on both binaries), but it is a fidelity divergence in an already-shipped writer. (B) AiBibleWhereAppears.For polish (read-only): added the aeon-growth case (an Atlas domain closed in v2.64.0 that was falling through to the empty fallback) with honest "where it appears / mapping" text; updated sphere-grid-node (was stale: "Unknown6 stays raw/unknown") to reflect AURON's burn-down — Node.Unknown6 is gameplay-inert (IDA-proved on both binaries), the possible residual = menu layout/visual (metadata-only), with no invented gameplay semantics. (fixes a latent fidelity bug in an already-shipped writer — "a consumer would benefit" → yes, byte fidelity — plus read-only text polish; tie-breaker doubt→PATCH). No new writer, no new UI, no provider/gate-count change (--spiraatlas-rt0 intact), no Common engine (ANIMA), runtime/AURORA/VALEFOR/native-menu/SIN, CHAPPU/save, or real grid topology (create/move/relink node stays a frontier) touched. Build 0 errors / 361 warnings; --spheregrid-edit-rt0 + --spheregrid-layout-rt0 + --spiraatlas-rt0 + --aicmdmeta-rt0 PASS. Doc: docs/ai/ATLAS_BARTHELLO_POLISH_RESULT_2026-06-10.md

v2.65.1

v2.65.3BETAPATCH

Seymour

SEYMOUR audit: the Atlas's 124 ATEL `unknown-call-shape` rows resolved into honest classes (read-only, byte-proven)

ATEL Call-Shape Audit (pre-SIN). The 124 field-writes the v8 semantic pass could not classify (field-write-unknown-call-shape) were audited at the bytecode level over the whole corpus (FFXDataParser target/text, 346 scripts) and decompose entirely into TWO proven classes, 0 remaining. (1) 0x70A8 btlSetMotionData — 116 rows (100%): uniform shape PUSHII actor · PUSHII field · (PUSHII|PUSHF) value · CALLPOPA = (actor, field, value) into the motionProperty namespace (fields 0x00..0x09, all named in AiMotionPropertyNames); the actor is a REAL ref (113× Self 0xFFF3 + Monster#01/#02/#03 0x15/16/17h), across 25 monsters → class set-motion-data-actor-field-value (the explicit-actor sibling of 0x70B2 setMotionField; NOT setStatField). (2) 0x7032 setActorFacingAngle — 8 rows: NOT a field write — it is (actor, angle), a 2-push call with no field-id; they leaked into the field catalog as a regex false-positive (only FLOAT angles render .facingAngle = 90.0 [42B40000h] and the v7 regex matched the .0 [hex] of the decimal; the 63 integer-angle sibling sites = 90 [5Ah] did not leak) → class not-field-write. Changes: the bridge tools/ffxdataparser_bridge/build_atel_semantic_catalog.ps1 (Get-FieldWriteSemanticKind + risk notes) reclassifies both call-ids; v8 regenerates with 0 unknown (116+8 in the new classes); the AtelFieldWriteShape count stays 202 (the relabel preserves the callId|fieldHex|fieldName|semanticKind|fieldCategory key cardinality, proven before/after); the unknown-health pass drops 199 → 75 (the 124 ATEL leave the "blocked-until-audited" bucket); the SpiraDataAtlasCatalog narratives (atlas:field-call-shapes + guardrail + atlas:unknown-health) are updated; the --spiraatlas-rt0 asserts move (detail entries 21316→21192, Unknown health rows 199→75) + 2 new lock-asserts (0 remaining unknown call-shapes + 0x70A8 motion-data present). Honesty: classifying ≠ authorizing a writer — every AtelFieldWriteShape stays read-only;not-writer-authority, 70A8 still has no button/write, nothing is promoted to SIN. Read-only fidelity finding: the EDITOR decoder (AiScript_File.cs) does not gloss 70A8 (absent from FieldArgBack/FieldNameForFunc) and AiStackModel lacks its arity → a display gap handed off to the MAGIC lane (docs/ai/HANDOFF_SEYMOUR_ATEL_DECODER_GLOSS_GAP_2026-06-10.md), with no gameplay effect. PATCH (refines/classifies an existing read-only catalog, resolves a SIN data-blocker; precedent v2.61.3/v2.62.1; tie-breaker doubt→PATCH). No new writer; does not touch runtime/AURORA/VALEFOR/native-menu/SIN-chain-builder/CHAPPU/save. Verified by an adversarial workflow (3 refuters + completeness critic). Build 0 errors / 361 warnings; --spiraatlas-rt0 + --aicmdmeta-rt0 + AiScriptLab --ai2 (346/346) PASS. Doc: docs/ai/ATLAS_SEYMOUR_ATEL_CALL_SHAPE_AUDIT_RESULT_2026-06-10.md

v2.65.2