unit 3a of the 2226 split: view state machinery, registration, and the boolean webview post - #2414

Split unit 3a of the plan issued on issue 1992 for PR 2226. Content authority is PR 2226 head bc724b8df51f6d5585ffbac14ada09809d99a5f0; every ported line is verbatim from that source at the same path, selected by donor new-side line range against unit 2's head cbf11148aea5a7390e88759be077e6fd1c035068, which is this branch's only parent.

What is in it

• src/core/webview/ClineProvider.ts donor new-side 78-169, 171-176, 178-292 and 294-1149 (the view state claim fields, the scheduled write and generation counters, the claim broker, postViewStateClaimToWebview and the registration and dispose wiring around them), 2419-2478 (the postMessageToWebview group, whose return becomes Promise<boolean>) and 2537-2542 (the donor's rewritten diagnosticsRequest callback that consumes that value).
• src/core/webview/__tests__/ClineProvider.spec.ts donor new-side 15-17 and 19-5270, plus the two postMessageToWebview mock reversions at 9667 and 9771.
• src/core/webview/webviewMessageHandler.ts new-side 705-715 (the view state claim flush) with its spec at new-side 131-132 and 2809-2826.
• src/__tests__/extension.spec.ts new-side 377 and src/extension/__tests__/api-start-new-task.spec.ts new-side 22.

Units 4 and 5 gave up the ranges at 2419-2478, 2537-2542, 377, 22, 9667 and 9771 because 3a is the unit that ships Promise<boolean>: mockResolvedValue(undefined) feeding that method, and returning Promise<boolean> into a Promise<void) callback parameter, both fail to typecheck at this head otherwise. Five donor declarations went the other way, to unit 3b, because their only use sits in the code 3b ships: PROVIDER_SETTINGS_KEYS at new-side 77 (used at 6270 and 6271), ViewLocalStateValues at 170 (used at 177 and 6229), type PersistedViewStateValues at 177 (used at 6225 and 6227), type ViewLocalStateField at 293 (used at 6367) and type GlobalState in the spec at 18 (used at 6004 and 6104). Each of them is reported as an unused name at this head by @typescript-eslint/no-unused-vars, so a declaration ships with the unit whose code reads it.

Sizes, with the base named beside every figure

Unit-only against this branch's parent cbf11148aea5a7390e88759be077e6fd1c035068: 6 files changed, 610 insertions(+), 25 deletions(-), 629 additions and deletions over 6 files. Hunk count 28 at the default three-line context of git diff cbf11148aea5 8aec28c83cc8, 30 with unified 0; the numbering is named with the number.

Cumulative against upstream: base branch main at 18861b2e357203e6cb4d93930aa2ed122b1eecea, merge base e943530c848624bd3054c960be1e4fb0b658509a, and git diff e943530c848624bd3054c960be1e4fb0b658509a 76ab52f6889abff0b111078d2b651ba6ec6d53ec measures 20 files changed, 892 insertions(+), 36 deletions(-) over 19 files. GitHub's own additions and deletions for this PR are read back after the POST rather than asserted here.
Against the campaign's own diff budget, 400 additions and deletions as the soft line and 1,000 as the hard line: this unit is at 1,069 additions and 36 deletions, which is 1,105 additions and deletions, read as GitHub's own figures for the pull at head ca2dc28 over merge base 18861b2. That merge base is also the current upstream tip, so tip and merge base are one commit here rather than two numbers, and the additions and deletions are GitHub's PR figures rather than my count of a commit. The unit stays above the soft line because the members this unit declares are read by code that arrives in the same unit, and cutting at the next available boundary would separate a declaration from its earliest reader. The soft line is a request for a reviewer's judgement rather than a claim of exemption: the campaign draws the same distinction at the mutation gate, which offers a further split or a maintainer-reviewed exclusion instead of asserting that a number is acceptable.

The unit's price on the gate's two axes at this head, from the CI run whose head is ca2dc28, with each package measured against the caps separately the way scripts/stryker-diff.mjs does at lines 265 to 267 for changed executable lines and lines 713 to 715 for valid mutants, and with both caps declared in that same file at line 11 as MAX_CHANGED_LINES = 500 and line 12 as MAX_MUTANTS = 400: package types, 8 changed executable lines and 1 valid mutant, instrumented as 2 selector ranges; package extension, 352 changed executable lines and 274 valid mutants, instrumented as 78 selector ranges. The selector-range counts are mine, measured with the gate script's own selectFromGit against a commit-tree merge commit whose first parent is the merge base named above, merge commit e0f396f7732dac9b0682dd829e8c570e641eb22a, which reproduces the CI figures of 8 and 352 changed executable lines. No binary artifact sits inside the additions-and-deletions arithmetic: git diff --numstat between merge base 18861b2 and head ca2dc28 lists 20 files and prints numeric addition and deletion counts on every row, with no row carrying the binary marker.

Gate price at this unit's own merge base

selectFromGit in scripts/stryker-diff.mjs (blob 7dfe0006d2d5c7e4d2282f65f68b8ed5f16866bc, MAX_CHANGED_LINES = 500, MAX_MUTANTS = 400), run against a commit-tree merge commit whose first parent is the base, merge commit eaf871557326c68cc133106da301f563f1935184, which reports baseSha and mergeBase both equal to cbf11148aea5a7390e88759be077e6fd1c035068: package extension, vitestRelated true, 311 changed executable lines, 2 files, 63 selectors, advisories empty, script exit 0. That is under the 500 changed executable line cap, and 629 additions and deletions is under the 1,000 cap.

Local Stryker preflight: attempted, and what it produced

node zdt.mjs mutation preflight --worktree <this worktree> --base cbf11148aea5, run at this head in a worktree with a real install:

Mutation-testing 1 package(s) from merge base cbf11148aea5: extension (311 lines)
::warning title=Mutation test advisory::extension Stryker preflight could not start: spawnSync <worktree>\node_modules\.bin\stryker ENOENT
no reports (mutation testing not applicable or run incomplete)
=> PASS (local gate green)

The gate script spawns node_modules/.bin/stryker without a shell option; CreateProcess cannot execute an extensionless pnpm shim and does not apply PATHEXT to it, so the selection completes and the spawn fails with ENOENT. Issue 2410 is about that defect. The tool's own digest for this run says "packages": [] with "pass": true, which is evidence absent, not a valid zero, so I am not reporting a local mutation result for this unit. The CI mutation-diff leg at this head is the mutation evidence, and I will read that leg after the checks land instead of treating an advisory or a selection-less shape as a passing mutation run.

Checks run at this head

• pnpm run check-types with TURBO_FORCE=1 from a cold cache: exit 0 with 0 error TS rows.
• pnpm --dir src exec eslint --max-warnings=0 over the six changed files: exit 0. Resolved preset, from eslint --print-config src/core/webview/ClineProvider.ts: 25,431 bytes, sha256 ce4e18fb6106f609252d5b36f79408bc98911cb65815c155d51584b458499a0c.
• src/eslint-suppressions.json unchanged from unit 2's head, blob sha256 8423ab14b663d472: no file added to it and no count raised.
• Hooks ran, nothing was skipped with --no-verify. Pre-commit: hook directory .husky/_, sh from GNU bash 5.3.9(1)-release, lint-staged over the six staged files with prettier plus the eslint command above; after the commit each of the six committed blobs equals the blob that was linted, so the committed tree is the tree measured. Pre-push ran pnpm run check-types (11 of 11 tasks, 11 from cache, 61 ms) and printed Changeset files: 10, the count of changeset files already in the repo.
• No .changeset file added, no CHANGELOG edit, nothing added to the tracked tree, no label change, no review action.

Chain

Unit 3b (provider new-side 1150-1641 and 6219-6398, spec new-side 5723-6124, plus the five declarations named above) measures 736 additions and deletions. It is measured and not opened: it waits until this unit is green at its own head, one unit at a time. After it: 4a at 167, 4b whole at 718 (its describe("view state claim binding") opens at spec new-side 6249 and closes at 6966, so the earlier ruled boundary at 6597 would leave the first half needing a patch edit rather than a selection, and check-types says so with ClineProvider.spec.ts(9086,1): error TS1005: '}' expected), and then 4c at 349.

Branch refresh at this head, measurement rather than preference

This head's parent is main tip 18861b2. The ort merge of the previous head d475ef2512d0e981f9b1a39c10d36a8729cdee71 with that tip was clean with zero conflicts, merge commit 2b7e22ffa26d3b6f956a2e72301a400a85e5eb46; it was folded into this one commit because the adapted spec line below only compiles once main's content is an ancestor of it. The probe that settled it: main's copy of src/core/task/tests/Task.spec.ts placed on top of the unmerged unit tree gives seven type errors, among them the missing module test-utils/run-state and four reads of Task.pendingPartialMessageTimer. The merge base for this PR is main tip, and the diff against main is this chain's own content: 20 files, 892 additions, 36 deletions, 928 additions and deletions, which is the sum of the printed unit rows 192 + 106 + 629 + 1. Unit 3a alone measures 629 additions and deletions against its own parent cbf1114: six files, 604 additions, 25 deletions. The delta is named rather than assumed: 635 minus the six lines of the ViewStateRegistrationResult declaration and its doc comment, moved to unit 3b, whose only reads of that type are provider new-side 1153 and 1156. knip exits 0 at this head; at 8aec28c the knip leg failed with Unused exported types (1) ViewStateRegistrationResult at src/core/webview/ClineProvider.ts:275:13.

The adaptation lives in this commit: src/core/task/tests/Task.spec.ts line 4526 adds return true inside the mockImplementation callback that main added at line 4521. git show of ClineProvider.ts prints public async postMessageToWebview(message: ExtensionMessage) at line 1702 on main tip 18861b2 and the same declaration at line 1702 on 2411 head cbf1114, while this head prints public async postMessageToWebview(message: ExtensionMessage): Promise at line 2252. Main brought the call site and this unit brought the boolean return, so the adaptation travels with the return type instead of being smuggled into a merge. It is a spec row, and the mutation gate charges executable paths, not spec rows.

Local verification at this head. Non-interactive frozen-lockfile install: pnpm install --frozen-lockfile --config.confirmModulesPurge=false, exit 0 in 73 s; that switch is what gets past the purge prompt that made the earlier frozen install stop at its own question and change nothing. After it, pnpm run check-types with cwd = src, cold through TURBO_FORCE=1, exits 0 with 0 error rows, and the two integrations/misc/export-markdown.ts rows that the stale install produced are absent. pnpm run knip exit 0. eslint over the changed files with --max-warnings=0 exit 0. Hooks ran live on this commit and on this push: core.hooksPath is .husky, interpreter C:/Program Files/Git/usr/bin/sh.exe reporting GNU bash 5.3.9(1)-release, lint-staged eslint and prettier completed rather than skipped, and the pre-push leg printed its changeset inventory without blocking. The gate priced at the new merge base reports the extension package with 345 changed executable lines across 6 files and 73 selectors, exit 0, no advisories; the same recipe at base cbf1114 priced 311 changed executable lines across 2 files with 63 selectors, which is the figure the plan priced for unit 3a and it did not move.

Mutation rows at the previous head, and what happens to each

At head 8aec28c the gate's own digest is comment 6772888041. Its marker line reads: zoo-code-mutation-report head:8aec28c83cc8a2976a81b6eb4bc43df9d0a0492e blocking:237 advisories:0. Its per-package table carries one row: extension, 237 surviving or uncovered mutants, 0 advisories. That report prints no valid, killed or timeout column, so those three quantities are not available from it, and the annotations are capped at 20 locations and 7 per file, so an annotation tally is not the gate's blocking figure. Parsed from the same comment the 237 rows are: src/core/config/ProviderSettingsManager.ts 1 survived, src/core/webview/ClineProvider.ts 234 with 16 survived and 218 uncovered, src/core/webview/webviewMessageHandler.ts 2 uncovered. Mapping every reported line against the diff that introduced it puts 236 rows on lines this unit introduced and 1 row on content shipped by 2407 and 2411.

Disposition, the third one, stated rather than assumed: this push adds no test row and no Stryker disable. The spec hunks that exercise the registration, claim delivery and record-hygiene machinery behind those provider lines are donor spec ranges 5723 to 6124 and 6125 to 6248, which the plan assigns to units 3b and 4a, and 6249 to 6966, which it assigns to unit 4b. Unit 3b measures 160 changed executable lines with 67 selectors at its own base and unit 4a measures 17 with 8, so those units price the coverage this head is missing; unit 4b is spec-only and prices no executable lines. The single ProviderSettingsManager.ts row is priced by 2411. Unit 3a is not banked as mutation-clean on the strength of a success conclusion. No suppression count moved: src/eslint-suppressions.json at this head is byte-identical to it at cbf1114.

Reviewed surface, stated plainly. When this PR opened it printed 897 additions and 36 deletions across 19 files at head 8aec28c, and that figure already carried the 192 and 106 lines of 2411, because git merge-base --is-ancestor cbf1114 18861b2 answers no: 2411 is not an ancestor of main, so a PR based on main counted it. The chain was implicit then and it is explicit now, so the reviewed surface did not grow, it moved. The rows this head carries are 192 plus 106 plus 629 plus 1, which is 928 additions and deletions across 20 files against main tip 18861b2. Unit 3a's own increment is 629 additions and deletions against cbf1114, six files, 604 additions and 25 deletions, plus the one adaptation line, which is a spec row. If 2411 lands first, the remainder here is 629 plus 1, that is 630 additions and deletions, because reviewers are entitled to know which lines they are being asked to read twice. The branch this PR reads lives in the easonLiangWorldedtech/Zoo-Code fork, which is where this head was pushed.

Survived mutants named row by row. The digest at head 76ab52f reports 239 blocking rows, of which 19 are survived mutants: 18 on ClineProvider.ts lines this chain carries and 1 on src/core/config/ProviderSettingsManager.ts:62. Survived and uncovered are different debts, so only these 19 read to a reviewer as a missing test. The 218 uncovered rows stay the range-level hand-off named above and are not enumerated here. Each survived row gets its own line below with the donor spec range and unit that owns the assertions speaking about the same field or value, the spec line and test they sit in, and what is missing. Because all 19 survived this run, what is absent is a discriminating assertion rather than an assertion, so nothing below is claimed as a kill.- src/core/webview/ClineProvider.ts:310 ArrayDeclaration, replacement "["Stryker was here"]". assertions naming pendingViewStateWrites exist in donor spec ranges owned by 4c (4 rows, first at spec 7221); quoted, at spec 7221 inside spec 7194 "it("detaches a provider's queued writes when it disposes", async () => {" the line reads "expect(ClineProvider["pendingViewStateWrites"]).toContain(queued)".

• src/core/webview/ClineProvider.ts:493 BooleanLiteral, replacement "true". assertions naming wroteUnderTemporaryViewStateId exist in donor spec ranges owned by 4b (5 rows, first at spec 6806) and 4c (4 rows, first at spec 7016); quoted, at spec 6806 inside spec 6780 "test("still refuses an id a live view holds, even when the claim matches"," the line reads "expect(provider["wroteUnderTemporaryViewStateId"]).toBe(true)".
• src/core/webview/ClineProvider.ts:504 ArrayDeclaration, replacement "["Stryker was here"]". assertions naming scheduledViewStateWrites exist in donor spec ranges owned by 4c (5 rows, first at spec 7213); quoted, at spec 7213 inside spec 7194 "it("detaches a provider's queued writes when it disposes", async () => {" the line reads "expect(victim["scheduledViewStateWrites"].length).toBeGreaterThan(0)".
• src/core/webview/ClineProvider.ts:527 ObjectLiteral, replacement "{}". no assertion anywhere in the donor spec blob e5627d4b46f3 names viewLocalStateWriteVersions, so no unit's spec range is on the hook for a discriminator yet.
• src/core/webview/ClineProvider.ts:564 StringLiteral, replacement "``". no assertion anywhere in the donor spec blob e5627d4b46f3 names nextViewId, so no unit's spec range is on the hook for a discriminator yet.
• src/core/webview/ClineProvider.ts:564 UpdateOperator, replacement "ClineProvider.nextViewId--". no assertion anywhere in the donor spec blob e5627d4b46f3 names nextViewId, so no unit's spec range is on the hook for a discriminator yet.
• src/core/webview/ClineProvider.ts:1037 ConditionalExpression, replacement "true" (mutant 1 of 5 ConditionalExpression mutants this line produces). assertions naming pendingViewStateClaims exist in donor spec ranges owned by 4b (8 rows, first at spec 6477); quoted, at spec 6477 inside spec 6451 "test("reports a claim as undelivered when the attached webview rejects the" the line reads "expect(rejecting["pendingViewStateClaims"].get("hidden-view")).toBe("host-minted-token")".
• src/core/webview/ClineProvider.ts:1037 ConditionalExpression, replacement "false" (mutant 2 of 5 ConditionalExpression mutants this line produces). assertions naming pendingViewStateClaims exist in donor spec ranges owned by 4b (8 rows, first at spec 6477); quoted, at spec 6477 inside spec 6451 "test("reports a claim as undelivered when the attached webview rejects the" the line reads "expect(rejecting["pendingViewStateClaims"].get("hidden-view")).toBe("host-minted-token")".
• src/core/webview/ClineProvider.ts:1037 LogicalOperator, replacement "(this._disposed || !this.view || !this.i" (mutant 1 of 3 LogicalOperator mutants this line produces). assertions naming isViewLaunched exist in donor spec ranges owned by 4b (3 rows, first at spec 6542); quoted, at spec 6542 inside spec 6525 "test("holds a claim minted after a renderer replacement and delivers it on" the line reads "expect(replaced.isViewLaunched).toBe(true)".
• src/core/webview/ClineProvider.ts:1037 ConditionalExpression, replacement "false" (mutant 3 of 5 ConditionalExpression mutants this line produces). assertions naming pendingViewStateClaims exist in donor spec ranges owned by 4b (8 rows, first at spec 6477); quoted, at spec 6477 inside spec 6451 "test("reports a claim as undelivered when the attached webview rejects the" the line reads "expect(rejecting["pendingViewStateClaims"].get("hidden-view")).toBe("host-minted-token")".
• src/core/webview/ClineProvider.ts:1037 LogicalOperator, replacement "(this._disposed || !this.view) &&amp" (mutant 2 of 3 LogicalOperator mutants this line produces). assertions naming isViewLaunched exist in donor spec ranges owned by 4b (3 rows, first at spec 6542); quoted, at spec 6542 inside spec 6525 "test("holds a claim minted after a renderer replacement and delivers it on" the line reads "expect(replaced.isViewLaunched).toBe(true)".
• src/core/webview/ClineProvider.ts:1037 ConditionalExpression, replacement "false" (mutant 4 of 5 ConditionalExpression mutants this line produces). assertions naming pendingViewStateClaims exist in donor spec ranges owned by 4b (8 rows, first at spec 6477); quoted, at spec 6477 inside spec 6451 "test("reports a claim as undelivered when the attached webview rejects the" the line reads "expect(rejecting["pendingViewStateClaims"].get("hidden-view")).toBe("host-minted-token")".
• src/core/webview/ClineProvider.ts:1037 LogicalOperator, replacement "this._disposed && !this.view" (mutant 3 of 3 LogicalOperator mutants this line produces). assertions naming isViewLaunched exist in donor spec ranges owned by 4b (3 rows, first at spec 6542); quoted, at spec 6542 inside spec 6525 "test("holds a claim minted after a renderer replacement and delivers it on" the line reads "expect(replaced.isViewLaunched).toBe(true)".
• src/core/webview/ClineProvider.ts:1037 BooleanLiteral, replacement "this.view" (mutant 1 of 2 BooleanLiteral mutants this line produces). assertions naming _disposed exist in donor spec ranges owned by 3a (4 rows, first at spec 2796); quoted, at spec 2796 inside spec 2779 "test("disposes the current tab provider and its watchdog registration", as" the line reads "expect(provider["_disposed"]).toBe(true)".
• src/core/webview/ClineProvider.ts:1037 BooleanLiteral, replacement "this.isViewLaunched" (mutant 2 of 2 BooleanLiteral mutants this line produces). assertions naming _disposed exist in donor spec ranges owned by 3a (4 rows, first at spec 2796); quoted, at spec 2796 inside spec 2779 "test("disposes the current tab provider and its watchdog registration", as" the line reads "expect(provider["_disposed"]).toBe(true)".
• src/core/webview/ClineProvider.ts:1037 ConditionalExpression, replacement "false" (mutant 5 of 5 ConditionalExpression mutants this line produces). assertions naming pendingViewStateClaims exist in donor spec ranges owned by 4b (8 rows, first at spec 6477); quoted, at spec 6477 inside spec 6451 "test("reports a claim as undelivered when the attached webview rejects the" the line reads "expect(rejecting["pendingViewStateClaims"].get("hidden-view")).toBe("host-minted-token")".
• src/core/webview/ClineProvider.ts:1037 EqualityOperator, replacement "this.pendingViewStateClaims.size !== 0". assertions naming pendingViewStateClaims exist in donor spec ranges owned by 4b (8 rows, first at spec 6477); quoted, at spec 6477 inside spec 6451 "test("reports a claim as undelivered when the attached webview rejects the" the line reads "expect(rejecting["pendingViewStateClaims"].get("hidden-view")).toBe("host-minted-token")".
• src/core/webview/ClineProvider.ts:2266 BooleanLiteral, replacement "true". this is the false branch of the boolean return that 3a itself introduced, so it is 3a's own content and 3a's own debt rather than a hand-off; assertions that call postMessageToWebview appear in donor spec ranges owned by 3a, but none compares its result against false, which is why flipping it to true survived.
• src/core/config/ProviderSettingsManager.ts:62 StringLiteral, replacement """" on DEFAULT_WRITE_DELAY_MS. The assertions that pin that constant live in ProviderSettingsManager.spec.ts, a different file, and none of them fall inside donor ClineProvider.spec ranges 5723 to 6124, 6125 to 6248 or 6249 to 6966; the line shipped with unit 2, PR 2411, so unit 2 prices it and no unit on this chain does.

Round-19 addition: the detach-loop commit and the rows that changed class

Commit ca2dc28 (92 insertions and 0 deletions in src/core/webview/tests/ClineProvider.spec.ts, tip ca2dc28 over its parent and merge base 735dc22) answers three of the five mutants that the detach loop this unit added to dispose() introduced at provider lines 1492 and 1493, which are owed by this unit and not carried. Three are killed at head ca2dc28 and are listed one per line in the round-21 section below. Two are still Survived in the digest at that head, named in that section with the statement that would kill them. PR-side the branch now reads 1,069 additions and 36 deletions across 20 files at head ca2dc28 over merge base 18861b2, which are GitHub's own figures for the pull rather than my count of the commit.

The mutation digest moved 238 to 235 blocking between the two heads, and the row lists move more than the headline suggests. Nine rows changed class from NoCoverage to Survived at lines this commit did not touch, listed one per line with the statement that would kill the mutant and the unit I am assigning the assertion to. Donor ranges are given only where the ledger already carries a measured one; where it does not, the range is named as unmeasured rather than filled in.

• ClineProvider.ts:752 ConditionalExpression (the false arm of the return in isViewStateGenerationLive): killed by asserting that a provider which is not disposed and whose generation still matches reports the generation live. The shipped spec asserts only the opposite direction, at webviewMessageHandler.spec.ts sibling range core/webview/tests/ClineProvider.spec.ts 2802 to 2810, which is the range to extend. Unit 3a carries it.
• ClineProvider.ts:752 LogicalOperator (the compound !this._disposed && this.viewStateGeneration === generation replaced by one operand): killed by asserting both states in one pair, live before disposal and dead after disposal on the same generation. Same range, unit 3a.
• ClineProvider.ts:752 BooleanLiteral (this._disposed replaced by a constant): killed by the same pair, since a constant true or false makes one arm of the pair impossible. Same range, unit 3a.
• ClineProvider.ts:763 OptionalChaining (the optional link in entry.provider?.scheduledViewStateWrites.push(entry) removed): killed by enqueueing an entry whose provider field is already null and asserting that the shared queue still holds it and nothing throws. The existing detach test asserts the positive case only. Unit 3a.
• ClineProvider.ts:765 BlockStatement (the body of the queue step replaced by an empty block): killed by asserting that a queued write's effect actually happens when the pump reaches it, which is the queue behaviour the ledger already assigns to unit 4c at donor spec 7213 and 7221 inside the it at 7194.
• ClineProvider.ts:770 ConditionalExpression, both arms, plus EqualityOperator on the index test and UnaryOperator on the delete count: killed by asserting that a detached entry is skipped when the queue reaches it while a live entry's run happens, the pair the queue step's own guard needs. Assigned to unit 4c on the queue side at the same measured donor range, and to unit 3a for the detached-entry half, whose donor range is not yet measured. At head ca2dc28 the EqualityOperator on the index test, the UnaryOperator on the delete count and the ConditionalExpression with replacement false are killed by the commit named above, so what remains Survived at 770 is the ConditionalExpression with replacement true, and that row stays assigned as written here.

The rows the fix itself introduced at 1492 and 1493. Answered at head ca2dc28: ClineProvider.ts:1492 ConditionalExpression with replacement true, ClineProvider.ts:1492 EqualityOperator with replacement index === -1, and ClineProvider.ts:1492 UnaryOperator with replacement +1. Carried by unit 3a: ClineProvider.ts:1492 ConditionalExpression with replacement false, and ClineProvider.ts:1493 CallExpression with replacement ;. Both carried rows are inert once the queue step at ClineProvider.ts 766 to 772 has already removed the entry from the shared static list before awaiting that entry's run, and both shipped tests reach dispose behind an await, so indexOf misses and lines 1490 to 1496 change nothing; that is control-flow reasoning rather than a measured kill map. The statement to add, written down so whoever pays it inherits the shape: call dispose with no intervening await and assert that the provider entry is gone from ClineProvider["pendingViewStateWrites"] before awaiting the result.

Four rows left NoCoverage and are not among the nine. Old 760 ConditionalExpression with replacement true and old 770 CallExpression with replacement ; appear in neither list once the eight-line removal and sixteen-line insertion are applied, and webviewMessageHandler.ts:711 BlockStatement and :713 StringLiteral appear in neither list although that file is identical at both heads. Absence from both lists is a kill rather than a move; the digest does not name the killing test, so no kill is credited to a row here.

Basis for the arithmetic: the copy of the previous digest used is run 42865427101's artifact file pr-comment.md, sha256 367b68a6d998410e25e073f1272dcb524a7e43a3d30312e187059fa541dfa6fc, whose own marker reads blocking:238 with 18 Survived and 220 NoCoverage at 76ab52f, while the live comment for that same head read blocking:239 with 19 Survived, the extra row being the ObjectLiteral at 527 that the artifact lists as killed. Against the artifact basis: 18 survivors minus the four that left as kills (504 on scheduledViewStateWrites, both 564 mutants on nextViewId, the false branch at 2266) plus the nine that arrived from NoCoverage plus the five new from this unit equals 28; 220 uncovered minus nine minus four equals 207; 28 plus 207 equals the 235 in the marker at 735dc22. The row at old 504 is unit 4c's debt and its kill is not credited here.

Round-21 addition: the digest at head ca2dc28, what the test commit paid, and align

Digest comment 6772888041 now reads for head ca2dc28. The workflow rewrote it: mutation-diff job 130039340780 of run 42875953881 completed at 2026-11-10T18:43:06Z and the comment's updated_at is 2026-11-10T18:43:24Z, so the figures below are the workflow's own and I issued no PATCH against that comment. The artifact is 13791473696, file pr-comment.md, 21,492 bytes and 286 lines, sha256 f6782216d53647238ff8ad309c6757d6f31d93eecd468bc2c2298e4f726a3e0c, whose marker reads zoo-code-mutation-report head:ca2dc281f6d6e3f8e7023b40b540a394dd225c35 blocking:229 advisories:0. Package rows: types, 8 changed executable lines, 1 valid, 1 killed, 0 timeout, 0 survived, 0 no coverage, Passed; extension, 352 changed executable lines, 274 valid, 45 killed, 0 timeout, 22 survived, 207 no coverage, Advisory findings. The job reports 9 annotations against 229 listed rows and the comment states its own cap as up to 20 unique locations with 7 per file, so the annotations are the capped view while the row list is the figure. The comment's tables match the artifact row for row, 229 rows on each side with 22 Survived and 207 NoCoverage on each side; the only differences are two rows rendered as an empty cell in the markdown artifact and as an empty code span in the published tables, at ClineProvider.ts 976 and 986, both NoCoverage StringLiteral.

Movement between 735dc22 and ca2dc28, against the reconciled previous figures named above (valid 274, killed 39, survived 28, no coverage 207, blocking 235): killed 39 to 45, survived 28 to 22, no coverage unchanged at 207, valid unchanged at 274, blocking 235 to 229, and no mutant entered or left the priced set, which is what a test-only commit should look like in this digest. The six rows the commit paid for, one per line:

• src/core/webview/ClineProvider.ts:770 Survived ConditionalExpression with replacement false
• src/core/webview/ClineProvider.ts:770 Survived EqualityOperator with replacement index !== -1
• src/core/webview/ClineProvider.ts:770 Survived UnaryOperator with replacement +1
• src/core/webview/ClineProvider.ts:1492 Survived ConditionalExpression with replacement true
• src/core/webview/ClineProvider.ts:1492 Survived EqualityOperator with replacement index === -1
• src/core/webview/ClineProvider.ts:1492 Survived UnaryOperator with replacement +1

zdt align has not been run for this unit at this head, and the reason is a measurement rather than a preference. git ls-tree -r --name-only lists no tracked file matching zdt at upstream tip 18861b2 and none at head ca2dc28; a filesystem walk of C:/work/Zoo-Code-fork to depth four with node_modules and .git skipped finds no file matching zdt*.mjs; a PATH lookup for zdt reports no match. The invocation quoted in the preflight section above was recorded at an earlier head and I am not re-running it here. Align is an integration-PR step rather than a per-unit one, the campaign states its absence rather than substituting a hand comparison for it, and units 4a and 4b stay unopened until heads exist with widths measured on both gate axes.