feat(chat): CT-B1 — bound the transcript by retail's character budget

Campaign CT slice B1.

CORRECTION TO THE PLAN: this slice was written as "the transcript grows for the
life of the session — a slow leak". That was wrong, and the plan said it
because I read the retail-side finding and inferred our side without checking.
ChatLog has always been bounded (ConcurrentQueue, maxEntries default 500, with
a dequeue loop in Append). There was no leak.

The real gap is the UNIT. Retail bounds the rendered transcript by CHARACTERS —
0x2710, beheaded toward 0x1D4C at a newline boundary — while we bounded the
model by messages. Two different things: a window of 500 messages is far more
scrollback than 10,000 characters, and the message cap is a safety limit on the
log rather than a display rule.

So the budget is applied where retail applies it: on the rendered window, not
the model. ChatLog's entry cap stays as the model-level bound.

Two deliberate simplifications, both registered as CT-1 rather than left
implicit:

  - ONE threshold, not retail's two. The hysteresis exists to stop retail
    re-trimming an accumulating buffer on every append; we rebuild the visible
    list each time, so there is nothing to damp, and a second threshold would
    only make the oldest visible line jump around as messages arrive.
  - Whole-line cutting rather than a newline search near an offset — our unit
    already IS the line, which is what retail's newline preference is for.

Filtered-out lines deliberately do not consume budget: a line this window
filters out is not in retail's buffer at all, so counting it would mean turning
a filter OFF silently shortened the visible history.

Solution builds clean; full hermetic gate green.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

docs/architecture/retail-divergence-register.md

@@ -490,6 +490,7 @@ equivalence argument (promote to AD/AP) or a fix.
 | UN-4 | GfxObj double-sided/negative-surface handling keeps WB's legacy logic (cull-mode double-siding, no reversed-winding duplicate, different neg-surface predicate) while the CellStruct path follows the retail-cited `ConstructMesh` reading | `src/AcDream.App/Rendering/Wb/ObjectMeshManager.cs:1059` (CellStruct contrast :1396-1410) | No recorded justification on the GfxObj side — it is the unmodified WB extraction; the retail citation was added only to the CellStruct path | GfxObj models retail draws via duplicated-reversed-winding get wrong back-face lighting (normals not inverted) or missing/extra negative faces — dark or absent faces from behind | `D3DPolyRender::ConstructMesh` 0x0059dfa0 |
 | UN-6 | Fixed 200 ms sleep between ConnectRequest and ConnectResponse; retail inserts no delay. Annotated only as "with 200ms race delay"; the 2026-06-04 audit flagged it, the follow-up refuted "forbidden workaround" but wrote no fuller rationale back | `src/AcDream.Core.Net/WorldSession.cs:484` | Presumed ACE port+1 listener race guard — four words, no citation | Every login eats a flat 200 ms; if the race needs longer on a loaded server, the handshake fails intermittently (ConnectResponse ignored → CharacterList never arrives, exit-29 shape) with no retry — a timing constant masking an unconfirmed root cause | (none recorded) |
 | UN-7 | Outdoor OBJECT point lighting uses `calc_point_light` (wrap/norm + per-channel cap, `~1/d²`) for ALL meshes including static buildings, but retail's object path is unconfirmed — `config_hardware_light` (0x0059ad30) sets D3D-FF point lights (`Diffuse=color×intensity`, `Attenuation=(0,1,0)`⇒`1/d`, `Range=falloff×1.5`, `material.diffuse=white`) yet that math would blow walls WHITE while retail stays DIM, so static buildings may instead use the `SetStaticLightingVertexColors` bake. Model + the brightness-scaling factor both UNRESOLVED (issue #140 / Fix D) | `src/AcDream.App/Rendering/Shaders/mesh_modern.vert` (`pointContribution`); `src/AcDream.Core/Lighting/LightManager.cs` (`SelectForObject`) | Fix A/B ported calc_point_light + per-object selection for objects without confirming retail uses that model for static buildings; cdb captured the D3D-FF path but it contradicts the observed dim result | Outdoor buildings blow out warm near torches (the #140 meeting-hall symptom); whichever model is wrong, the object torch contribution is too strong | `config_hardware_light` 0x0059ad30; `SetStaticLightingVertexColors` 0x0059cfe0; `rangeAdjust=1.5` 0x00820cc4 — see docs/research/2026-06-18-lighting-a7-fixABC-shipped-fixD-handoff.md |
+| CT-1 | Transcript truncation uses ONE character threshold (10,000) where retail uses two — it beheads to ~7,500 (`0x1D4C`) on passing 10,000 (`0x2710`), so its buffer oscillates between the two. acdream also cuts at whole LINES rather than searching for a newline near a byte offset | `src/AcDream.App/UI/Layout/ChatTranscriptRenderer.cs` (`MaxTranscriptCharacters`, `FirstLineWithinBudget`) | Retail's hysteresis exists to avoid re-trimming an ACCUMULATING buffer on every append; we rebuild the visible list from the log each time, so there is nothing to damp and a second threshold would only make the oldest visible line jump around as messages arrive. Whole-line cutting is what retail's newline preference is trying to achieve — our unit already is the line | acdream shows up to ~2,500 characters more scrollback than retail at the moment retail has just trimmed. Visible only as a slightly longer history; no state, wire or memory effect (ChatLog's own entry cap still bounds the model) | `ChatInterface::TruncateChatLog @0x004F4290`; threshold read at `RecvNotice_DisplayFinalStringInfo @0x004F4640` |
 
 ---