Skip to content

[PR] kernel: compress length-rejection advice must match the failure direction (#6730) - #6751

Open
ranxianglei wants to merge 5 commits into
masterfrom
2026-11-14_compress-length-remediation-direction
Open

ranxianglei wants to merge 5 commits into
masterfrom
2026-11-14_compress-length-remediation-direction

Conversation

@ranxianglei

Copy link
Copy Markdown
Owner

What

kernel-change: fix a self-contradictory error-message literal (mechanical bug, owner policy 2026-11-06 exemption) — no algorithmic / state / wire-shape change, only the remediation text of the batch length-rejection.

When a compress batch is rejected for a summary-length failure, the kernel returned ONE generic remediation string for every direction:

Fix it one way or the other and resend: shorten or split each offending range's summary to fit its cap, then resend the whole batch; or drop the offending range(s)...

That advice is correct for a too-long summary but exactly backwards for a too-short one (the reporter's case, Summary too short (NN chars, min 50)). Following it made the summary shorter, so every retry failed again — a guaranteed-fail loop that re-read 300K+ tokens of context each round until the circuit breaker fired.

Fix

validateSummaryLength now throws a typed SummaryLengthError carrying the direction (empty / tooShort / tooLong). The batch screen tracks which directions fired and renders direction-aware remediation:

  • too-short / empty → expand each too-short/empty summary until it reaches its minimum — write out the key information the range actually contains (do NOT shorten it)
  • too-long → shorten or split each too-long summary to fit its cap (unchanged)
  • mixed → both, so each offending range gets the right advice
  • the drop the offending range(s) exit is always offered

The per-range failure lines (Summary too short (…) / Summary too long (…) / Batch summary too short (…)) are unchanged; only the trailing remediation sentence is now direction-aware.

Behavior change (old → new)

Old: every length-class batch rejection ended with shorten or split each offending range's summary to fit its cap, then resend the whole batch; or drop the offending range(s) and resend only the rest that validate.

New: the trailing sentence is built from the set of failure directions actually present (see above). For a too-long-only batch the text keeps its shorten or split … to fit its cap advice; for a too-short batch it now says to expand, not shorten. This is the intended fix — the old text was the bug.

Tests

kernel/tests/issue6730-length-remediation-direction.test.ts (new) pins both directions (#3617):

  • too-short rejection must NOT carry shorten or split / to fit its cap, and MUST carry expand … do NOT shorten it
  • too-long rejection still carries shorten or split … to fit its cap (control)
  • mixed batch carries both directions
  • batch-total-too-short (multi-candidate skipMin path) advises grow

Existing issue3003 / issue5894 length-gate tests still pass (18/18 targeted). Kernel typecheck clean.

Out of scope

The reporter's optional ask (reject the 3rd consecutive failure on the same range) is already covered by the existing circuit breaker (3 consecutive failures → COMPRESS CIRCUIT BREAKER); the real defect was the backwards advice, which is fixed here.

Closes #6730

…tion (#6730)

The batch rejection message told a too-short summary to "shorten or split
... to fit its cap" — backwards advice that made the summary shorter, so the
model retried into a guaranteed-fail loop (full context re-read each round)
until the circuit breaker fired.

validateSummaryLength now throws a typed SummaryLengthError carrying the
direction (empty / tooShort / tooLong); the batch screen tracks which
directions fired and renders direction-aware remediation: too-short/empty
-> "expand ... do NOT shorten it", too-long -> "shorten or split ... to fit
its cap", mixed -> both. The drop-the-range exit is always offered.

kernel-change: fix a self-contradictory error-message literal (mechanical
bug) — no algorithmic/state/wire-shape change, only the remediation text of
the batch length-rejection. Regression test:
kernel/tests/issue6730-length-remediation-direction.test.ts.
@ranxianglei ranxianglei changed the title kernel: compress length-rejection advice must match the failure direction (#6730) [PR] kernel: compress length-rejection advice must match the failure direction (#6730) Nov 14, 2026
@github-actions

github-actions Bot commented Nov 14, 2026 •

Copy link
Copy Markdown
Contributor

📦 Built Package Artifact

Branch: 2026-11-14_compress-length-remediation-direction (ea5a658)

Latest build — install this exact version

npm install -g billion-context@0.0.11192-pr.6751.27954fe

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
ea5a658 billion-context@0.0.11192-pr.6751.27954fe build
ea03daa billion-context@0.0.10911-pr.6751.dcfa43b build
8cf97f4 billion-context@0.0.10869-pr.6751.f721bc3 build
9935eea billion-context@0.0.10868-pr.6751.53fe94c build
928f312 billion-context@0.0.10003-pr.6751.62f4056 build

Download artifact

  1. Download the artifact from the Actions run
  2. Extract the tarball and install:
tar xzf billion-context-pr6751.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 — 检索 open PR,只有本 PR 针对「compress 长度拒绝建议方向」这一缺陷;其余 compress 相关 PR(#6753/#6714/#6074/#5545 等)都是不同面(nudge、ref 号、CJK guard、overlap 原子性),无并列重复。

Reviewed PR #6751 (kernel: direction-aware compress length-rejection advice, fixes #6730). Pulled the head branch (2026-11-14_compress-length-remediation-direction, 928f312) and read the full diff (4 files: version bumps + kernel/src/compress.ts +49/-4 + new test +173).

Verification

  • Root cause, not a mask. The batch-rejection remediation was a single hardcoded string ("shorten or split … to fit its cap") emitted for every length direction. It's now direction-aware: validateSummaryLength throws a typed SummaryLengthError (empty/tooShort/tooLong), the batch screen records the fired directions in a Set, and lengthRemediation() renders per-direction advice. Per-range failure lines are byte-unchanged.
  • Invariant holds. validateSummaryLength throws only SummaryLengthError (3 sites), and all three lengthFailures.push sites also populate failureKinds — so the remediation can never render with an empty direction set (no Fix it and resend the whole batch: ; or drop… degenerate sentence).
  • Other call sites unaffected. applyRefolds (L325) and validateCompressionRange (L2043) just let it throw; since SummaryLengthError extends Error, their catch handling and the per-range message text are unchanged.
  • Regression: none. Traced every assertion in issue3003/issue5894 — all still hold (the lone match(/shorten or split/) at issue3003:94 is a too-long case, which keeps that phrase). No other test or golden file pins the old string.
  • Local targeted run: issue6730 + issue3003 + issue5894 = 18/18 pass; kernel tsc --noEmit clean.
  • Kernel Boundary: kernel-change declared under the mechanical-bug exemption (owner policy 2026-11-06); version 0.0.197 → 0.0.198 (lockfile in step); reproducing regression test added; human-only merge tier still applies.

CI: still in flight on the pushed head (main CI run queued; 0 failures anywhere so far). Targeted local verification is green.

结构化八问

  1. 是什么问题: compress 批量按摘要长度被拒时,结尾整改建议对所有方向都用同一句「shorten or split … to fit its cap」;对「摘要太短」的失败方向这句正好反了(让它更短),模型照做只会越来越不合格,陷入每轮重读 300K+ token 的必败循环,直到 3 连败熔断。
  2. 什么原因导致的: 根因是批量拒绝消息里那句整改建议被写死成一个通用字符串,没区分长度失败的方向(太短/空/太长)。逐 range 的失败行(Summary too short/long)方向本身是对的,只有结尾那句整改建议错了。
  3. 影响范围与严重程度: 触发条件 = 一次 compress 批量因摘要长度被拒、且失败方向是「太短/空」(报告者正是这个场景)。严重度:p2——不是数据损坏或不可用(compress 仍能正确拒绝、熔断器兜底),但模型被误导进必败重试、每轮重读全量上下文,体验/成本有坑,需等熔断或模型自行绕过。
  4. 怎么修复的: 让整改建议方向感知——validateSummaryLength 改抛带方向的类型化 SummaryLengthError,批量筛查用 Set 记录实际触发的方向,lengthRemediation() 按方向渲染(太短/空→expand,太长→shorten or split,混合→两者,drop 出口始终保留);只改结尾那句,逐 range 失败行字节不变。
  5. 是否有回归问题: 无。逐条核过 issue3003/issue5894 现有断言全部仍成立;无其它测试/golden 钉死旧串;本地定向 18/18 绿 + kernel typecheck 干净;语义/阈值/时序/输出格式(除这句整改文案)均未漂移。
  6. 是否破坏缓存: 不涉及。纯错误文案变更,不触碰缓存/重试/增量游标/去重键;熔断器行为不变。
  7. 是否改了设计方向: 不涉及架构/边界/数据流转向。只把一句写死文案改成按方向渲染,机制层(长度筛查 all-or-nothing、批量 floor、skipMin)完全不动。
  8. 是否引入了新的配置: 不涉及。无新 env/config/默认值。

判定:可合(建议)。修复打到根因,测试双向钉死(#3617),无回归,diff 干净,Kernel Boundary 合规。kernel 变更走人工合并档——等 CI 全绿后由 owner 合。

ci-check added 2 commits November 19, 2026 02:17
# Conflicts:
#	kernel/package-lock.json
#	kernel/package.json
The merge-resolution bump of kernel/ to 0.0.200 left the root package-lock.json workspace entry at 0.0.199 (master's value); lock-sync (npm install --package-lock-only + git diff) requires the three version sites to agree.
@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

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

lock-sync fixed and pushed — new head 8cf97f4ae4.

What failed: on head 9935eea8b1 (the conflict-resolution merge), the lock-sync job ran npm install --package-lock-only + git diff --exit-code -- package-lock.json and found the root package-lock.json's packages["kernel"] workspace entry still at 0.0.199 (master's value) while kernel/package.json was bumped to 0.0.200 by the merge resolution. Root cause: the PR's original diff never touched the root lockfile (its CI was all-skipped while the PR sat in conflict state), so the three kernel-version sites — kernel/package.json, kernel/package-lock.json, root package-lock.json — only agreed in the kernel-local pair after my merge; the root entry kept master's 0.0.199.

Fix: one line in root package-lock.json — packages["kernel"].version 0.0.199 → 0.0.200, making all three sites agree. Verified locally with the exact CI commands: npm install --package-lock-only --ignore-scripts --no-fund --no-audit reports no further drift, and the committed lockfile blob (dd0c2f32d2) is byte-identical to the canonical output the CI job itself produced (its log shows index 6fedef6..dd0c2f3). All other checks on 9935eea8b1 were already green (kernel-guard, version-guard, gate, lane-guard, architecture-docs, unused-exports, mysql-contract — 29/30, only lock-sync red).

Pushed as the final head without the [ework:ci-held] marker — this is the branch's one full CI pass; the mirror check cache had not yet picked up the new run at reply time (poller refresh ~60s).

七问:

  1. 是什么问题:lock-sync 检查红——根 package-lock.json 里 kernel workspace 条目版本(0.0.199)与 kernel/package.json(0.0.200)不一致。
  2. 什么原因:PR 原始 diff 只 bump 了 kernel 本地两个版本文件、从未更新根 lockfile(PR 在冲突态期间 CI 全 skip,lock-sync 从未跑过);我解冲突时把 kernel 版本解到 0.0.200,根 lockfile 保留了 master 的 0.0.199,三处版本失步。
  3. 影响范围:仅 lockfile 一致性,零代码/行为影响;其余 29 项检查全绿。
  4. 怎么修的:根 package-lock.json 一行改动(kernel 条目 0.0.199→0.0.200),用 CI 同款命令本地验证 lockfile 已收敛到规范态(blob 与 CI 产物逐字节一致)。
  5. 回归风险:无——纯 lockfile 元数据对齐,不触碰任何依赖解析结果(npm 确认 "up to date",无其他条目变化)。
  6. 是否破坏缓存:不涉及。
  7. 当前状态:新 head 8cf97f4ae4 已推(不带 ci-held 标记,触发全量 CI 一轮);等 CI 结果。

…subset (#6730)

The verif-subset-guard CI job rejects the class form: CLASS_DECL for
'class SummaryLengthError extends Error' and UNKNOWN_GLOBAL for the three
'new SummaryLengthError' throws (the guarded-file subset only allows
NEWABLE constructors: Map/Set/Error/RangeError/TypeError/RegExp).

Rewrite as a plain-Error factory (summaryLengthError) plus a type guard
(isSummaryLengthError) — same error name, same kind field, same messages,
behavior byte-identical; the issue6730 regression tests pin the visible
rejection text and pass unchanged. No exception added to
verif/exceptions.ts: the code now conforms to the subset instead of
weakening the guard.
@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

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

Red test shards fixed — pushed final head ea03daa639.

What was red (verified live from the job logs this run): all 3 failing shards on head 8cf97f4ae4 — test (windows-latest, 22, 4), test (ubuntu-latest, 24, 2), test (ubuntu-latest, 22, 2) — failed the SAME single test, tests/verif-subset-guard.test.ts ("guarded kernel files stay inside the verifiable subset"), with the same 4 violations, all caused by this PR's new code in kernel/src/compress.ts:

  • 2038 [CLASS_DECL] — class SummaryLengthError extends Error
  • 2071 / 2084 / 2092 [UNKNOWN_GLOBAL] — new SummaryLengthError

Root cause: kernel/src/compress.ts is a verif-guarded file (KDD #14). The verifiable subset excludes class declarations entirely, and new X is only allowed for the NEWABLE set (Map/Set/Error/RangeError/TypeError/RegExp) — so the class form could never pass the guard. The guard's own message offers two paths: fix the code, or add a reviewed exception in verif/exceptions.ts.

Fix (code side, no guard weakening): rewrote the error as a plain-Error factory summaryLengthError(kind, message) plus a type guard isSummaryLengthError(error) (checks instanceof Error + name === "SummaryLengthError" + kind is one of the three directions). Same error name, same kind field, same messages — behavior is byte-identical; the 4 issue6730 regression tests pin the visible rejection text and pass unchanged. I deliberately did NOT add an entry to verif/exceptions.ts (an audited, human-justified inventory with no CLASS_DECL precedent) — the code now conforms to the subset instead of weakening the guard.

Local verification (targeted, per local-test discipline):

  • tests/verif-subset-guard.test.ts: 2/2 pass (was 1 fail, 4 violations)
  • kernel/tests/issue6730-length-remediation-direction.test.ts: 4/4 pass
  • all 7 compress-related kernel test files: 189/189 pass
  • npm --prefix kernel run typecheck: clean
  • all tests/verif-*.test.ts: 40/41 — the one failure (verif-snapshot-render buildView judge, a solver degradation in a HOST file src/server/snapshot-render.ts) is pre-existing and environment-local: A/B via git stash shows it fails identically WITHOUT this change, and it is not among the 3 red shards (CI runs it green).

Push notes: the mirror git endpoint was returning 502 this run (its API was up); per push discipline I verified the live GitHub head (8cf97f4ae4) matched my local base — no concurrent pushes — and pushed directly to GitHub. ea03daa639 is the final head with NO [ework:ci-held] marker: it is this branch's one full CI pass. CI on the new head is still starting; I will confirm the result on the next wake if anything is still red.


七问

  1. 是什么问题:PR 新增的 SummaryLengthError 类写在 verif 守护文件 kernel/src/compress.ts 里,而可验证子集不允许类声明,导致 3 个测试分片同一测试(verif-subset-guard)红。
  2. 什么原因导致的:PR 用 class ... extends Error + new SummaryLengthError 实现方向感知的长度错误;守护子集只允许 NEWABLE 构造器(Map/Set/Error/RangeError/TypeError/RegExp)且完全排除 CLASS_DECL——类形态在守护文件里结构上不可通过。
  3. 影响范围与严重程度:仅 CI 门禁红(3 个分片同一测试),不影响运行时行为;不修则 PR 无法合并。
  4. 怎么修复的:把类改写为纯 Error 工厂函数 + 类型守卫(错误名、kind 字段、消息全部不变,行为字节级一致);不改 verif/exceptions.ts——让代码符合子集,而不是削弱守护。
  5. 是否有回归问题:无。issue6730 四个回归测试(钉住对外可见的拒绝文案)全过;compress 相关 189 个 kernel 测试全过;kernel typecheck 干净。本地另有 1 个 verif 求解器测试失败,已用 stash A/B 证实与本改动无关(改动前后同样失败,属本地求解器环境差异)。
  6. 是否破坏缓存:不涉及——纯错误对象构造方式改写,不触碰缓存/重试/增量游标/去重键。
  7. 是否改了设计方向 / 新配置:不涉及——无架构变化、无新配置;kernel 版本维持 0.0.200(本 PR 已含 bump,无需再动)。

判定:修复已推送(ea03daa639,无 ci-held 标记,即本分支唯一一次完整 CI),待 CI 结果确认。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

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

Independent review of PR #6751 (head ea03daa639) — findings only, no approval (kernel human-only tier; see the 🔶 需要决策 comment below).

What I verified (own pass on the diff, not a copy of the earlier review):

  1. Correctness — the fix hits the producer, not a mask. The batch length-rejection in kernel/src/compress.ts used ONE hardcoded remediation sentence for every direction. Now validateSummaryLength throws a typed error carrying kind (empty/tooShort/tooLong), the batch screen records the fired directions in failureKinds, and lengthRemediation() renders per-direction advice (too-short/empty → expand, too-long → shorten-or-split, mixed → both, drop-exit always offered). Per-range failure lines are byte-unchanged (diff-verified).
  2. The no-degenerate-sentence invariant holds structurally. lengthFailures.push has exactly three sites (solo catch L738, multi-candidate catch L751, batch-total L766) and each is paired with a failureKinds.add — the two catch sites via isSummaryLengthError, which is always true because validateSummaryLength throws ONLY the typed error (all three throw sites are summaryLengthError), and the batch-total site adds "tooShort" directly. lengthRemediation can never render with an empty direction set.
  3. Other call sites unaffected. applyRefolds (L321) and validateCompressionRange (L2019) just let the error propagate; message text is unchanged and no code anywhere consumes error.name for these errors (grep: only an unrelated AbortError check in src/agent/pi-inline.ts:50). The host failure-class regex in src/stream.ts:1562 matches the message head Batch rejected before folding any range — byte-unchanged.
  4. verif subset conformance. The factory + type-guard form (instead of a class) is what the guarded-file subset requires; nothing added to verif/exceptions.ts (guard not weakened). tests/verif-subset-guard.test.ts 2/2 locally.
  5. Tests have teeth (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 both directions). The 4 new tests assert the exact model-visible strings: too-short must NOT carry shorten or split / to fit its cap and MUST carry the expand advice; the too-long control still carries shorten or split each too-long summary to fit its cap; mixed and batch-floor covered. Ran locally: issue6730 + issue3003 + issue5894 = 18/18 pass; kernel tsc --noEmit clean.
  6. Regression sweep of the touched path. Existing pins of the rejection text (issue3003:94 /shorten or split/, issue5765:264, issue5894:*, validation.test.ts:301) all match the unchanged head/phrase — verified by grep + local run. The apply-level rejection path (compress.ts:893, content-class failures) is untouched; circuit-breaker logic untouched; no cache/retry/cursor/dedup impact.
  7. Model-facing text change (disclosed). This is prompt surface — the compress rejection wording the model reads. Old→new is disclosed in the PR body's "Behavior change" section: too-long-only rejections keep the shorten or split … to fit its cap advice (framing sentence reworded); too-short/empty rejections now say expand, never shorten. No other behavior moved.
  8. Kernel Boundary compliance. kernel-change: declared in the PR body under the mechanical-bug exemption (owner policy 2026-11-06); kernel version 0.0.199 → 0.0.200 with all three version/lockfile sites in sync (lock-sync green); reproducing regression test added; the kernel commit 928f312575 carries the kernel: prefix and the declaration.

CI: green at head ea03daa639 per the dispatch (mirror cache at 23:59Z showed 0 failed / 3 pending, since completed; no gh CLI in this environment to re-poll).

Findings: none blocking; no small findings to fix on the spot — the diff is 5 files (3 version sites + kernel/src/compress.ts +62/−4 + new test +173), every line on-purpose, naming clean, comments explain why.

八问

  1. 是什么问题: compress 批量按摘要长度被拒时,结尾整改建议对所有方向共用一句「shorten or split … to fit its cap」——对「太短/空」的失败方向正好说反,模型照做越改越短,陷入每轮全量重读 300K+ token 的必败循环直到熔断。
  2. 什么原因导致的: 根因是批量拒绝消息的整改句被写死成单一字符串,不区分 validateSummaryLength 实际抛出的失败方向(空/太短/太长);逐 range 失败行本身方向正确,只有结尾那句建议错。
  3. 影响范围与严重程度: 所有触发「摘要太短/空」批量长度拒绝的会话(任意宿主,summary 低于 minSummaryLength=50 或为空)。p2:压缩功能本身可用(写长即成功),但错误建议方向反了,造成成本浪费(每轮全量重读 + 缓存反复作废)与困惑循环,需模型自行绕过或等熔断兜底。
  4. 怎么修复的: 让整改建议方向感知——validateSummaryLength 改抛带 kind 的类型化错误(工厂函数 + 类型守卫,符合 verif 可验证子集),批量筛检用 Set 记录实际触发的方向,lengthRemediation() 按方向渲染;只改结尾一句,逐 range 失败行字节不变。
  5. 是否有回归问题: 无。既有断言(issue3003:94 的 /shorten or split/、issue5765:264、issue5894、validation.test.ts:301 钉的消息头)全部仍成立,本地 18/18 绿;内容类拒绝路径(compress.ts:893)、熔断器、缓存/重试/游标均未触碰;「太长」方向的建议语义保持(仅框架句措辞变化,已在 PR 正文 old→new 披露)。
  6. 是否破坏缓存: 不涉及——纯错误文案变更,不碰缓存/重试/增量游标/去重键。
  7. 是否改了设计方向: 不涉及——机制层(长度筛查 all-or-nothing、批量 floor、skipMin)完全不动,只把一句硬编码建议改成按方向渲染。
  8. 是否引入了新的配置: 不涉及——无新 env/config/默认值。

判定: 代码层面干净、修复打到根因、测试双向钉死、无回归——review 通过(findings-only);因触碰 kernel/,按 Kernel Boundary 人工合并档,不做 approve,等 owner 授权(见 🔶 需要决策 评论)。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

🔶 需要决策

PR #6751(kernel: compress 长度拒绝建议方向须与失败方向一致,Closes #6730)触碰 kernel/ 并带 kernel-change: 声明(机械缺陷豁免,owner policy 2026-11-06)。按 Kernel Boundary,kernel 变更一律人工合并档——上面的 findings-only review 即豁免要求的线程内报告:

  • 改了什么: kernel/src/compress.ts —— validateSummaryLength 改抛带方向的类型化错误(工厂函数 + 类型守卫,符合 verif 可验证子集),批量长度拒绝的结尾整改句按实际失败方向渲染(太短/空→扩写、太长→缩短或拆分、混合→两者、drop 出口保留);kernel 版本 0.0.199 → 0.0.200(三处版本/lockfile 同步);新增回归测试 kernel/tests/issue6730-length-remediation-direction.test.ts。
  • 为何机械安全: 无算法/状态/线形变更——只改批量长度拒绝消息的结尾一句(旧句对「太短」方向说反,即缺陷本身);逐 range 失败行字节不变;既有测试钉全部仍成立(本地 18/18);属「错误字面量自相矛盾」类,正是豁免条款的目标。
  • 回归测试: kernel/tests/issue6730-length-remediation-direction.test.ts(4 例:太短/太长/混合/批量 floor,双向钉死)。
  • 被 gate 的 diff: kernel/src/compress.ts(+62/−4)、kernel/package.json、kernel/package-lock.json、package-lock.json(版本同步)、新测试(+173)。

请求 owner 授权合并(head ea03daa639,CI 绿)。

…tion (#6751 conflict)

Resolve the dirty merge state against the advanced master:
- kernel/src/compress.ts auto-merged cleanly (PR's #6730 length-remediation
  direction change + master's batch-seam double-claim fix coexist)
- kernel version collision: PR had 0.0.200, master advanced to 0.0.201
  (batch seam fix) -> bump to 0.0.202 in kernel/package.json,
  kernel/package-lock.json (both occurrences), and root package-lock.json

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

Projects

None yet

1 participant