feat(agent-workspace): explain history tradeoffs and drift

Extend expanded current-atom history items with explicit primary-vs-
secondary tradeoff copy and adjacent-event drift explanation, keep the
surface bilingual, and refresh the progress docs so mainline now records
M8.34/M8.35 as shipped.

Author: Jacobinwwey

docs/diataxis/en/explanation/agent-conversation-focus-mode-plan.md

@@ -469,13 +469,14 @@ It is conversation-product hardening on top of the shipped shell.
 `M8.31` now makes those follow-up suggestions continuity-aware, so older history entries skip already-completed recent steps when a fresher deterministic action exists.
 `M8.32` now explains those freshness-aware suggestions and exposes one secondary alternative, so expanded history items read like bounded decision surfaces instead of opaque action picks.
 `M8.33` now layers confidence signals onto that rationale surface, so each expanded history item exposes rank/freshness/alternative-availability evidence instead of leaving suggestion certainty implicit.
+`M8.34` now explains why the primary follow-up stayed ahead of the surfaced secondary alternative, so expanded history items no longer leave the primary-vs-secondary choice implicit.
+`M8.35` now explains recommendation drift across adjacent history events, so the current-atom history surface no longer requires manual cross-card comparison for that change.
 
 Priority order:
 
-1. `M8.34`: explain why the primary follow-up beat the surfaced secondary alternative using existing ranking inputs only,
-2. `M8.35`: explain confidence drift when adjacent history events change the recommended next step,
-3. only then deeper personalization work,
-4. markdown-reader improvements only when they clearly compound the same study loop.
+1. deepen personalization on top of the shipped primary/secondary/drift explanation surface,
+2. expand broader operator rollups only when they clearly compound the same current-atom continuity loop,
+3. keep markdown-reader improvements secondary unless they clearly compound the same study loop.
 
 ## Related Pages
 

docs/diataxis/en/explanation/development-progress-dashboard.md

@@ -6,7 +6,7 @@ It explicitly separates branch-level implementation status from mainline integra
 ## Scope
 
 - Focus area: local-first knowledge mastery platform (ingest, retrieval, learning path, tutor, memory, governance).
-- Time window: `v1.7.0` through the 2026-04-21 foundation-integration checkpoint, the 2026-04-21 UTC CI review cycle, and the post-`M8.33` reconciliation pass.
+- Time window: `v1.7.0` through the 2026-04-21 foundation-integration checkpoint, the 2026-04-21 UTC CI review cycle, and the post-`M8.35` reconciliation pass.
 - Evidence rule: every progress claim must map to:
   - contract layer (`src/learning/api.ts`, `src/learning/types.ts`),
   - route/runtime layer (`src/server.ts`),
@@ -18,12 +18,12 @@ Current top priority is not feature-count expansion.
 Current top priority is **conversation-product hardening on top of the already-green backend baseline**:
 
 - keep backend sufficiency as a standing decision gate rather than an active execution lane,
-- keep closing the correction/product/operator gap exposed after `M8.20-M8.33`,
-- treat `M8.23` as the operator-visibility baseline for conversation-memory/correction actions, `M8.24` as the active-atom action-surface baseline, `M8.25` as the active-atom pane-state continuity baseline, `M8.26` as the active-atom study-loop summary baseline, `M8.27` as the active-atom recent-result continuity baseline, `M8.28` as the active-atom result-preview baseline, `M8.29` as the active-atom history-drill-down baseline, `M8.30` as the active-atom history follow-up baseline, `M8.31` as the active-atom history freshness-ranking baseline, `M8.32` as the active-atom history rationale/alternative baseline, and `M8.33` as the active-atom history confidence-signal baseline,
-- keep the remaining near-term budget on deeper continuity explanation quality around alternative tradeoffs and confidence drift instead of reopening backend scope,
+- keep closing the correction/product/operator gap exposed after `M8.20-M8.35`,
+- treat `M8.23` as the operator-visibility baseline for conversation-memory/correction actions, `M8.24` as the active-atom action-surface baseline, `M8.25` as the active-atom pane-state continuity baseline, `M8.26` as the active-atom study-loop summary baseline, `M8.27` as the active-atom recent-result continuity baseline, `M8.28` as the active-atom result-preview baseline, `M8.29` as the active-atom history-drill-down baseline, `M8.30` as the active-atom history follow-up baseline, `M8.31` as the active-atom history freshness-ranking baseline, `M8.32` as the active-atom history rationale/alternative baseline, `M8.33` as the active-atom history confidence-signal baseline, `M8.34` as the primary-vs-secondary tradeoff explanation baseline, and `M8.35` as the adjacent-history drift explanation baseline,
+- keep the remaining near-term budget on deeper personalization and operator rollup quality now that the continuity-explanation gap is closed on the active history surface,
 - keep remote `main` CI interpretation evidence-based: the latest fully completed `main` workflow set is green, and new pushes should be treated as pending until their workflow set finishes.
 
-## Current Mainline Snapshot (2026-04-21 Baseline Split: Foundation M8.18 / Conversation M8.33)
+## Current Mainline Snapshot (2026-04-21 Baseline Split: Foundation M8.18 / Conversation M8.35)
 
 - Foundation baseline on `main` is currently:
   - status `integrated`
@@ -37,15 +37,14 @@ Current top priority is **conversation-product hardening on top of the already-g
 - Historical increment sections below remain chronological archive entries.
 - When a historical section says `in_progress` / `no-go`, read it as past checkpoint truth rather than current repo state.
 
-## Reconciliation Notes (2026-04-21 Post-M8.33)
+## Reconciliation Notes (2026-04-21 Post-M8.35)
 
 - The active mainline question is no longer whether backend sufficiency work should continue.
   That lane is closed-by-default while its gate stays green.
-- The active mainline question is also no longer whether confidence signals should be added.
-  `M8.33` already shipped that slice on the expanded history decision surface.
-- The remaining explanation gap is narrower and code-shaped:
-  - the UI still does not explain why the primary follow-up beat the surfaced secondary alternative,
-  - the UI still does not explain recommendation drift across adjacent history events without manual comparison.
+- The active mainline question is also no longer whether tradeoff and drift explanation should be added.
+  `M8.34` and `M8.35` now ship that continuity-explanation layer on the expanded history decision surface.
+- The next gap is no longer local explanation opacity.
+  The remaining product question is how much deeper personalization and operator rollup value should sit on top of this now-readable current-atom history surface.
 - Remote `main` CI is currently green on the latest push.
   Current-facing docs must treat CI repair as dormant unless a fresh failing run exists.
 
@@ -236,6 +235,38 @@ Execution anchor:
   - diagnostics snapshot coverage for stored preview payloads,
   - EN/ZH re-render coverage for localized preview text.
 
+## Latest Mainline Increment (2026-04-21 M8.35 Active-Atom Adjacent History Drift Explanation)
+
+- Extended the expanded history decision surface with adjacent-event drift explanation:
+  - older current-atom history items now explain when a newer adjacent event changed the recommended next step,
+  - drift copy stays bounded to the already-available ranking inputs instead of introducing new scoring state,
+  - the rail now makes recommendation drift readable without forcing manual comparison across expanded cards.
+- Kept the slice frontend-bounded:
+  - no backend route, API contract, persistence model, or capability schema changed,
+  - drift resolution reuses the same bounded recent-history list plus the shipped follow-up descriptor logic.
+- Closed the second post-`M8.33` explanation gap:
+  - operators no longer have to infer adjacent-event recommendation changes by hand,
+  - the continuity surface now explains both confidence posture and recommendation drift in place.
+- Expanded regression coverage for:
+  - adjacent history items rendering explicit drift explanation when the recommended next step changes,
+  - EN/ZH rerender coverage for the new drift label plus tradeoff explanation copy.
+
+## Latest Mainline Increment (2026-04-21 M8.34 Active-Atom Primary-vs-Secondary Tradeoff Explanation)
+
+- Extended the expanded history follow-up surface with explicit primary-vs-secondary tradeoff explanation:
+  - each expanded current-atom history item now states why the primary follow-up stayed ahead of the surfaced secondary alternative,
+  - tradeoff copy stays bounded to existing ranking inputs: candidate order, freshness, and typed-vs-fallback availability,
+  - secondary alternatives remain actionable while the rail now explains why they were not promoted.
+- Kept the slice frontend-bounded:
+  - no backend route, API contract, persistence model, or capability schema changed,
+  - tradeoff resolution reuses the shipped follow-up candidate ordering plus the existing point capability/fallback model.
+- Closed the first post-`M8.33` explanation gap:
+  - operators no longer have to infer why one surfaced alternative remained secondary,
+  - the current-atom history card now explains the primary-vs-secondary decision before deeper personalization work.
+- Expanded regression coverage for:
+  - explicit tradeoff copy on expanded history items,
+  - EN/ZH rerender coverage for the new tradeoff label and explanation copy.
+
 ## Latest Mainline Increment (2026-04-21 M8.33 Active-Atom History Confidence Signals)
 
 - Extended the expanded history follow-up surface with explicit confidence evidence:
@@ -1374,11 +1405,10 @@ Note: foundation-lane gates now exist on `main`; keep them in the normal verific
 
 1. Treat backend sufficiency as closed-by-default while the gate remains green; do not reopen heavier backend work without new measured pressure.
 2. Treat remote `main` CI as evidence-bound to the latest fully completed workflow set; only reopen CI-triage narrative when a fresh failing run exists.
-3. Make the next planning round about post-`M8.33` bounded conversation-product hardening:
-   - `M8.34`: explain why the primary follow-up beat the surfaced secondary alternative using existing ranking inputs only,
-   - `M8.35`: explain confidence drift when adjacent history events change the recommended next step,
-   - only then deeper personalization work,
-   - markdown-reader improvements only when they compound the same study loop.
+3. Make the next planning round about post-`M8.35` bounded conversation-product compounding:
+   - deepen personalization on top of the now-readable primary/secondary/drift explanation surface,
+   - add broader operator rollup/reporting only when it compounds the same current-atom continuity loop,
+   - keep markdown-reader improvements secondary unless they strengthen the same study loop.
 4. Treat notification-threshold governance as closed after `M7.29`; do not plan `M7.30+` unless a materially new operator requirement appears.
 5. Keep markdown-reader governance bounded and secondary until the conversation-product lane is explicitly reprioritized.
 

docs/diataxis/en/explanation/local-backend-sufficiency-and-escalation-plan.md

@@ -248,7 +248,8 @@ Recommended next product order:
    - the minimum correction / feedback closure is now integrated through explicit next-step guidance on missing managed state,
    - operator-facing visibility for conversation-memory and correction actions is now integrated through persisted diagnostics summary/index counts,
    - stronger agent action surfaces around focus mode and learning path are now integrated through the active-atom quick-action rail, toolbar routing, continuity notes/path-target diagnostics, capability-taxonomy summary cards, the recent-activity history card, localized result-detail previews on that history surface, expandable history drill-down, deterministic history follow-up affordances, continuity-aware freshness ranking for those affordances, and bounded rationale/secondary-alternative rendering,
-   - next prioritize `M8.34` primary-vs-secondary tradeoff explanation and `M8.35` confidence-drift explanation on top of the shipped signal-enabled current-atom recent-activity card before deeper personalization work.
+   - `M8.34` primary-vs-secondary tradeoff explanation and `M8.35` adjacent-history drift explanation are now integrated on top of the shipped signal-enabled current-atom recent-activity card,
+   - next only deepen personalization or broader operator rollups when they clearly compound that same readable current-atom continuity loop.
 2. markdown-reader improvement:
    - only when it clearly compounds knowledge review or focus-mode study loops.
 

docs/diataxis/zh/explanation/agent-conversation-focus-mode-plan.md

@@ -457,13 +457,14 @@
 `M8.31` 现在又让这条 follow-up suggestion 具备 continuity-aware freshness：只要存在更新鲜的确定性动作，较旧 history item 就会跳过已经完成过的近期步骤。
 `M8.32` 现在又给这条 freshness-aware suggestion 补上原因解释与一个 secondary alternative，让展开后的 history item 更像有界决策面，而不是黑盒动作推荐。
 `M8.33` 现在又在这层 rationale surface 之上补上置信信号，让每条展开的 history item 直接暴露排序 / 新鲜度 / 备选可用性，而不是把建议可信度继续留给操作者自己猜。
+`M8.34` 现在又解释主建议为什么继续排在当前 secondary alternative 前面，让展开后的 history item 不再把主建议 vs 备选决策留成隐含信息。
+`M8.35` 现在又解释相邻 history event 之间的 recommendation drift，让当前 atom history surface 不再要求操作者手工对比多张展开卡片。
 
 优先顺序：
 
-1. `M8.34`：只用现有 ranking 输入解释主建议为什么赢过当前展示的 secondary alternative，
-2. `M8.35`：解释相邻 history event 之间推荐动作为何发生 confidence drift，
-3. 之后才进入更深的 personalization 工作，
-4. Markdown 阅读器提升只有在能明确复利同一学习闭环时才进入前列。
+1. 在当前已落地的 primary/secondary/drift explanation surface 之上继续加深 personalization，
+2. 只有当更宽的 operator rollup 能明确复利同一 current-atom continuity loop 时才继续扩展，
+3. Markdown 阅读器提升继续保持次优先级，除非它能明确加强同一学习闭环。
 
 ## 关联页面