docs: contract #480 equipped publication and pin Rynthid destination

docs/research/2026-09-01-overhaul/478-480-readonly-diagnosis.md

@@ -47,11 +47,27 @@ effective part replacements and cleanup; do not manually inject membership.
 Existing controller fixtures use empty Setups; registry fixtures inject
 handmade child shapes, and selection tests inject visible parts directly.
 
-Agent named-retail pointers (lead binary re-verification still owed before
-any AC-specific repair contract): CPhysicsObj::add_shadows_to_cells recursively
-visits children; CPartArray::AddPartsShadow visits every non-null visual part.
-Named pseudo-C lines282819–282901 and285933–285955. No new retail claim is
-accepted merely from a successful attachment log.
+Lead independently verified the named-retail loops and paired executable:
+CPhysicsObj::add_shadows_to_cells recursively visits children with the same
+CELLARRAY; CPartArray::AddPartsShadow visits every non-null visual part.
+Named pseudo-C lines282819–282901 and285933–285955. These establish expected
+child/part publication, not the modern prepared-cache mechanism itself.
+
+Paired `C:\Turbine\Asheron's Call\acclient.exe` SHA-256:
+`006FFEADC5D679C871497112A5BD1F87714D0E273E2166BAE5052DDE369297B1`.
+Lead mapped VAs using its PE image base/section table and hashed exact ranges:
+
+| Function | VA / file offset / length | SHA-256 |
+| --- | --- | --- |
+| CPhysicsObj::add_shadows_to_cells | 00514AE0 / 00114AE0 / 307 bytes | E651AE6CC5C09374356B4389D44C2F5615DAF50BD2CADFFE8C4565A41C078388 |
+| CPartArray::AddPartsShadow | 00517E40 / 00117E40 / 89 bytes | 63C6ACFF37CE039B6929EBD51A08889F6EB4904B53F29C77C0E162B19CE95C75 |
+
+Verified call bytes: 00514BA2 `E8 99 32 00 00` targets00517E40;
+00514BF7 `E8 E4 FE FF FF` recursively targets00514AE0. The part loop reads
+parts[i], skips null, and calls the virtual add-part slot via `FF 55 00`
+at00517E87 before increment/num_parts comparison. The pseudo-C's00517E76
+annotation identifies the expression, not that native call instruction.
+No new retail claim is accepted merely from a successful attachment log.
 
 ## #478 — distinguish local-light selection from High postprocessing
 

docs/research/2026-09-01-overhaul/479-rynthid-location-evidence.md

@@ -0,0 +1,50 @@
+# #479 — Rynthid identity, before a transform diagnosis
+
+2026-09-05, read-only lead investigation on production `de427d2c0`, docs
+HEAD710cbb554. No fix or visual acceptance. The owner's hanging-structure
+placement failure remains open; no owner PNG/exact affected part was captured.
+
+## Verified destination
+
+Rynthid Recall is spell6150/0x1806 (also present in the generated local
+SpellId enum). The primary ACE world-data row is
+[06150 Rynthid Recall.sql](https://github.com/ACEmulator/ACE-World-16PY-Patches/blob/master/Database/Patches/2%20SpellTableExtendedData/06150%20Rynthid%20Recall.sql).
+Root fetched its exact Git blob `fd301a1f944e2585a823fa7a17bcb35926786e2e`
+through the public GitHub API, without accessing server credentials or writing
+the database. The row specifies cell3032001C, XYZ(83.3,89.2,133),
+WXYZ(-0.642788,0,0,-0.766044). This is upstream ACE data, not a claim to have
+queried this running server's database.
+
+The owner's recorded reveal generation4 at3032001C,
+XYZ(83.3,89.2,132.00499), therefore DOES corroborate the Rynthid visit.
+The small Z difference is the observed settled pose, not evidence of the
+reported structure displacement. No owner heading/affected-object identity
+is recovered from this correlation alone.
+
+Historical `tools/connected-r6-soak.route.txt` labels3032001C "Sawato" and
+F6820033 "Rynthid". Those labels must NOT be used to identify the failed
+scene. Keep old result/checkpoint identifiers traceable; this note corrects
+their semantic use without rewriting historical evidence or changing route
+coordinates. A coordinate-validity check supersedes a copied label.
+
+## Installed-DAT candidates
+
+Root ran the existing Release A8CellAudit read-only `buildings 30320000 1`
+and `buildings 30320000 4` against Documents/Asheron's Call installed DATs.
+Before execution, both AcDream.App/acclient were absent; ACE UDP9000 PID13340
+was up. No graphical client was launched and no DAT was modified.
+
+- 3032 itself has35 static objects and0 buildings: the recall encampment.
+- The radius4 scan includes dense2D31 with535 objects and0 buildings; many
+  objects sit at Z205–225, including Setups02001BEE/02001BF0/02001BF2 and
+  Gfx01004D35–01004D3A. These are candidate elevated complex components,
+  NOT a proven match to the displaced part.
+- The radius4 scan totals129 shell entities and0 registry buildings (shells
+  without portal cells count separately). Tool output was too large for the
+  model result and truncated; no whole-scan numerical/geometry PASS is claimed.
+
+Wegener is doing a bounded read-only object/part/transform trace while #477
+is tested. Next proof needs exact candidate Setup/GfxObj/part transforms from
+the current static pipeline and the installed authored frames, then a stable
+connected view after GPU repair. Do not assert rebase, quaternion, animation,
+or DAT corruption as the cause without that evidence.

docs/research/2026-09-01-overhaul/480-equipped-render-publication-contract.md

@@ -0,0 +1,101 @@
+# #480 — publish equipped visual-part geometry before attachment
+
+Lead contract2026-09-05. NOT IMPLEMENTED. Independent of the in-flight #477
+native synchronization repair. The commit containing this contract is the
+exact scratch base; production remains `de427d2c0` until reviewed integration.
+
+## Evidence and objective
+
+The owner sees no wielded hand items. `478-480-readonly-diagnosis.md` records
+the lead-verified producer-to-consumer defect: the controller resolves GfxObj
+availability through DAT but builds child render parts from an unpopulated
+PhysicsDataCache. Cold parts can therefore attach with zero RetailPartEntry
+rows and never reach the world walk. The ordinary materializer already uses
+LiveCollisionAssetPublisher to publish the required prepared views/bounds.
+The attached, Position-null route bypasses that ordinary materializer.
+
+The note contains the lead's exact named-retail/paired-executable checks for
+CPhysicsObj::add_shadows_to_cells and CPartArray::AddPartsShadow. Retail
+recurses through attached children using the parent's CELLARRAY and publishes
+their non-null parts. Repair the missing modern producer edge; do not change
+those rules or infer success from an attachment log. The exact owner-item
+cache state was not captured, so a real cold-cache regression is required.
+
+## Scope and implementation contract
+
+- EquippedChildRenderController: reuse the existing strict live collision
+  publisher to publish each DAT-resolved effective GfxObj before building
+  the child's render-part membership and before materializing/announcing its
+  new projection. BuildPartAvailability already visits those actual objects;
+  use that path or an equally narrow shared visit, not a second decoder.
+- LivePresentationComposition: inject the SAME composition-owned publisher
+  used by ordinary live hydration. A required Action<uint,GfxObj> bound to
+  CacheGfxObj is acceptable to keep the public controller from exposing an
+  internal concrete type. No nullable/no-op production fallback or second
+  cache/source/publisher owner. Existing fixture constructor adaptations are
+  permitted in the two network-presentation suites and controller suite.
+- Focused tests in the existing equipped-child controller fixture, plus at
+  most one narrowly scoped test file if separating the composed witness is
+  clearer. Reuse current scene/walk fixtures and prepared-source test support.
+- Update the relevant architecture paragraph and this contract's outcome.
+  Every intentional deviation gets a register row in the same landing commit;
+  none is expected. Do not change unrelated register history.
+
+Preserve effective AnimPartChanged IDs, part ordering, absent-DAT behavior,
+Setup scale/holding-location/pose composition, exact entity/authority checks,
+parent CELLARRAY inheritance and all detach/reparent/withdrawal sequencing.
+Cache immutable asset data only; children still acquire NO independent
+collision registration or flood. Do not repair the ignored AttachChild
+Boolean by inventing a new rollback/lifetime protocol in this chunk.
+If an actual required failure-path change emerges, report it before expanding.
+
+Strict prepared-source failure must remain a truthful failure before a new
+child projection/membership is published. Never manufacture bounds or silently
+fall back to parsed collision, root-cell draw selection, parent mesh parts,
+or an all-entities sweep. No changes to shaders, GPU code, Runtime authority,
+lighting, capture timing, package format or streaming policy.
+
+## Discriminating proof
+
+1. Drive the REAL OnSpawn attached Position-null path with a nonempty child
+   Setup and real drawable GfxObj metadata, an initially empty PhysicsDataCache
+   and a recording IPreparedCollisionSource behind the REAL strict publisher.
+   Give the parent actual registered retail CELLARRAY membership and normal
+   materialized scene projection. Do not prepopulate child cache/rows manually.
+2. Assert the child relation/materialization as a control, then exact child
+   local-ID, part indices and effective GfxObj identities in every inherited
+   cell. Feed the same registry and actual resulting projection into
+   WalkProductionWorldData; assert its EquippedChild record. The unfixed
+   producer must fail at the missing child rows, not fixture/setup failure.
+3. Cover visual-only/no-physics-BSP geometry, a cold AnimPartChanged replacement
+   distinct from the original Setup ID, and repeat realization/reappearance
+   without duplicate rows or unnecessary prepared reads. Missing DAT parts
+   retain existing unavailable semantics. Missing prepared content must fail
+   before new projection/ready publication and must not publish graph-only data.
+4. Detach/withdraw the exercised child through the real controller lifecycle;
+   its rows disappear without deleting the parent's rows or registering child
+   collision shapes. Preserve existing withdrawal/reparent/network tests.
+5. Bind production composition to the actual shared publisher (existing
+   source/IL test idiom if needed); a fixture-only publisher call is not proof.
+
+Two actual-production mutation checks: remove the new publication call;
+then publish original Setup IDs instead of effective replacement IDs. The
+corresponding cold-cache/replacement assertions must fail, not compilation.
+Restore exact raw source hashes after each and finish Release/focused green.
+No new test framework, detached model oracle, increased tolerances or skips.
+
+## Workflow and closure
+
+One OpenAI implementer in a fresh codex/s5-480-openai-impl scratch. No commits,
+extra agents or graphical launch by the implementer. Check both AcDream.App
+and retail acclient before EVERY build/test; neither may be running. Other
+scratches are untouched. Root independently verifies the return and repeats
+discriminating evidence; sequential retail/behavior and production/lifetime
+review lenses follow, within the owner's ten-review limit.
+
+Return exact changed files/hashes, commands, green Release/focused TRX paths,
+mutation failures and raw restoration proof. Root commits/integrates and
+updates the ledger. Coordinate graphics only after #477 validation is clean:
+verify equipped hand objects in a connected client, preserve PNG paths and
+label visual PASS provisional until owner inspection. This chunk cannot
+close the overall failed owner gate or authorize a main merge.

docs/research/2026-09-01-overhaul/s5-consumers-material-closeout-packet.md

@@ -4672,3 +4672,25 @@ Dispatch: contract commit `29730506c57b44cfa3ba454a8594e170c95d7233`, fresh
 base verified clean. James implements only that contract; no agent graphical
 launches or extra agents. Lead owns verification/review/integration and
 unchanged production remains `de427d2c0` until accepted. No repair PASS yet.
+
+## 50. Parallel bounded held-item repair; Rynthid identity correction
+
+#477's three production dependency fixes are written in its scratch; James
+is compiling/testing them, not yet returned or accepted. No second graphical
+run until lead verification/review/integration. Both clients absent and
+ACEUDP9000 PID13340 up at the lead's latest read-only DAT execution.
+
+Lead independently re-verified #480's child/part retail loops against the
+paired executable; range hashes/call bytes are in478-480-readonly-diagnosis.md.
+`480-equipped-render-publication-contract.md` bounds the missing shared
+prepared-publication edge, real cold-cache OnSpawn-to-walk proof, replacement
+IDs, strict failure/withdrawal, and two actual-production mutations. It may
+proceed in an independent fresh scratch while GPU testing continues; no
+changes to GPU/lifetime policy, no fallback cell membership or new owner.
+Implementation/visual PASS remain owed. Lead owns commits/integration.
+
+Primary ACE spell6150 resolves to3032001C and matches the owner's generation4;
+old route labels "Sawato"/"Rynthid" are unreliable identifiers. Evidence and
+candidate elevated2D31 complex are in479-rynthid-location-evidence.md. A
+read-only object/part trace is in progress; transform cause still unproven.
+All #478/#481, #476, exact route, C1c/C2/full-G4 obligations in§49 remain open.