|
| 1 | +--- |
| 2 | +date: 2026-04-21 |
| 3 | +topic: mainline-conversation-product-hardening-after-m8-21 |
| 4 | +--- |
| 5 | + |
| 6 | +# Mainline Conversation-Product Hardening After M8.21 Requirements |
| 7 | + |
| 8 | +## Problem Frame |
| 9 | + |
| 10 | +`main` no longer has an unresolved backend-readiness question. |
| 11 | +It also no longer has a failing remote-`main` CI story that should dictate direction by itself. |
| 12 | + |
| 13 | +As of `3a1926a`, the current `main` reality is: |
| 14 | + |
| 15 | +- backend sufficiency governance is integrated, |
| 16 | +- managed-memory inspection and managed-memory state visibility are integrated, |
| 17 | +- remote `main` CI is green on the latest push, |
| 18 | +- the remaining gap is not "can the platform run locally?" but "which product-facing conversation lane should absorb the next bounded budget?" |
| 19 | + |
| 20 | +The stale part of the current document chain is now different: |
| 21 | + |
| 22 | +- `docs/brainstorms/2026-04-21-mainline-foundation-integration-and-post-checkpoint-direction-requirements.md` correctly shifted the lane away from backend replay, but it stopped before naming the concrete post-`M8.21` product slice, |
| 23 | +- `docs/diataxis/*/explanation/development-progress-dashboard.md` still describes the current top priority as backend sufficiency governance closure even though that lane already landed, |
| 24 | +- `docs/diataxis/*/explanation/agent-conversation-focus-mode-plan.md` still treats deeper memory work as something that comes after focus/path pane work, while the repo already ships focus/path coexistence and has moved into bounded managed-memory execution and visibility. |
| 25 | + |
| 26 | +This document supersedes the earlier 2026-04-21 direction note as the current source-of-truth item for post-`M8.21` execution priority. |
| 27 | + |
| 28 | +## Verified Mainline Reality (2026-04-21) |
| 29 | + |
| 30 | +- `main` currently points at `3a1926a`. |
| 31 | +- Latest remote `main` CI runs for that commit are green: |
| 32 | + - `Migration Gates` |
| 33 | + - `Fixrisk Operational Readiness` |
| 34 | + - `Mobile E2E Detox Contracts` |
| 35 | + - `Docs Diataxis Site` |
| 36 | + - `Docs GitHub Pages Publish` |
| 37 | +- The local-first baseline remains integrated: |
| 38 | + - foundation status `integrated` |
| 39 | + - backend sufficiency decision `go` |
| 40 | + - sqlite graph persistence on `main` |
| 41 | + - bounded local ANN retrieval on `main` |
| 42 | +- The agent conversation lane already includes: |
| 43 | + - focus-mode and learning-path actions, |
| 44 | + - tutor actions, |
| 45 | + - session/history/quality inspection, |
| 46 | + - session/unit/long-term memory inspection, |
| 47 | + - managed note/correction write, |
| 48 | + - managed-key eviction, |
| 49 | + - deterministic managed-memory state visibility with present/missing key reporting. |
| 50 | +- The current local verification baseline still passes: |
| 51 | + - `npm run test:agent-workspace:contracts` |
| 52 | + - `npm test -- src/learning/KnowledgeLearningPlatform.test.ts --runInBand --testNamePattern "agent conversation returns local knowledge points with executable capabilities|tutor actions and memory policy APIs are operational"` |
| 53 | + - `npm run verify:agent-workspace:runtime` |
| 54 | + - `npm run docs:diataxis:check` |
| 55 | + - `npm run docs:site:build` |
| 56 | + |
| 57 | +## Deep Comparison: Prior Direction vs Current Code |
| 58 | + |
| 59 | +| Prior source | Earlier direction | Current code reality | Judgment | Consequence | |
| 60 | +| --- | --- | --- | --- | --- | |
| 61 | +| `docs/brainstorms/2026-04-21-mainline-foundation-integration-and-post-checkpoint-direction-requirements.md` | stop replaying backend-independence work; decide whether backend sufficiency is enough | backend sufficiency gate landed and remote `main` CI is green; the remaining question is now which conversation-product slice gets the next budget | direction still broadly correct but now incomplete | preserve as checkpoint-turn document, replace as current top item | |
| 62 | +| `docs/diataxis/en/explanation/development-progress-dashboard.md` + ZH counterpart | current focus is backend sufficiency governance closure | that closure already landed through `M8.19-M8.21` | stale current-focus framing | update top focus and next-direction sections | |
| 63 | +| `docs/diataxis/en/explanation/agent-conversation-focus-mode-plan.md` + ZH counterpart | deeper memory/personalization work comes only after focus/path/fullscreen contract work | focus/path/fullscreen coexistence is already shipped; bounded managed-memory execution and visibility also shipped | stale priority ordering | update current program priority to post-`M8.21` reality | |
| 64 | +| `docs/diataxis/en/explanation/local-backend-sufficiency-and-escalation-plan.md` + ZH counterpart | next lane should be conversation-product hardening, but still at a broad category level | current repo now proves the first two managed-memory increments; next step needs narrower product-unit selection | correct but too broad | refine next-lane wording to concrete bounded units | |
| 65 | + |
| 66 | +## Critical Findings |
| 67 | + |
| 68 | +1. Backend sufficiency governance is no longer the active execution lane. |
| 69 | + It is now a standing gate and interpretation rule. |
| 70 | + |
| 71 | +2. Remote `main` CI is not currently the blocker. |
| 72 | + The latest `main` runs are green, so "fix remote `main` CI" should not be used as cover for keeping old checkpoint language alive. |
| 73 | + |
| 74 | +3. The next lane should stay inside conversation-product hardening, but it must narrow from category-level wording to bounded product units. |
| 75 | + The most defensible order now is: |
| 76 | + - correction-flow closure, |
| 77 | + - operator-facing visibility for conversation-memory actions, |
| 78 | + - stronger focus/path action surfaces, |
| 79 | + - only then markdown-reader work if it clearly compounds learning loops. |
| 80 | + |
| 81 | +4. The repo now has a partial correction loop, not a complete one. |
| 82 | + It can: |
| 83 | + - record notes, |
| 84 | + - record corrections, |
| 85 | + - inspect managed-state presence, |
| 86 | + - evict deterministic managed keys. |
| 87 | + |
| 88 | + It still does not fully close: |
| 89 | + - operator-facing reporting of correction-state presence/absence over time, |
| 90 | + - product-level guidance for when missing correction state should trigger explicit next actions, |
| 91 | + - stronger contextual action surfacing around focus mode and learning path after conversation resolution. |
| 92 | + |
| 93 | +## Requirements |
| 94 | + |
| 95 | +**Truth Refresh** |
| 96 | +- R1. Add a fresh post-`M8.21` requirements record instead of silently stretching the earlier 2026-04-21 checkpoint document beyond its scope. |
| 97 | +- R2. `docs/brainstorms/index.md` must list this document as the first current item. |
| 98 | +- R3. EN/ZH development dashboards must point to this document first in their execution anchors. |
| 99 | +- R4. EN/ZH dashboard top-level current-focus text must describe conversation-product hardening as the active lane, not backend sufficiency closure. |
| 100 | +- R5. EN/ZH dashboard next-direction text must record that latest remote `main` CI is green and therefore not an active execution blocker. |
| 101 | + |
| 102 | +**Next-Lane Narrowing** |
| 103 | +- R6. The current product lane must be narrowed to three bounded units rather than a generic "improve conversation product" bucket: |
| 104 | + - correction-flow closure, |
| 105 | + - operator-facing visibility for conversation-memory and correction actions, |
| 106 | + - stronger focus-mode / learning-path action surfaces. |
| 107 | +- R7. Markdown-reader work must remain secondary until those three units are explicitly reprioritized away. |
| 108 | +- R8. No new direction text may imply that backend escalation is still the default lane while the sufficiency gate remains green. |
| 109 | + |
| 110 | +**Focus-Mode Plan Refresh** |
| 111 | +- R9. EN/ZH focus-mode delivery pages must stop presenting deeper memory work as strictly post-focus/post-path future work. |
| 112 | +- R10. EN/ZH focus-mode delivery pages must reflect that focus/path coexistence is integrated and that the current remaining work is around correction/product/operator hardening on top of that surface. |
| 113 | + |
| 114 | +**CI Interpretation** |
| 115 | +- R11. Current-facing docs must state the present CI truth: latest remote `main` workflows for `3a1926a` are green. |
| 116 | +- R12. Historical CI hardening notes should remain archived, but they must not be read as the current blocker unless a new failing run exists. |
| 117 | + |
| 118 | +## Success Criteria |
| 119 | + |
| 120 | +- a new 2026-04-21 post-`M8.21` requirements document exists, |
| 121 | +- `docs/brainstorms/index.md` lists it first, |
| 122 | +- EN/ZH dashboards use it as the first execution anchor and update their current-priority wording, |
| 123 | +- EN/ZH focus-mode and backend-sufficiency pages are refreshed to current product-lane reality, |
| 124 | +- local docs verification passes, |
| 125 | +- `main` is updated, |
| 126 | +- the worktree is clean after push. |
| 127 | + |
| 128 | +## Scope Boundaries |
| 129 | + |
| 130 | +- This document does not reopen backend-escalation implementation. |
| 131 | +- This document does not claim markdown-reader work is canceled forever. |
| 132 | +- This document does not add a new memory platform. |
| 133 | +- This document does not claim operator diagnostics already have full correction-history visibility. |
| 134 | +- This document does not reopen `M7.30+` notification-threshold work. |
| 135 | + |
| 136 | +## Key Decisions |
| 137 | + |
| 138 | +- **Preserve checkpoint history, replace current direction**: older 2026-04-20 and earlier 2026-04-21 docs remain historical evidence, not present-tense execution truth. |
| 139 | +- **Treat green remote `main` CI as verified current state**: document the present fact instead of continuing to write as if CI is still red. |
| 140 | +- **Narrow the product lane explicitly**: post-`M8.21` work should compete inside a bounded conversation-product list, not inside an open-ended "improve product" category. |
| 141 | +- **Keep backend pressure evidence-based**: no heavier backend work unless the sufficiency gate turns with measured evidence. |
| 142 | + |
| 143 | +## Outstanding Questions |
| 144 | + |
| 145 | +### Deferred to Planning |
| 146 | + |
| 147 | +- What is the smallest user-visible correction-flow closure that compounds the newly landed managed-memory state visibility? |
| 148 | +- What operator-facing report should become the authoritative correction/memory action surface: existing agent-workspace diagnostics, a dedicated memory-governance slice, or both? |
| 149 | +- Which focus/path action-surface improvement has the highest leverage after correction/operator visibility work: stronger contextual actions on the current atom, richer result cards, or tighter pane-state diagnostics? |
| 150 | + |
| 151 | +## Recommended Next Planning Order |
| 152 | + |
| 153 | +1. `M8.22` correction-flow closure on top of managed-memory state visibility. |
| 154 | +2. `M8.23` operator-facing visibility for conversation-memory and correction actions. |
| 155 | +3. `M8.24` stronger focus-mode / learning-path action surfaces tied to the active conversation atom. |
| 156 | +4. Re-evaluate markdown-reader scope only after those units are bounded or displaced by new evidence. |
| 157 | + |
| 158 | +## Next Steps |
| 159 | + |
| 160 | +-> `/ce:plan` for `post-M8.21 conversation-product hardening`, using the three bounded units above as the only first-pass candidates. |
0 commit comments