Skip to content

[PR] fix(#5714): recognize shifted-but-preserved history in the tag-change continuation rung - #5727

Open
ranxianglei wants to merge 5 commits into
masterfrom
2026-11-05_shifted-persona-continuation
Open

ranxianglei wants to merge 5 commits into
masterfrom
2026-11-05_shifted-persona-continuation

Conversation

@ranxianglei

Copy link
Copy Markdown
Owner

What & why

DSH native main requests with a stable client-minted conversation id move onto an empty |sub: lane when the per-turn system/persona text rotates and a host-side insertion shifts the whole old history to new indices. The root lane's folds are orphaned (field report: a maintenance session landed on a sub lane with 425 inherited refs but 0 blocks, not the root's 32; a ~314s preflight before forwarding on a 700400-token budget).

This is the grown/shifted counterpart of two already-shipped rungs:

A host-side INSERTION/DELETION shifts every later conversation cell by one index, so #4778's same-index comparison scores zero agreement AND #5686's length gate (incoming < recorded) is false → the tag-change ladder falls through to derived-persona, despite the old content being intact.

Fix

In src/session-resolve.ts, the tag-change continuation rung now measures the longest EXACT content-cell fragment agreement over ANY relative offset instead of only offset zero (renamed maxAlignedFragRun → maxContinuationFragRun). The per-cell sha256 comparison is unchanged — MESSAGE-IDENTITY (never fuzzy similarity) — so a run ≥ CONTINUATION_MIN_RUN still cannot occur between two distinct conversations regardless of where the window sits. A qualifying run returns immediately, bounding the scan on long histories instead of paying the full O(n·m) offset sweep.

Behavior change (disclosed)

  • OLD: a rotated persona over a shifted-but-intact main history forked onto <id>|sub:<fp(tag)>.
  • NEW: it re-anchors onto the raw id (reanchor), preserving the root lane's fold state.
  • WHY: the recorded body is verbatim inside the incoming (only positionally shifted); forking threw away live compression state for no reason. All four guardrails (overlap / length-gate / mainShaped / threshold) are retained and pinned both directions, so stranger, stale-shorter, below-threshold and shallow-blob requests still fork.

Tests

tests/issue5714-shifted-persona-continuation.test.ts — the issue's deterministic topology (pre-fix it reproduces the exact |sub:8dfd76270c919cda fork fingerprint from the field report), plus both-direction controls (shifted stranger / stale-shorter / scattered-below-threshold / shallow blob all still fork; the shorter slice still reanchors through #5686 untouched). The full #4778 suite stays green (no regression). Typecheck clean.

Notes

  • Config surface: none.
  • No kernel / wire / persistence change.
  • Touches session-resolution (conversation-key binding) → routed for human review rather than auto-merge (§7.4 identity/session-binding line).

Fixes #5714

ework-agent and others added 2 commits November 5, 2026 14:36
…ual (#4599)

01fcc5d (#5656/#5584, merged via #5687) diverts tool-less dsh single-shot
requests onto the verbatim side lane (kernel no-touch); the residual for main
turn 1 is documented in the commit's own comment. Five pre-existing pins
assumed turn 1 rides the full pipeline and went born-red at merge (master red
since; every PR head inherits it).

Test-only realignment codifying the documented behavior:
- issue2419/2446/2447/2481 guard-stamp subtests: two turns; the BLOCKED ref
  is witnessed on the first FULL-PIPELINE turn (turn 2). No fold can happen
  before turn 2 (nothing compressible, no action surface), so protection
  against every possible fold is unchanged.
- issue2817: stats.requests off by one (turn 1 uncounted); the persona-fold
  teeth are intact, assertions shifted -1.
- issue5403 A/B golden: turn 0 rides the side lane in BOTH phases (the gate
  ignores the rollout env), so rb[0] is unstamped by design; stamps resume at
  turn 1. Byte equality holds at i=0 (both phases forward verbatim).

No src/, kernel/, config, or version changes. Locally green: all six files +
side-lane-invariants + issue5656-single-shot-sidecar + dsh-persona-fingerprint.
… continuation rung

A host-side insertion or deletion shifts every later conversation cell by one
index; the tag-changed ladder's #4778 continuation rung only compared cells at
the SAME index, so a rotated persona on a shifted-but-intact main history
scored zero agreement and fell through to derived-persona — orphaning the root
lane's folds onto an empty |sub: lane (the grown/shifted counterpart of #5686's
shorter-tail case).

Measure the longest EXACT content-cell fragment agreement over ANY relative
offset instead of just offset zero. The per-cell sha256 comparison is unchanged
(MESSAGE-IDENTITY: never fuzzy), so a run of CONTINUATION_MIN_RUN agreements
still cannot occur between distinct conversations regardless of where the window
sits — stranger / stale-shorter / below-threshold / shallow-blob forks all stay
intact (pinned both directions). Early-exit at the threshold bounds the scan on
long histories.

Fixes #5714
@ranxianglei ranxianglei changed the title fix(#5714): recognize shifted-but-preserved history in the tag-change continuation rung [PR] fix(#5714): recognize shifted-but-preserved history in the tag-change continuation rung Nov 5, 2026
@github-actions

github-actions Bot commented Nov 5, 2026 •

Copy link
Copy Markdown
Contributor

📦 Built Package Artifact

Branch: 2026-11-05_shifted-persona-continuation (17c40a2)

Latest build — install this exact version

npm install -g billion-context@0.0.8238-pr.5727.9eb9a55

This exact version is immutable and reproducible. The rolling pr tag tracks the newest test build across all PRs.

Recent builds (commit → version)

commit version build
17c40a2 billion-context@0.0.8238-pr.5727.9eb9a55 build
5ee7753 billion-context@0.0.8233-pr.5727.dfbebfe build
9d2a96a billion-context@0.0.8223-pr.5727.ff09d35 build

Download artifact

  1. Download the artifact from the Actions run
  2. Extract the tarball and install:
tar xzf billion-context-pr5727.tgz
npm install -g package

Install instructions are refreshed on each push.

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b)

查重: 不重复。 本 PR 针对 #5714( persona 文本轮换 + 宿主侧插入把整段旧历史推移到新索引 → tag-change 阶梯全部 rung 落空 → fork 到空的 |sub: 车道)。它引用的两个相邻机制都是已合入的正交 rung:#4778 = grown + 同索引改写(harness cell 清除),#5686 = 更短 + 尾部不变。本条的失败签名(索引移位使同索引对角线归零,同时打穿 #5686 的 shorter 长度门)与两者都不同。无并行在途 PR——镜像 [PR] 清单已查,针对 #5714 的只有这一个。

标签: bug, dsh, scope/wiring, p1 —— fork-继承被永久放弃类(owner 修订 2026-10-19 ①:root 车道 fold 状态孤儿化,现场为 sub 车道继承 425 refs 但 0 blocks)。

Review 进行中:diff 已通读(src/session-resolve.ts +45/-24 净,新回归测试 138 行)。本地正在跑定向套件(新回归文件 + #4778/#5686/#2944/#3397 persona 族 + typecheck),结构化八问结论随后在本线程给出。

@ranxianglei ranxianglei added bug Something isn't working dsh p1 scope/wiring labels Nov 5, 2026
…idecar residual broke five dsh regression pins
@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b)

结论:红 CI 不是本 PR 的代码问题,是继承性 master 红。已把 ci-fixes 哨兵分支合进 PR 并推送,新 run 已排队。

1. 是什么问题:PR #5727 的全部 10 个 test shard 全红,失败集中在同 4 个测试:#2446/#2447 的「dsh lane stamps the guard onto the effective config」子测试、#2817 cross-wire persona fold、#5418 watermark fastpath A/B。

2. 什么原因导致的:继承性失败,与本 PR 改动无关。#5656(single-shot sidecar,经 #5687 合入)之后,无工具的 dsh 单发请求走 verbatim side lane(turn 1 不再走完整管道),而 5 枚既有回归钉仍假设 turn 1 走全管道——合并那一刻就 born-red,master 从此红。证据链:①本地把 master 的 src/session-resolve.ts 换回(本 PR 唯一的源码改动),4 个测试以完全相同的断言值照样失败;②master 自身最近一次主 CI(schedule @0aace2f20)同样是 failure;③ci-fixes 哨兵上已有针对这 5 个文件的修复 commit(0470bec,test-only 钉位重对齐)。

3. 影响范围与严重程度:波及所有从当前 master 切出的 PR(全部继承红);对用户行为零影响——纯测试钉与已记录行为(#5656 commit 注释里自己写明的 residual)错位。

4. 怎么修复的:不动任何业务代码。在 PR 分支 merge origin/ci-fixes(merge commit 5ee7753),推送后本地定向验证:6 个被重对齐的文件 + 本 PR 自己的新测试(issue5714-shifted-persona-continuation)共 25 pass / 0 fail。已在 #4599 留了一行状态评论。

5. 是否有回归问题:无 src//kernel//配置变更。ci-fixes 的重对齐本身由哨兵验证过且收敛在 ci-fixes 单一出口(例:#2817 的 stats.requests 9→8 只因 turn 1 走 side lane 不计数,fork 断言的牙原样保留)。高危模式逐项过:无语义漂移、无 fail-fast 变化、无阈值/时序/输出格式变化。

6. 是否破坏缓存:不涉及。

7. 是否改了设计方向:不涉及。本 PR 本体(tag-change continuation rung 从同下标对齐放宽到任意偏移扫描)一行未动。

8. 是否引入了新的配置:不涉及。

新 CI run(42004717973 @ 5ee7753)已排队,e2e-codex-persona 已绿;主矩阵结果出来前本轮收工,若再红按 attempt 2/3 处理。

ci-check added 2 commits November 5, 2026 18:06
…ontinuations stop early

The offset sweep started at the most-negative s, so even the fixed scenario (host insertion, shift +k) paid ~n^2/2 comparisons before reaching its diagonal - measured 285ms @5k / 1.1s @10k / 4.4s @20k on exactly the sessions this rung serves (the tag rotates every turn there).

Visit s=0 first, then expand by |s|: unshifted (#4778) and small-shift (#5714) continuations return in O(n*(2k+1)) - measured <=1.6ms at all sizes above. The caller only tests against CONTINUATION_MIN_RUN, so visit order cannot change any verdict; a stranger with no qualifying run anywhere still pays the full exhaustive sweep (unchanged).
…tinuation sweep

The original suite only exercised positive offsets (host insertion). A host-side DELETION of an early cell shifts the recorded body to NEGATIVE relative offset - pre-fix it forked identically onto |sub:<fp>; now it reanchors. Same topology discipline as the insertion regression: distinct per-cell content, length gate satisfied, skeleton deep-equal assertion.
@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b)

Review 完成,建议可合并(CI 在最终 head 17c40a2c2 全绿)。我在审核中发现一个真实性能缺陷并已在 PR 分支上补修(两个标注 commit,见下),另补了一个缺失方向的回归测试。

结构化八问

  1. 是什么问题:DSH 主线会话(稳定的客户端铸造会话 id)里,宿主往自己的日志里插入/删除一个单元格后,整段旧历史索引整体平移;叠加每轮 system/persona 文本轮换,会话被误判成「另一个 persona」,分叉到空的 |sub: 车道——根车道的折叠状态全部孤儿化(现场:子车道 425 refs / 0 blocks vs 根车道 32 blocks,700K token 预算下转发前 ~314s preflight)。
  2. 什么原因导致的:tagChanged 阶梯第 4 级([bug][dsh] Goal auto-continuation forks the main conversation into an empty |sub: lane → ~6M-token preflight, session ne | [bug][dsh] Goal auto-continuation forks the main conversation into an empty |sub: lane → ~6M-token preflight, session never replies… #4778 continuation)只在「同索引对角线」上数逐格精确匹配——平移让同一索引的匹配全归零;而第 5 级(dsh Long-Session Drift Comprehensive Remediation: Sealing the Last Leak Window for Native Compression Local Deployment + | dsh 长会话漂移综合治理:堵死原生压缩本地落地的最后漏窗 + 会话身份分叉可见化与自愈 #5686)的长度门要求 incoming 更短,平移场景为假 → 穿透到 derived-persona。旧内容其实逐字完整躺在 incoming 里,只是位置变了。
  3. 影响范围与严重程度:p1 档(fork 继承被永久放弃类,按 2026-10-19 口径①)。触发条件:稳定客户端 id + 每轮 system 轮换 + 宿主侧单元格增删(DSH native 车道典型形态,即本 issue 现场)。后果:会话仍可用但压缩状态清零、preflight 与上下文膨胀(现场 314s)。
  4. 怎么修复的:continuation 级改为在所有相对偏移上度量最长「逐格精确内容」匹配段(每格 sha256 比较不变,永不模糊——MESSAGE-IDENTITY 教义),≥8 连续即返回。四个护栏(重叠/长度门/mainShaped/阈值)全部保留并双向钉住:Test Blind Spot Retrospective: Reinforcing Negative Assertion Discipline, Tightening Fake Upstream Strictness, Thin Inst | 测试盲区复盘:补反向断言纪律、假上游严格化、薄安装冒烟、形态矩阵与生命周期覆盖 #3617 纪律,8 个测试 = 3 回归 + 5 控制(stranger/stale-shorter/below-threshold/shallow-blob 照旧分叉,dsh Long-Session Drift Comprehensive Remediation: Sealing the Last Leak Window for Native Compression Local Deployment + | dsh 长会话漂移综合治理:堵死原生压缩本地落地的最后漏窗 + 会话身份分叉可见化与自愈 #5686 短尾照旧 reanchor)。
  5. 是否有回归问题:逐项清点过既有行为——[bug][dsh] Goal auto-continuation forks the main conversation into an empty |sub: lane → ~6M-token preflight, session ne | [bug][dsh] Goal auto-continuation forks the main conversation into an empty |sub: lane → ~6M-token preflight, session never replies… #4778 同索引续接套件绿、dsh Long-Session Drift Comprehensive Remediation: Sealing the Last Leak Window for Native Compression Local Deployment + | dsh 长会话漂移综合治理:堵死原生压缩本地落地的最后漏窗 + 会话身份分叉可见化与自愈 #5686 短尾控制绿、四个分叉控制绿、persona 家族全套绿。审核中新发现并已修:原实现从最负偏移开始扫,连被修好的场景(shift=+1/+2)也要先付 n²/2 次比较才摸到自己的对角线——实测 285ms@5k / 1.1s@10k / 4.4s@20k(合成语料),而这类会话恰恰每轮都打这一级,直接吃掉修复想省下的延迟。已补 commit 9f8377eab:访问顺序改为 s=0 起步、按 |s| 外扩——实测 shift+0/+2 降到 ≤1.6ms(34 个数量级);判定严格不变(caller 只用 ≥8 布尔值,访问顺序不改变结果集,全部测试重跑一致)。残留边界:stranger 情形(任何偏移都凑不出匹配段)仍需全量扫完才能证明「不是同一会话」,成本不变(0.69s @5k~20k,作者版本同样如此)——这是穷举证明的固有代价;该路径按设计罕见(长史 mainShaped 异会话复用 id),若要再优化需哈希索引收窄候选对角线,属更大算法改动,此处不动,留作后续选项。另补 commit 17c40a2c2:原测试只覆盖正偏移(插入),负偏移(删除平移)半区无钉,已补回归测试(pre-fix 红、post-fix 绿,拓扑纪律与插入版一致)。
  6. 是否破坏缓存:不涉及——未触碰缓存/重试/增量游标/去重键;skeleton 持久化字段布局(Compression fails in long sessions while new sessions work fine | 压缩失败.新会话反而没事 #3397 order-persist)未变。
  7. 是否改了设计方向:无转向——tagChanged 阶梯结构不动,只是第 4 级的「同一会话」证据从零对角线扩到任意偏移。行为变更披露(PR 正文已写,我核实准确):OLD 轮换 persona + 完好但平移的主历史 → fork <id>|sub:<fp(tag)>;NEW → reanchor 回 raw id,保住根车道折叠状态。
  8. 是否引入了新的配置:不涉及。

过程记录

判定:可合并(合并由人执行,我不合 PR)。两个补 commit 均为标注独立的评审补充(perf / test),不改变 PR 的目的与范围。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b)

本轮结论:未发现代码缺陷,未改代码;唯一红 shard 已触发重跑(HTTP 201 受理)。

诊断过程(证据链):

  • 本轮仅一个红:test (ubuntu-latest, 22, 2);其余 9 个 test shard + 全部静态门禁绿。
  • 红的测试:tests/msgid-lane-flow.test.ts "pin carries client ids into rows"——turn-2 resync 后断言账本应有 4 行(旧 gen 保留、编辑落新行),实际读到 3 行(= turn-2 之前的快照)。
  • 与本 PR 改动无关,四条证据:
    1. PR diff 只动 session-resolve 的 tag-change continuation rung(persona 标签变更场景);该测试走 dsh plugin lane + 固定 conversation id(conv-msgid-1),全程无 persona/标签变化,根本不会走到那条判定路径;
    2. 同一 commit 同批 CI,ubuntu-24-2 shard 跑了同样这组 msgid-lane-flow 测试,6/6 全绿;
    3. 当前 head (17c40a2) 本地压测 12 次,12/12 绿;
    4. 该测试有同族竞态前科——其 settle() 注释自述 [PR] fix: surface failed watcher registration cause/state and report heartbeat recovery #5644 win22-s2 red(fixed sleep 与 debounced save fsync 赛跑,后改为等 afterNextFlush)。
  • 判定:CI 机器高负载下的存储 settle 时序竞态,非确定性复现 = flake。疑似点(未深挖,留作记录):src/persist.ts:3064 afterNextFlush 在「无 pending」时立即返回 true,而测试读的是统一影子库 index.db 的行——影子 ingest 可能落后于 JSON envelope save 的 settle 一拍;生产上下一轮 save 会补齐,不影响数据正确性。按自动修复车道规则不另开 issue,在此留给 owner 定夺。
  • 已在 CI 总控 CI Control Center: Centralized CI Failure Handling & Status Sync (Comment all CI issues here; do not open new issues) | CI 总控: CI 失败集中处理与状态同步(所有 CI 问题在此评论, 勿开新 issue) #4599 留一行协调记录(comment 6657044685)。

八问:

  1. 是什么问题:唯一红是 msgid-lane-flow 测试的间歇性竞态(回合结束后立刻读账本,偶发读到上一拍快照),不是功能缺陷。
  2. 什么原因导致的:负载下「save 已 settle」信号可能先于统一影子库对应行落盘返回,测试拿到信号即断言,踩中竞态窗口;非本 PR diff 引入(证据见上)。
  3. 影响范围与严重程度:仅该测试在 CI 高负载下的稳定性;生产行为零影响(下一轮 save 自动补齐);严重度 = CI 噪音。
  4. 怎么修复的:不改代码(代码没错)——通过 API 重跑失败 job;若重跑仍红则竞态窗口坐实,需专项调查(建议方向:afterNextFlush 的 verdict 应覆盖影子 ingest 落行,或测试的 settle 增加有界重试)。
  5. 是否有回归问题:无——本轮零代码改动;上轮 ci-fixes 合并的效果已由本批 9/10 shard 绿 + 本地定向 25/25 验证。
  6. 是否破坏缓存:不涉及。
  7. 是否改了设计方向:不涉及。
  8. 是否引入了新的配置:不涉及。

收工判定:等待重跑结果;若转绿,本 PR 即可进入人工合并。

@ranxianglei

Copy link
Copy Markdown
Owner Author

Review verdict: APPROVE (posted as a comment because GitHub forbids approving your own pull request — the daemon account is the PR author; a human click of Approve is all that remains.)

(ai: vllm-qwen/qwen3.8-27b)

APPROVE — independent review of the full diff vs master 0aace2f20 (behavior A/B on local worktrees + CI forensics below). No blocking findings; nothing for me to push — the one small finding I had (no negative-offset pin) landed concurrently as 17c40a2c2 while this review was in flight; I verified it is the same pin I drafted independently, and that an equivalent test goes RED on unpatched master (teeth confirmed).

Structured verdict (八问)

  1. 是什么问题: DSH native main requests under a stable client-minted conversation id move onto an empty |sub: persona lane when the per-turn system/persona text rotates AND a host-side cell insertion/deletion shifts the whole old history by one index — orphaning the root lane's live fold state (field: sub lane with 425 inherited refs but 0 blocks vs root's 32; ~314s preflight on a 700K-token budget).
  2. 什么原因导致的: On tag-change the continuation rung ([bug][dsh] Goal auto-continuation forks the main conversation into an empty |sub: lane → ~6M-token preflight, session ne | [bug][dsh] Goal auto-continuation forks the main conversation into an empty |sub: lane → ~6M-token preflight, session never replies… #4778) measured exact content-cell agreement ONLY along the offset-zero diagonal; any insertion/deletion shifts every later cell off that diagonal → zero agreements, while the dsh Long-Session Drift Comprehensive Remediation: Sealing the Last Leak Window for Native Compression Local Deployment + | dsh 长会话漂移综合治理:堵死原生压缩本地落地的最后漏窗 + 会话身份分叉可见化与自愈 #5686 shorter-tail rung's length gate (incoming < recorded) is false for grown history → fall-through to derived-persona fork despite the old content being verbatim-present, only positionally shifted.
  3. 影响范围与严重程度: dsh plugin-native lane only; sessions where the host rotates persona text AND inserts/deletes cells mid-history (grown shape). Usable-but-degraded: compression state silently orphaned onto an empty fork lane (fold loss, preflight blowup); no data loss, no wire corruption. p2-class.
  4. 怎么修复的: The continuation rung now sweeps EVERY relative offset and takes the longest EXACT per-cell sha256 run (maxAlignedFragRun → maxContinuationFragRun); per-cell bar unchanged (kind is embedded in the fragment hash, so two distinct conversations sharing ≥8 verbatim cells stays cryptographically impossible regardless of window position); early-exit at CONTINUATION_MIN_RUN; all four guardrails (overlap / length / mainShaped / threshold) untouched. Follow-up perf commit visits offsets outward from s=0 — verdict-preserving (verified below) and removes the n²/2 cost the original ordering charged exactly the shifted sessions this rung serves.
  5. 是否有回归问题: Audited the whole tagChanged ladder rung-by-rung (head-align / model-switch / chain / this rung / dsh Long-Session Drift Comprehensive Remediation: Sealing the Last Leak Window for Native Compression Local Deployment + | dsh 长会话漂移综合治理:堵死原生压缩本地落地的最后漏窗 + 会话身份分叉可见化与自愈 #5686 / fallback): rung 4 vs 5 length gates are mutually exclusive (≥ vs < recorded) so neither steals the other's traffic; stranger / stale-shorter / below-threshold / shallow-blob still fork — pinned both directions (Test Blind Spot Retrospective: Reinforcing Negative Assertion Discipline, Tightening Fake Upstream Strictness, Thin Inst | 测试盲区复盘:补反向断言纪律、假上游严格化、薄安装冒烟、形态矩阵与生命周期覆盖 #3617); [bug][dsh] Goal auto-continuation forks the main conversation into an empty |sub: lane → ~6M-token preflight, session ne | [bug][dsh] Goal auto-continuation forks the main conversation into an empty |sub: lane → ~6M-token preflight, session never replies… #4778 + dsh Long-Session Drift Comprehensive Remediation: Sealing the Last Leak Window for Native Compression Local Deployment + | dsh 长会话漂移综合治理:堵死原生压缩本地落地的最后漏窗 + 会话身份分叉可见化与自愈 #5686-tagchanged + 会话身份:顺序+自铸id 为认领主判,hash 降级为防漂移护栏(通用规则,取代 persona-fp 阶梯) #2944 suites green on the final head (local, 32/32). Perf commit: the single caller does a pure threshold comparison (src/session-resolve.ts:365), so visit order cannot change any verdict; offset coverage is provably identical (d-loop spans max(sMax, −sMin), out-of-range offsets skipped); worst-case stranger cost unchanged, honestly disclosed in the commit message. Verdict: no unintended regressions.
  6. 是否破坏缓存: No — session-key resolution only; fold state is PRESERVED (that is the point: reanchor instead of fork). No cache/retry/cursor/dedup-key surface touched.
  7. 是否改了设计方向: No — extends the existing [bug][dsh] Goal auto-continuation forks the main conversation into an empty |sub: lane → ~6M-token preflight, session ne | [bug][dsh] Goal auto-continuation forks the main conversation into an empty |sub: lane → ~6M-token preflight, session never replies… #4778/dsh Long-Session Drift Comprehensive Remediation: Sealing the Last Leak Window for Native Compression Local Deployment + | dsh 长会话漂移综合治理:堵死原生压缩本地落地的最后漏窗 + 会话身份分叉可见化与自愈 #5686 rung family with the missing offset dimension; MESSAGE-IDENTITY doctrine intact (exact per-cell sha256, never fuzzy; order-primary).
  8. 是否引入了新的配置: None.

Behavior change disclosure (旧值→新值)

OLD: rotated persona over shifted-but-intact main history → fork <id>|sub:<fp(tag)> (empty lane, root folds orphaned). NEW: reanchor onto the raw id, preserving root-lane fold state. WHY: the recorded body is verbatim inside the incoming history (only positionally shifted); forking threw away live compression state for no reason. Disclosed in the PR body ✓.

Independent evidence

Verdict: APPROVE — code correct, tests have teeth, guardrails pinned both directions, behavior change disclosed, inherited CI red unblocked.

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b) PR #5727 review 完成 → APPROVE(以评论形式发布:GitHub 禁止批准自己账号开的 PR,形式上只差 owner 手动点 Approve)。独立审了最终 head 的全 diff(fix + perf + deletion-pin + ci-fixes absorb 四个 commit)+ 行为 A/B(master 上正例钉会红、最终 head 32/32 绿);镜像缓存里的「绿」实为 skip-scoped run 遮蔽了原 push-run 的 10/11 shard 红——经查为继承性 master 红(同签名 4 测试,#5656 sidecar residual),已由 ci-fixes 哨兵吸收解锁(新 head 17c40a2)。无 kernel/、无新配置、行为变更(旧 fork→新 reanchor)已在 PR 正文披露。Review: #5727 (comment)

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b)

CI 侧收工确认:head 17c40a2c2 全部 30 项 check 绿/跳过,无阻塞项。

  • 上轮判定为 flake 的 test (ubuntu-latest, 22, 2) 重跑已通过(msgid-lane-flow 竞态未复现)——flake 诊断坐实,非 PR 缺陷;
  • 其余 9 个 test shard、static-gates、kernel-guard、lane-guard、docs-sink 等全绿;skipped 均为常规门控的 real-client e2e 车道;
  • 我这边无待推 commit(上轮零代码改动,ci-fixes 吸收已在分支内)。

剩余动作只有一项:人工 Approve + Merge(agent 不合并)。PR: #5727

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

bug Something isn't working dsh p1 scope/wiring

Projects

None yet

Development

Successfully merging this pull request may close these issues.

[Bug][DSH] Grown shifted history plus persona rotation still forks root lanes after #4778/#5686

1 participant