tooling(pm): take node:path as a namespace in dispatch-gates so no fixture can shadow it

[claude]

Fixes#14688

selfTest() binds fixture values to two names this module imports from node:path, so a later case reaching for either gets the fixture instead of the function and dies at run time. Parse-time checks cannot see it; the run-time message names a real binding, so it reads like a broken import rather than a shadow.

What changed

The module took five bare names from node:path. It now takes one namespace binding, and the 149 genuine call sites are spelled through it. Nothing else moved — this is a binding-name change, as the triage scope ruling requires.

The namespace is spelled nodePath, not path — measured, not stylistic

The obvious spelling is refuted by this file. Measured on the merged tree with a scope-resolving AST pass over every identifier reference (not a grep — the file is full of resolve(/join( inside fixture strings and prose):

namespace candidate	conflicts
path	1
nodePath	0

This file already binds path locally 16 times — 3 block consts and 13 parameters. One of them is load-bearing:

functionreadTrackedSource(path){returnreadFileSync(join(ROOT,path),'utf8');}

Under a path namespace that line becomes path.join(ROOT, path), calling .join on a string parameter — re-creating the exact defect this PR closes, in the same file, on the same day. nodePath occurs nowhere in scripts/.

Why the namespace rather than renaming the two fixtures

Renaming the fixtures fixes today's two instances. The namespace removes the possibility: after this change zero module-scope node:path bindings remain, so there is nothing left for a fixture to shadow. The AST pass confirms the end state — the only surviving bare references to the five names are the 3 deliberate fixture uses, each binding to its own local fixture.

There is a second, structural gain. The two shadowed names were dangerous precisely because no existing case inside selfTest() called them, so the shadow armed a trap for a future case instead of breaking a present one. nodePath is called ~100 times inside selfTest() itself, so any future binding of that name fails immediately, at the moment it is written, rather than lying in wait.

The diff is mechanically reviewable

Every changed line is a prefix insertion. Stripping the inserted prefix from the result reproduces the previous file byte for byte except the import line:

$ sed 's/nodePath\.//g' scripts/pm/dispatch-gates.mjs ...stripped
$ diff previous-file stripped ...one hunk, at line 363:
previous: import { basename, join, dirname, relative, resolve } from 'node:path';
now: import * as nodePath from 'node:path';

That single hunk is the whole semantic content of a 140-line diff. No assertion, fixture value, string literal or comment moved; the file is the same 17,974 lines.

Verification

The battery is the verdict, and it is identical.pnpm check:pm-dispatch-gates before and after, both under scripts/pm/os-verify-lock.sh:

before (base 5bc2f27): ✓ dispatch-gates self-test: 1288 cases pass.
after (head 28943b59): ✓ dispatch-gates self-test: 1288 cases pass.

Stronger than the count: all 1,289 streamed case lines diff byte-identical between the two runs, 0 failures on each side.

Reverse verification, both legs injected at the same program point (immediately after the fixture binding), each mutation proven on disk by occurrence count and object hash, each restored to the HEAD blob with git diff HEAD empty:

leg	tree	probe	reading
A	unfixed base	relative(ROOT, ROOT)	TypeError: relative is not a function
B	this branch	nodePath.relative(ROOT, ROOT)	returned "", and typeof relative is still string

Leg B's second reading is the point worth stating plainly: the fixture is untouched and the bare name is still a string. The defect is closed by removing the module binding that could be shadowed, not by making the bare spelling work.

node --check passes on both trees and is not offered as evidence — by construction it cannot see this defect class.

Gates. Re-derived after the final commit from the actual change set (node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack, no paths): 20 runnable families. 19 ran green; check-test-completeness.mjs exits 3, which is its own NOT MEASURED code — it grades a saved turbo run test log and states "⛔ It is not a red, and there is nothing here to fix."

ESLint narrowed to the changed file, with the invariance evidence: eslint's own --print-config reports 2 active rules for this file, neither type-aware; --format json reports 1 file linted, 0 errors, 0 warnings. No project/projectService is configured anywhere in eslint.config.mjs, so this diff cannot move the verdict of any file it does not touch.

Scope

Binding names only, one file. Out of scope and untouched, as the dispatch fences them: the --tier --residue refusal table (#14753) and the derivation coverage work (#14880) both remain open and queued behind this. #14672 stays exactly as merged — its ROOT-derived workaround is left in place and is still correct; only the comment above it now describes a hazard that no longer exists, which is a follow-up rather than a rider here.

skip-changeset: scripts/pm/** publishes nothing from any package.

---

🤖 Generated with Claude Code

Generated by Claude Code

---

Generated by Claude Code

[claude]

Fixes#14688

selfTest() binds fixture values to two names this module imports from node:path, so a later case reaching for either gets the fixture instead of the function and dies at run time. Parse-time checks cannot see it; the run-time message names a real binding, so it reads like a broken import rather than a shadow.

What changed

The module took five bare names from node:path. It now takes one namespace binding, and the 149 genuine call sites are spelled through it. Nothing else moved — this is a binding-name change, as the triage scope ruling requires.

The namespace is spelled nodePath, not path — measured, not stylistic

The obvious spelling is refuted by this file. Measured on the merged tree with a scope-resolving AST pass over every identifier reference (not a grep — the file is full of resolve(/join( inside fixture strings and prose):

namespace candidate	conflicts
path	1
nodePath	0

This file already binds path locally 16 times — 3 block consts and 13 parameters. One of them is load-bearing:

functionreadTrackedSource(path){returnreadFileSync(join(ROOT,path),'utf8');}

Under a path namespace that line becomes path.join(ROOT, path), calling .join on a string parameter — re-creating the exact defect this PR closes, in the same file, on the same day. nodePath occurs nowhere in scripts/.

Why the namespace rather than renaming the two fixtures

Renaming the fixtures fixes today's two instances. The namespace removes the possibility: after this change zero module-scope node:path bindings remain, so there is nothing left for a fixture to shadow. The AST pass confirms the end state — the only surviving bare references to the five names are the 3 deliberate fixture uses, each binding to its own local fixture.

There is a second, structural gain. The two shadowed names were dangerous precisely because no existing case inside selfTest() called them, so the shadow armed a trap for a future case instead of breaking a present one. nodePath is called ~100 times inside selfTest() itself, so any future binding of that name fails immediately, at the moment it is written, rather than lying in wait.

The diff is mechanically reviewable

Every changed line is a prefix insertion. Stripping the inserted prefix from the result reproduces the previous file byte for byte except the import line:

$ sed 's/nodePath\.//g' scripts/pm/dispatch-gates.mjs ...stripped
$ diff previous-file stripped ...one hunk, at line 363:
previous: import { basename, join, dirname, relative, resolve } from 'node:path';
now: import * as nodePath from 'node:path';

That single hunk is the whole semantic content of a 140-line diff. No assertion, fixture value, string literal or comment moved; the file is the same 17,974 lines.

Verification

The battery is the verdict, and it is identical.pnpm check:pm-dispatch-gates before and after, both under scripts/pm/os-verify-lock.sh:

before (base 5bc2f27): ✓ dispatch-gates self-test: 1288 cases pass.
after (head 28943b59): ✓ dispatch-gates self-test: 1288 cases pass.

Stronger than the count: all 1,289 streamed case lines diff byte-identical between the two runs, 0 failures on each side.

Reverse verification, both legs injected at the same program point (immediately after the fixture binding), each mutation proven on disk by occurrence count and object hash, each restored to the HEAD blob with git diff HEAD empty:

leg	tree	probe	reading
A	unfixed base	relative(ROOT, ROOT)	TypeError: relative is not a function
B	this branch	nodePath.relative(ROOT, ROOT)	returned "", and typeof relative is still string

Leg B's second reading is the point worth stating plainly: the fixture is untouched and the bare name is still a string. The defect is closed by removing the module binding that could be shadowed, not by making the bare spelling work.

node --check passes on both trees and is not offered as evidence — by construction it cannot see this defect class.

Gates. Re-derived after the final commit from the actual change set (node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack, no paths): 20 runnable families. 19 ran green; check-test-completeness.mjs exits 3, which is its own NOT MEASURED code — it grades a saved turbo run test log and states "⛔ It is not a red, and there is nothing here to fix."

ESLint narrowed to the changed file, with the invariance evidence: eslint's own --print-config reports 2 active rules for this file, neither type-aware; --format json reports 1 file linted, 0 errors, 0 warnings. No project/projectService is configured anywhere in eslint.config.mjs, so this diff cannot move the verdict of any file it does not touch.

Scope

Binding names only, one file. Out of scope and untouched, as the dispatch fences them: the --tier --residue refusal table (#14753) and the derivation coverage work (#14880) both remain open and queued behind this. #14672 stays exactly as merged — its ROOT-derived workaround is left in place and is still correct; only the comment above it now describes a hazard that no longer exists, which is a follow-up rather than a rider here.

skip-changeset: scripts/pm/** publishes nothing from any package.

---

🤖 Generated with Claude Code

Generated by Claude Code

---

Generated by Claude Code

[claude]

Fixes#14688

selfTest() binds fixture values to two names this module imports from node:path, so a later case reaching for either gets the fixture instead of the function and dies at run time. Parse-time checks cannot see it; the run-time message names a real binding, so it reads like a broken import rather than a shadow.

What changed

The module took five bare names from node:path. It now takes one namespace binding, and the 149 genuine call sites are spelled through it. Nothing else moved — this is a binding-name change, as the triage scope ruling requires.

The namespace is spelled nodePath, not path — measured, not stylistic

The obvious spelling is refuted by this file. Measured on the merged tree with a scope-resolving AST pass over every identifier reference (not a grep — the file is full of resolve(/join( inside fixture strings and prose):

namespace candidate	conflicts
path	1
nodePath	0

This file already binds path locally 16 times — 3 block consts and 13 parameters. One of them is load-bearing:

functionreadTrackedSource(path){returnreadFileSync(join(ROOT,path),'utf8');}

Under a path namespace that line becomes path.join(ROOT, path), calling .join on a string parameter — re-creating the exact defect this PR closes, in the same file, on the same day. nodePath occurs nowhere in scripts/.

Why the namespace rather than renaming the two fixtures

Renaming the fixtures fixes today's two instances. The namespace removes the possibility: after this change zero module-scope node:path bindings remain, so there is nothing left for a fixture to shadow. The AST pass confirms the end state — the only surviving bare references to the five names are the 3 deliberate fixture uses, each binding to its own local fixture.

There is a second, structural gain. The two shadowed names were dangerous precisely because no existing case inside selfTest() called them, so the shadow armed a trap for a future case instead of breaking a present one. nodePath is called ~100 times inside selfTest() itself, so any future binding of that name fails immediately, at the moment it is written, rather than lying in wait.

The diff is mechanically reviewable

Every changed line is a prefix insertion. Stripping the inserted prefix from the result reproduces the previous file byte for byte except the import line:

$ sed 's/nodePath\.//g' scripts/pm/dispatch-gates.mjs ...stripped
$ diff previous-file stripped ...one hunk, at line 363:
previous: import { basename, join, dirname, relative, resolve } from 'node:path';
now: import * as nodePath from 'node:path';

That single hunk is the whole semantic content of a 140-line diff. No assertion, fixture value, string literal or comment moved; the file is the same 17,974 lines.

Verification

The battery is the verdict, and it is identical.pnpm check:pm-dispatch-gates before and after, both under scripts/pm/os-verify-lock.sh:

before (base 5bc2f27): ✓ dispatch-gates self-test: 1288 cases pass.
after (head 28943b59): ✓ dispatch-gates self-test: 1288 cases pass.

Stronger than the count: all 1,289 streamed case lines diff byte-identical between the two runs, 0 failures on each side.

Reverse verification, both legs injected at the same program point (immediately after the fixture binding), each mutation proven on disk by occurrence count and object hash, each restored to the HEAD blob with git diff HEAD empty:

leg	tree	probe	reading
A	unfixed base	relative(ROOT, ROOT)	TypeError: relative is not a function
B	this branch	nodePath.relative(ROOT, ROOT)	returned "", and typeof relative is still string

Leg B's second reading is the point worth stating plainly: the fixture is untouched and the bare name is still a string. The defect is closed by removing the module binding that could be shadowed, not by making the bare spelling work.

node --check passes on both trees and is not offered as evidence — by construction it cannot see this defect class.

Gates. Re-derived after the final commit from the actual change set (node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack, no paths): 20 runnable families. 19 ran green; check-test-completeness.mjs exits 3, which is its own NOT MEASURED code — it grades a saved turbo run test log and states "⛔ It is not a red, and there is nothing here to fix."

ESLint narrowed to the changed file, with the invariance evidence: eslint's own --print-config reports 2 active rules for this file, neither type-aware; --format json reports 1 file linted, 0 errors, 0 warnings. No project/projectService is configured anywhere in eslint.config.mjs, so this diff cannot move the verdict of any file it does not touch.

Scope

Binding names only, one file. Out of scope and untouched, as the dispatch fences them: the --tier --residue refusal table (#14753) and the derivation coverage work (#14880) both remain open and queued behind this. #14672 stays exactly as merged — its ROOT-derived workaround is left in place and is still correct; only the comment above it now describes a hazard that no longer exists, which is a follow-up rather than a rider here.

skip-changeset: scripts/pm/** publishes nothing from any package.

---

🤖 Generated with Claude Code

Generated by Claude Code

---

Generated by Claude Code

[claude]

Fixes#14688

selfTest() binds fixture values to two names this module imports from node:path, so a later case reaching for either gets the fixture instead of the function and dies at run time. Parse-time checks cannot see it; the run-time message names a real binding, so it reads like a broken import rather than a shadow.

What changed

The module took five bare names from node:path. It now takes one namespace binding, and the 149 genuine call sites are spelled through it. Nothing else moved — this is a binding-name change, as the triage scope ruling requires.

The namespace is spelled nodePath, not path — measured, not stylistic

The obvious spelling is refuted by this file. Measured on the merged tree with a scope-resolving AST pass over every identifier reference (not a grep — the file is full of resolve(/join( inside fixture strings and prose):

namespace candidate	conflicts
path	1
nodePath	0

This file already binds path locally 16 times — 3 block consts and 13 parameters. One of them is load-bearing:

functionreadTrackedSource(path){returnreadFileSync(join(ROOT,path),'utf8');}

Under a path namespace that line becomes path.join(ROOT, path), calling .join on a string parameter — re-creating the exact defect this PR closes, in the same file, on the same day. nodePath occurs nowhere in scripts/.

Why the namespace rather than renaming the two fixtures

Renaming the fixtures fixes today's two instances. The namespace removes the possibility: after this change zero module-scope node:path bindings remain, so there is nothing left for a fixture to shadow. The AST pass confirms the end state — the only surviving bare references to the five names are the 3 deliberate fixture uses, each binding to its own local fixture.

There is a second, structural gain. The two shadowed names were dangerous precisely because no existing case inside selfTest() called them, so the shadow armed a trap for a future case instead of breaking a present one. nodePath is called ~100 times inside selfTest() itself, so any future binding of that name fails immediately, at the moment it is written, rather than lying in wait.

The diff is mechanically reviewable

Every changed line is a prefix insertion. Stripping the inserted prefix from the result reproduces the previous file byte for byte except the import line:

$ sed 's/nodePath\.//g' scripts/pm/dispatch-gates.mjs ...stripped
$ diff previous-file stripped ...one hunk, at line 363:
previous: import { basename, join, dirname, relative, resolve } from 'node:path';
now: import * as nodePath from 'node:path';

That single hunk is the whole semantic content of a 140-line diff. No assertion, fixture value, string literal or comment moved; the file is the same 17,974 lines.

Verification

The battery is the verdict, and it is identical.pnpm check:pm-dispatch-gates before and after, both under scripts/pm/os-verify-lock.sh:

before (base 5bc2f27): ✓ dispatch-gates self-test: 1288 cases pass.
after (head 28943b59): ✓ dispatch-gates self-test: 1288 cases pass.

Stronger than the count: all 1,289 streamed case lines diff byte-identical between the two runs, 0 failures on each side.

Reverse verification, both legs injected at the same program point (immediately after the fixture binding), each mutation proven on disk by occurrence count and object hash, each restored to the HEAD blob with git diff HEAD empty:

leg	tree	probe	reading
A	unfixed base	relative(ROOT, ROOT)	TypeError: relative is not a function
B	this branch	nodePath.relative(ROOT, ROOT)	returned "", and typeof relative is still string

Leg B's second reading is the point worth stating plainly: the fixture is untouched and the bare name is still a string. The defect is closed by removing the module binding that could be shadowed, not by making the bare spelling work.

node --check passes on both trees and is not offered as evidence — by construction it cannot see this defect class.

Gates. Re-derived after the final commit from the actual change set (node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack, no paths): 20 runnable families. 19 ran green; check-test-completeness.mjs exits 3, which is its own NOT MEASURED code — it grades a saved turbo run test log and states "⛔ It is not a red, and there is nothing here to fix."

ESLint narrowed to the changed file, with the invariance evidence: eslint's own --print-config reports 2 active rules for this file, neither type-aware; --format json reports 1 file linted, 0 errors, 0 warnings. No project/projectService is configured anywhere in eslint.config.mjs, so this diff cannot move the verdict of any file it does not touch.

Scope

Binding names only, one file. Out of scope and untouched, as the dispatch fences them: the --tier --residue refusal table (#14753) and the derivation coverage work (#14880) both remain open and queued behind this. #14672 stays exactly as merged — its ROOT-derived workaround is left in place and is still correct; only the comment above it now describes a hazard that no longer exists, which is a follow-up rather than a rider here.

skip-changeset: scripts/pm/** publishes nothing from any package.

---

🤖 Generated with Claude Code

Generated by Claude Code

---

Generated by Claude Code

[claude]

Fixes#14688

selfTest() binds fixture values to two names this module imports from node:path, so a later case reaching for either gets the fixture instead of the function and dies at run time. Parse-time checks cannot see it; the run-time message names a real binding, so it reads like a broken import rather than a shadow.

What changed

The module took five bare names from node:path. It now takes one namespace binding, and the 149 genuine call sites are spelled through it. Nothing else moved — this is a binding-name change, as the triage scope ruling requires.

The namespace is spelled nodePath, not path — measured, not stylistic

The obvious spelling is refuted by this file. Measured on the merged tree with a scope-resolving AST pass over every identifier reference (not a grep — the file is full of resolve(/join( inside fixture strings and prose):

namespace candidate	conflicts
path	1
nodePath	0

This file already binds path locally 16 times — 3 block consts and 13 parameters. One of them is load-bearing:

functionreadTrackedSource(path){returnreadFileSync(join(ROOT,path),'utf8');}

Under a path namespace that line becomes path.join(ROOT, path), calling .join on a string parameter — re-creating the exact defect this PR closes, in the same file, on the same day. nodePath occurs nowhere in scripts/.

Why the namespace rather than renaming the two fixtures

Renaming the fixtures fixes today's two instances. The namespace removes the possibility: after this change zero module-scope node:path bindings remain, so there is nothing left for a fixture to shadow. The AST pass confirms the end state — the only surviving bare references to the five names are the 3 deliberate fixture uses, each binding to its own local fixture.

There is a second, structural gain. The two shadowed names were dangerous precisely because no existing case inside selfTest() called them, so the shadow armed a trap for a future case instead of breaking a present one. nodePath is called ~100 times inside selfTest() itself, so any future binding of that name fails immediately, at the moment it is written, rather than lying in wait.

The diff is mechanically reviewable

Every changed line is a prefix insertion. Stripping the inserted prefix from the result reproduces the previous file byte for byte except the import line:

$ sed 's/nodePath\.//g' scripts/pm/dispatch-gates.mjs ...stripped
$ diff previous-file stripped ...one hunk, at line 363:
previous: import { basename, join, dirname, relative, resolve } from 'node:path';
now: import * as nodePath from 'node:path';

That single hunk is the whole semantic content of a 140-line diff. No assertion, fixture value, string literal or comment moved; the file is the same 17,974 lines.

Verification

The battery is the verdict, and it is identical.pnpm check:pm-dispatch-gates before and after, both under scripts/pm/os-verify-lock.sh:

before (base 5bc2f27): ✓ dispatch-gates self-test: 1288 cases pass.
after (head 28943b59): ✓ dispatch-gates self-test: 1288 cases pass.

Stronger than the count: all 1,289 streamed case lines diff byte-identical between the two runs, 0 failures on each side.

Reverse verification, both legs injected at the same program point (immediately after the fixture binding), each mutation proven on disk by occurrence count and object hash, each restored to the HEAD blob with git diff HEAD empty:

leg	tree	probe	reading
A	unfixed base	relative(ROOT, ROOT)	TypeError: relative is not a function
B	this branch	nodePath.relative(ROOT, ROOT)	returned "", and typeof relative is still string

Leg B's second reading is the point worth stating plainly: the fixture is untouched and the bare name is still a string. The defect is closed by removing the module binding that could be shadowed, not by making the bare spelling work.

node --check passes on both trees and is not offered as evidence — by construction it cannot see this defect class.

Gates. Re-derived after the final commit from the actual change set (node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack, no paths): 20 runnable families. 19 ran green; check-test-completeness.mjs exits 3, which is its own NOT MEASURED code — it grades a saved turbo run test log and states "⛔ It is not a red, and there is nothing here to fix."

ESLint narrowed to the changed file, with the invariance evidence: eslint's own --print-config reports 2 active rules for this file, neither type-aware; --format json reports 1 file linted, 0 errors, 0 warnings. No project/projectService is configured anywhere in eslint.config.mjs, so this diff cannot move the verdict of any file it does not touch.

Scope

Binding names only, one file. Out of scope and untouched, as the dispatch fences them: the --tier --residue refusal table (#14753) and the derivation coverage work (#14880) both remain open and queued behind this. #14672 stays exactly as merged — its ROOT-derived workaround is left in place and is still correct; only the comment above it now describes a hazard that no longer exists, which is a follow-up rather than a rider here.

skip-changeset: scripts/pm/** publishes nothing from any package.

---

🤖 Generated with Claude Code

Generated by Claude Code

---

Generated by Claude Code

[claude]

Fixes#14688

selfTest() binds fixture values to two names this module imports from node:path, so a later case reaching for either gets the fixture instead of the function and dies at run time. Parse-time checks cannot see it; the run-time message names a real binding, so it reads like a broken import rather than a shadow.

What changed

The module took five bare names from node:path. It now takes one namespace binding, and the 149 genuine call sites are spelled through it. Nothing else moved — this is a binding-name change, as the triage scope ruling requires.

The namespace is spelled nodePath, not path — measured, not stylistic

The obvious spelling is refuted by this file. Measured on the merged tree with a scope-resolving AST pass over every identifier reference (not a grep — the file is full of resolve(/join( inside fixture strings and prose):

namespace candidate	conflicts
path	1
nodePath	0

This file already binds path locally 16 times — 3 block consts and 13 parameters. One of them is load-bearing:

functionreadTrackedSource(path){returnreadFileSync(join(ROOT,path),'utf8');}

Under a path namespace that line becomes path.join(ROOT, path), calling .join on a string parameter — re-creating the exact defect this PR closes, in the same file, on the same day. nodePath occurs nowhere in scripts/.

Why the namespace rather than renaming the two fixtures

Renaming the fixtures fixes today's two instances. The namespace removes the possibility: after this change zero module-scope node:path bindings remain, so there is nothing left for a fixture to shadow. The AST pass confirms the end state — the only surviving bare references to the five names are the 3 deliberate fixture uses, each binding to its own local fixture.

There is a second, structural gain. The two shadowed names were dangerous precisely because no existing case inside selfTest() called them, so the shadow armed a trap for a future case instead of breaking a present one. nodePath is called ~100 times inside selfTest() itself, so any future binding of that name fails immediately, at the moment it is written, rather than lying in wait.

The diff is mechanically reviewable

Every changed line is a prefix insertion. Stripping the inserted prefix from the result reproduces the previous file byte for byte except the import line:

$ sed 's/nodePath\.//g' scripts/pm/dispatch-gates.mjs ...stripped
$ diff previous-file stripped ...one hunk, at line 363:
previous: import { basename, join, dirname, relative, resolve } from 'node:path';
now: import * as nodePath from 'node:path';

That single hunk is the whole semantic content of a 140-line diff. No assertion, fixture value, string literal or comment moved; the file is the same 17,974 lines.

Verification

The battery is the verdict, and it is identical.pnpm check:pm-dispatch-gates before and after, both under scripts/pm/os-verify-lock.sh:

before (base 5bc2f27): ✓ dispatch-gates self-test: 1288 cases pass.
after (head 28943b59): ✓ dispatch-gates self-test: 1288 cases pass.

Stronger than the count: all 1,289 streamed case lines diff byte-identical between the two runs, 0 failures on each side.

Reverse verification, both legs injected at the same program point (immediately after the fixture binding), each mutation proven on disk by occurrence count and object hash, each restored to the HEAD blob with git diff HEAD empty:

leg	tree	probe	reading
A	unfixed base	relative(ROOT, ROOT)	TypeError: relative is not a function
B	this branch	nodePath.relative(ROOT, ROOT)	returned "", and typeof relative is still string

Leg B's second reading is the point worth stating plainly: the fixture is untouched and the bare name is still a string. The defect is closed by removing the module binding that could be shadowed, not by making the bare spelling work.

node --check passes on both trees and is not offered as evidence — by construction it cannot see this defect class.

Gates. Re-derived after the final commit from the actual change set (node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack, no paths): 20 runnable families. 19 ran green; check-test-completeness.mjs exits 3, which is its own NOT MEASURED code — it grades a saved turbo run test log and states "⛔ It is not a red, and there is nothing here to fix."

ESLint narrowed to the changed file, with the invariance evidence: eslint's own --print-config reports 2 active rules for this file, neither type-aware; --format json reports 1 file linted, 0 errors, 0 warnings. No project/projectService is configured anywhere in eslint.config.mjs, so this diff cannot move the verdict of any file it does not touch.

Scope

Binding names only, one file. Out of scope and untouched, as the dispatch fences them: the --tier --residue refusal table (#14753) and the derivation coverage work (#14880) both remain open and queued behind this. #14672 stays exactly as merged — its ROOT-derived workaround is left in place and is still correct; only the comment above it now describes a hazard that no longer exists, which is a follow-up rather than a rider here.

skip-changeset: scripts/pm/** publishes nothing from any package.

---

🤖 Generated with Claude Code

Generated by Claude Code

---

Generated by Claude Code

[claude]

Fixes#14688

selfTest() binds fixture values to two names this module imports from node:path, so a later case reaching for either gets the fixture instead of the function and dies at run time. Parse-time checks cannot see it; the run-time message names a real binding, so it reads like a broken import rather than a shadow.

What changed

The module took five bare names from node:path. It now takes one namespace binding, and the 149 genuine call sites are spelled through it. Nothing else moved — this is a binding-name change, as the triage scope ruling requires.

The namespace is spelled nodePath, not path — measured, not stylistic

The obvious spelling is refuted by this file. Measured on the merged tree with a scope-resolving AST pass over every identifier reference (not a grep — the file is full of resolve(/join( inside fixture strings and prose):

namespace candidate	conflicts
path	1
nodePath	0

This file already binds path locally 16 times — 3 block consts and 13 parameters. One of them is load-bearing:

functionreadTrackedSource(path){returnreadFileSync(join(ROOT,path),'utf8');}

Under a path namespace that line becomes path.join(ROOT, path), calling .join on a string parameter — re-creating the exact defect this PR closes, in the same file, on the same day. nodePath occurs nowhere in scripts/.

Why the namespace rather than renaming the two fixtures

Renaming the fixtures fixes today's two instances. The namespace removes the possibility: after this change zero module-scope node:path bindings remain, so there is nothing left for a fixture to shadow. The AST pass confirms the end state — the only surviving bare references to the five names are the 3 deliberate fixture uses, each binding to its own local fixture.

There is a second, structural gain. The two shadowed names were dangerous precisely because no existing case inside selfTest() called them, so the shadow armed a trap for a future case instead of breaking a present one. nodePath is called ~100 times inside selfTest() itself, so any future binding of that name fails immediately, at the moment it is written, rather than lying in wait.

The diff is mechanically reviewable

Every changed line is a prefix insertion. Stripping the inserted prefix from the result reproduces the previous file byte for byte except the import line:

$ sed 's/nodePath\.//g' scripts/pm/dispatch-gates.mjs ...stripped
$ diff previous-file stripped ...one hunk, at line 363:
previous: import { basename, join, dirname, relative, resolve } from 'node:path';
now: import * as nodePath from 'node:path';

That single hunk is the whole semantic content of a 140-line diff. No assertion, fixture value, string literal or comment moved; the file is the same 17,974 lines.

Verification

The battery is the verdict, and it is identical.pnpm check:pm-dispatch-gates before and after, both under scripts/pm/os-verify-lock.sh:

before (base 5bc2f27): ✓ dispatch-gates self-test: 1288 cases pass.
after (head 28943b59): ✓ dispatch-gates self-test: 1288 cases pass.

Stronger than the count: all 1,289 streamed case lines diff byte-identical between the two runs, 0 failures on each side.

Reverse verification, both legs injected at the same program point (immediately after the fixture binding), each mutation proven on disk by occurrence count and object hash, each restored to the HEAD blob with git diff HEAD empty:

leg	tree	probe	reading
A	unfixed base	relative(ROOT, ROOT)	TypeError: relative is not a function
B	this branch	nodePath.relative(ROOT, ROOT)	returned "", and typeof relative is still string

Leg B's second reading is the point worth stating plainly: the fixture is untouched and the bare name is still a string. The defect is closed by removing the module binding that could be shadowed, not by making the bare spelling work.

node --check passes on both trees and is not offered as evidence — by construction it cannot see this defect class.

Gates. Re-derived after the final commit from the actual change set (node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack, no paths): 20 runnable families. 19 ran green; check-test-completeness.mjs exits 3, which is its own NOT MEASURED code — it grades a saved turbo run test log and states "⛔ It is not a red, and there is nothing here to fix."

ESLint narrowed to the changed file, with the invariance evidence: eslint's own --print-config reports 2 active rules for this file, neither type-aware; --format json reports 1 file linted, 0 errors, 0 warnings. No project/projectService is configured anywhere in eslint.config.mjs, so this diff cannot move the verdict of any file it does not touch.

Scope

Binding names only, one file. Out of scope and untouched, as the dispatch fences them: the --tier --residue refusal table (#14753) and the derivation coverage work (#14880) both remain open and queued behind this. #14672 stays exactly as merged — its ROOT-derived workaround is left in place and is still correct; only the comment above it now describes a hazard that no longer exists, which is a follow-up rather than a rider here.

skip-changeset: scripts/pm/** publishes nothing from any package.

---

🤖 Generated with Claude Code

Generated by Claude Code

---

Generated by Claude Code

[claude]

Fixes#14688

selfTest() binds fixture values to two names this module imports from node:path, so a later case reaching for either gets the fixture instead of the function and dies at run time. Parse-time checks cannot see it; the run-time message names a real binding, so it reads like a broken import rather than a shadow.

What changed

The module took five bare names from node:path. It now takes one namespace binding, and the 149 genuine call sites are spelled through it. Nothing else moved — this is a binding-name change, as the triage scope ruling requires.

The namespace is spelled nodePath, not path — measured, not stylistic

The obvious spelling is refuted by this file. Measured on the merged tree with a scope-resolving AST pass over every identifier reference (not a grep — the file is full of resolve(/join( inside fixture strings and prose):

namespace candidate	conflicts
path	1
nodePath	0

This file already binds path locally 16 times — 3 block consts and 13 parameters. One of them is load-bearing:

functionreadTrackedSource(path){returnreadFileSync(join(ROOT,path),'utf8');}

Under a path namespace that line becomes path.join(ROOT, path), calling .join on a string parameter — re-creating the exact defect this PR closes, in the same file, on the same day. nodePath occurs nowhere in scripts/.

Why the namespace rather than renaming the two fixtures

Renaming the fixtures fixes today's two instances. The namespace removes the possibility: after this change zero module-scope node:path bindings remain, so there is nothing left for a fixture to shadow. The AST pass confirms the end state — the only surviving bare references to the five names are the 3 deliberate fixture uses, each binding to its own local fixture.

There is a second, structural gain. The two shadowed names were dangerous precisely because no existing case inside selfTest() called them, so the shadow armed a trap for a future case instead of breaking a present one. nodePath is called ~100 times inside selfTest() itself, so any future binding of that name fails immediately, at the moment it is written, rather than lying in wait.

The diff is mechanically reviewable

Every changed line is a prefix insertion. Stripping the inserted prefix from the result reproduces the previous file byte for byte except the import line:

$ sed 's/nodePath\.//g' scripts/pm/dispatch-gates.mjs ...stripped
$ diff previous-file stripped ...one hunk, at line 363:
previous: import { basename, join, dirname, relative, resolve } from 'node:path';
now: import * as nodePath from 'node:path';

That single hunk is the whole semantic content of a 140-line diff. No assertion, fixture value, string literal or comment moved; the file is the same 17,974 lines.

Verification

The battery is the verdict, and it is identical.pnpm check:pm-dispatch-gates before and after, both under scripts/pm/os-verify-lock.sh:

before (base 5bc2f27): ✓ dispatch-gates self-test: 1288 cases pass.
after (head 28943b59): ✓ dispatch-gates self-test: 1288 cases pass.

Stronger than the count: all 1,289 streamed case lines diff byte-identical between the two runs, 0 failures on each side.

Reverse verification, both legs injected at the same program point (immediately after the fixture binding), each mutation proven on disk by occurrence count and object hash, each restored to the HEAD blob with git diff HEAD empty:

leg	tree	probe	reading
A	unfixed base	relative(ROOT, ROOT)	TypeError: relative is not a function
B	this branch	nodePath.relative(ROOT, ROOT)	returned "", and typeof relative is still string

Leg B's second reading is the point worth stating plainly: the fixture is untouched and the bare name is still a string. The defect is closed by removing the module binding that could be shadowed, not by making the bare spelling work.

node --check passes on both trees and is not offered as evidence — by construction it cannot see this defect class.

Gates. Re-derived after the final commit from the actual change set (node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack, no paths): 20 runnable families. 19 ran green; check-test-completeness.mjs exits 3, which is its own NOT MEASURED code — it grades a saved turbo run test log and states "⛔ It is not a red, and there is nothing here to fix."

ESLint narrowed to the changed file, with the invariance evidence: eslint's own --print-config reports 2 active rules for this file, neither type-aware; --format json reports 1 file linted, 0 errors, 0 warnings. No project/projectService is configured anywhere in eslint.config.mjs, so this diff cannot move the verdict of any file it does not touch.

Scope

Binding names only, one file. Out of scope and untouched, as the dispatch fences them: the --tier --residue refusal table (#14753) and the derivation coverage work (#14880) both remain open and queued behind this. #14672 stays exactly as merged — its ROOT-derived workaround is left in place and is still correct; only the comment above it now describes a hazard that no longer exists, which is a follow-up rather than a rider here.

skip-changeset: scripts/pm/** publishes nothing from any package.

---

🤖 Generated with Claude Code

Generated by Claude Code

---

Generated by Claude Code