docs(adr): fix ADR-011 module table and pin it with a contract test by Chemaclass · Pull Request #964 · TypedDevs/bashunit

Chemaclass · 2026-08-02T09:24:07Z

🤔 Background

ADR-011's module table was wrong in two places. Both errors came from counting with a one-level glob instead of from the tree.

💡 Changes

doubles/ was missing outright. The table is ordered by load order, so it was generated from the entrypoint's source lines — and doubles/ is the one module the entrypoint doesn't source (assert/index.sh does). It fell through silently while the prose directly above said "Seventeen", a number that came from counting directories. The two disagreed and nothing noticed. It's now row 12, where it actually loads, and the row records where it's sourced from.

learn/ was listed at 5 files / 240 lines. It's 14 / 1296. The difference is learn/lessons/, which src/learn/*.sh doesn't descend into — the same one-level-glob mistake Makefile:67 made with nested tests. Other rows had drifted 2-3 lines from recent comment edits; refreshed too.

docs/project-overview.md listed 15 of 17 modules, missing doubles and dev. Spies and mocks are a headline feature to be absent from the source-layout overview.

🧪 The contract test

Asserts every src/ module has a row, and that the Files column sums to 118 — the tracked .sh count. Summing catches an under-count in any single row without pinning every number exactly, which would make the table tedious to maintain for no extra safety.

Mutation-tested, not assumed:

Mutation	Result
Delete the `doubles/` row	2 assertions red
`learn/` files 14 → 5	1 assertion red
Rename the ADR away	all 3 red

That last row is why the file-exists assertion exists — without it a rename would leave the greps matching nothing and the whole file passing while checking nothing, exactly how a layering contract rotted unnoticed in #946.

✅ Verification

make sa · make lint · bash build.sh bin -v → ✅ Build verified ✅. 1633 sequential / 1592 parallel-simple-strict — baseline + the 3 new tests.

ADR-011's module table was wrong in two places, and both errors came from counting with a one-level glob instead of from the tree. `doubles/` was missing outright. The table is ordered by load order, so it was generated from the entrypoint's `source` lines -- and `doubles/` is the one module the entrypoint does not source, `assert/index.sh` does. It fell through silently while the prose directly above said "Seventeen", a number that came from counting directories. The two disagreed and nothing noticed. The module is now row 12, where it actually loads, and the row says where it is sourced from so the next reader does not have to rediscover the discrepancy. `learn/` was listed at 5 files / 240 lines. It is 14 files / 1296. The difference is `learn/lessons/`, which `src/learn/*.sh` does not descend into -- the same one-level-glob mistake `Makefile:67` made with nested tests. Every other row's line count had drifted by two or three from recent comment edits; those are refreshed too. `docs/project-overview.md` listed 15 of the 17 modules, missing `doubles` and `dev`. Spies and mocks are a headline feature to be absent from the overview of the source layout. The new test asserts every `src/` module has a row, and that the Files column sums to the number of tracked `.sh` files in `src/` -- 118. Summing catches an under-count in any single row without pinning every number to an exact value, which would make the table annoying to maintain for no extra safety. Mutation-tested rather than assumed: deleting the `doubles/` row turns two assertions red, changing `learn/` from 14 to 5 turns one red, and renaming the ADR turns all three red. That last one is why the file-exists assertion is there -- without it a rename would leave the greps matching nothing and the whole file passing while checking nothing, which is exactly how a layering contract rotted unnoticed in #946. 1633 sequential / 1592 parallel; both are baseline + the 3 new tests.

… the ADR contract The Bash 3.0 jobs run in a container where git refuses to read the repo (dubious-ownership), so `git ls-files` exited non-zero and produced nothing. Both count-based assertions then compared against an empty listing and failed for a reason unrelated to the ADR. `find` needs no git and no index.

Chemaclass · 2026-08-02T09:24:07Z

🤔 Background

ADR-011's module table was wrong in two places. Both errors came from counting with a one-level glob instead of from the tree.

💡 Changes

doubles/ was missing outright. The table is ordered by load order, so it was generated from the entrypoint's source lines — and doubles/ is the one module the entrypoint doesn't source (assert/index.sh does). It fell through silently while the prose directly above said "Seventeen", a number that came from counting directories. The two disagreed and nothing noticed. It's now row 12, where it actually loads, and the row records where it's sourced from.

learn/ was listed at 5 files / 240 lines. It's 14 / 1296. The difference is learn/lessons/, which src/learn/*.sh doesn't descend into — the same one-level-glob mistake Makefile:67 made with nested tests. Other rows had drifted 2-3 lines from recent comment edits; refreshed too.

docs/project-overview.md listed 15 of 17 modules, missing doubles and dev. Spies and mocks are a headline feature to be absent from the source-layout overview.

🧪 The contract test

Asserts every src/ module has a row, and that the Files column sums to 118 — the tracked .sh count. Summing catches an under-count in any single row without pinning every number exactly, which would make the table tedious to maintain for no extra safety.

Mutation-tested, not assumed:

Mutation	Result
Delete the `doubles/` row	2 assertions red
`learn/` files 14 → 5	1 assertion red
Rename the ADR away	all 3 red

That last row is why the file-exists assertion exists — without it a rename would leave the greps matching nothing and the whole file passing while checking nothing, exactly how a layering contract rotted unnoticed in #946.

✅ Verification

make sa · make lint · bash build.sh bin -v → ✅ Build verified ✅. 1633 sequential / 1592 parallel-simple-strict — baseline + the 3 new tests.

ADR-011's module table was wrong in two places, and both errors came from counting with a one-level glob instead of from the tree. `doubles/` was missing outright. The table is ordered by load order, so it was generated from the entrypoint's `source` lines -- and `doubles/` is the one module the entrypoint does not source, `assert/index.sh` does. It fell through silently while the prose directly above said "Seventeen", a number that came from counting directories. The two disagreed and nothing noticed. The module is now row 12, where it actually loads, and the row says where it is sourced from so the next reader does not have to rediscover the discrepancy. `learn/` was listed at 5 files / 240 lines. It is 14 files / 1296. The difference is `learn/lessons/`, which `src/learn/*.sh` does not descend into -- the same one-level-glob mistake `Makefile:67` made with nested tests. Every other row's line count had drifted by two or three from recent comment edits; those are refreshed too. `docs/project-overview.md` listed 15 of the 17 modules, missing `doubles` and `dev`. Spies and mocks are a headline feature to be absent from the overview of the source layout. The new test asserts every `src/` module has a row, and that the Files column sums to the number of tracked `.sh` files in `src/` -- 118. Summing catches an under-count in any single row without pinning every number to an exact value, which would make the table annoying to maintain for no extra safety. Mutation-tested rather than assumed: deleting the `doubles/` row turns two assertions red, changing `learn/` from 14 to 5 turns one red, and renaming the ADR turns all three red. That last one is why the file-exists assertion is there -- without it a rename would leave the greps matching nothing and the whole file passing while checking nothing, which is exactly how a layering contract rotted unnoticed in #946. 1633 sequential / 1592 parallel; both are baseline + the 3 new tests.

… the ADR contract The Bash 3.0 jobs run in a container where git refuses to read the repo (dubious-ownership), so `git ls-files` exited non-zero and produced nothing. Both count-based assertions then compared against an empty listing and failed for a reason unrelated to the ADR. `find` needs no git and no index.

Chemaclass · 2026-08-02T09:24:07Z

🤔 Background

ADR-011's module table was wrong in two places. Both errors came from counting with a one-level glob instead of from the tree.

💡 Changes

doubles/ was missing outright. The table is ordered by load order, so it was generated from the entrypoint's source lines — and doubles/ is the one module the entrypoint doesn't source (assert/index.sh does). It fell through silently while the prose directly above said "Seventeen", a number that came from counting directories. The two disagreed and nothing noticed. It's now row 12, where it actually loads, and the row records where it's sourced from.

learn/ was listed at 5 files / 240 lines. It's 14 / 1296. The difference is learn/lessons/, which src/learn/*.sh doesn't descend into — the same one-level-glob mistake Makefile:67 made with nested tests. Other rows had drifted 2-3 lines from recent comment edits; refreshed too.

docs/project-overview.md listed 15 of 17 modules, missing doubles and dev. Spies and mocks are a headline feature to be absent from the source-layout overview.

🧪 The contract test

Asserts every src/ module has a row, and that the Files column sums to 118 — the tracked .sh count. Summing catches an under-count in any single row without pinning every number exactly, which would make the table tedious to maintain for no extra safety.

Mutation-tested, not assumed:

Mutation	Result
Delete the `doubles/` row	2 assertions red
`learn/` files 14 → 5	1 assertion red
Rename the ADR away	all 3 red

That last row is why the file-exists assertion exists — without it a rename would leave the greps matching nothing and the whole file passing while checking nothing, exactly how a layering contract rotted unnoticed in #946.

✅ Verification

make sa · make lint · bash build.sh bin -v → ✅ Build verified ✅. 1633 sequential / 1592 parallel-simple-strict — baseline + the 3 new tests.

ADR-011's module table was wrong in two places, and both errors came from counting with a one-level glob instead of from the tree. `doubles/` was missing outright. The table is ordered by load order, so it was generated from the entrypoint's `source` lines -- and `doubles/` is the one module the entrypoint does not source, `assert/index.sh` does. It fell through silently while the prose directly above said "Seventeen", a number that came from counting directories. The two disagreed and nothing noticed. The module is now row 12, where it actually loads, and the row says where it is sourced from so the next reader does not have to rediscover the discrepancy. `learn/` was listed at 5 files / 240 lines. It is 14 files / 1296. The difference is `learn/lessons/`, which `src/learn/*.sh` does not descend into -- the same one-level-glob mistake `Makefile:67` made with nested tests. Every other row's line count had drifted by two or three from recent comment edits; those are refreshed too. `docs/project-overview.md` listed 15 of the 17 modules, missing `doubles` and `dev`. Spies and mocks are a headline feature to be absent from the overview of the source layout. The new test asserts every `src/` module has a row, and that the Files column sums to the number of tracked `.sh` files in `src/` -- 118. Summing catches an under-count in any single row without pinning every number to an exact value, which would make the table annoying to maintain for no extra safety. Mutation-tested rather than assumed: deleting the `doubles/` row turns two assertions red, changing `learn/` from 14 to 5 turns one red, and renaming the ADR turns all three red. That last one is why the file-exists assertion is there -- without it a rename would leave the greps matching nothing and the whole file passing while checking nothing, which is exactly how a layering contract rotted unnoticed in #946. 1633 sequential / 1592 parallel; both are baseline + the 3 new tests.

… the ADR contract The Bash 3.0 jobs run in a container where git refuses to read the repo (dubious-ownership), so `git ls-files` exited non-zero and produced nothing. Both count-based assertions then compared against an empty listing and failed for a reason unrelated to the ADR. `find` needs no git and no index.

Chemaclass · 2026-08-02T09:24:07Z

🤔 Background

ADR-011's module table was wrong in two places. Both errors came from counting with a one-level glob instead of from the tree.

💡 Changes

doubles/ was missing outright. The table is ordered by load order, so it was generated from the entrypoint's source lines — and doubles/ is the one module the entrypoint doesn't source (assert/index.sh does). It fell through silently while the prose directly above said "Seventeen", a number that came from counting directories. The two disagreed and nothing noticed. It's now row 12, where it actually loads, and the row records where it's sourced from.

learn/ was listed at 5 files / 240 lines. It's 14 / 1296. The difference is learn/lessons/, which src/learn/*.sh doesn't descend into — the same one-level-glob mistake Makefile:67 made with nested tests. Other rows had drifted 2-3 lines from recent comment edits; refreshed too.

docs/project-overview.md listed 15 of 17 modules, missing doubles and dev. Spies and mocks are a headline feature to be absent from the source-layout overview.

🧪 The contract test

Asserts every src/ module has a row, and that the Files column sums to 118 — the tracked .sh count. Summing catches an under-count in any single row without pinning every number exactly, which would make the table tedious to maintain for no extra safety.

Mutation-tested, not assumed:

Mutation	Result
Delete the `doubles/` row	2 assertions red
`learn/` files 14 → 5	1 assertion red
Rename the ADR away	all 3 red

That last row is why the file-exists assertion exists — without it a rename would leave the greps matching nothing and the whole file passing while checking nothing, exactly how a layering contract rotted unnoticed in #946.

✅ Verification

make sa · make lint · bash build.sh bin -v → ✅ Build verified ✅. 1633 sequential / 1592 parallel-simple-strict — baseline + the 3 new tests.

ADR-011's module table was wrong in two places, and both errors came from counting with a one-level glob instead of from the tree. `doubles/` was missing outright. The table is ordered by load order, so it was generated from the entrypoint's `source` lines -- and `doubles/` is the one module the entrypoint does not source, `assert/index.sh` does. It fell through silently while the prose directly above said "Seventeen", a number that came from counting directories. The two disagreed and nothing noticed. The module is now row 12, where it actually loads, and the row says where it is sourced from so the next reader does not have to rediscover the discrepancy. `learn/` was listed at 5 files / 240 lines. It is 14 files / 1296. The difference is `learn/lessons/`, which `src/learn/*.sh` does not descend into -- the same one-level-glob mistake `Makefile:67` made with nested tests. Every other row's line count had drifted by two or three from recent comment edits; those are refreshed too. `docs/project-overview.md` listed 15 of the 17 modules, missing `doubles` and `dev`. Spies and mocks are a headline feature to be absent from the overview of the source layout. The new test asserts every `src/` module has a row, and that the Files column sums to the number of tracked `.sh` files in `src/` -- 118. Summing catches an under-count in any single row without pinning every number to an exact value, which would make the table annoying to maintain for no extra safety. Mutation-tested rather than assumed: deleting the `doubles/` row turns two assertions red, changing `learn/` from 14 to 5 turns one red, and renaming the ADR turns all three red. That last one is why the file-exists assertion is there -- without it a rename would leave the greps matching nothing and the whole file passing while checking nothing, which is exactly how a layering contract rotted unnoticed in #946. 1633 sequential / 1592 parallel; both are baseline + the 3 new tests.

… the ADR contract The Bash 3.0 jobs run in a container where git refuses to read the repo (dubious-ownership), so `git ls-files` exited non-zero and produced nothing. Both count-based assertions then compared against an empty listing and failed for a reason unrelated to the ADR. `find` needs no git and no index.

Chemaclass · 2026-08-02T09:24:07Z

🤔 Background

ADR-011's module table was wrong in two places. Both errors came from counting with a one-level glob instead of from the tree.

💡 Changes

doubles/ was missing outright. The table is ordered by load order, so it was generated from the entrypoint's source lines — and doubles/ is the one module the entrypoint doesn't source (assert/index.sh does). It fell through silently while the prose directly above said "Seventeen", a number that came from counting directories. The two disagreed and nothing noticed. It's now row 12, where it actually loads, and the row records where it's sourced from.

learn/ was listed at 5 files / 240 lines. It's 14 / 1296. The difference is learn/lessons/, which src/learn/*.sh doesn't descend into — the same one-level-glob mistake Makefile:67 made with nested tests. Other rows had drifted 2-3 lines from recent comment edits; refreshed too.

docs/project-overview.md listed 15 of 17 modules, missing doubles and dev. Spies and mocks are a headline feature to be absent from the source-layout overview.

🧪 The contract test

Asserts every src/ module has a row, and that the Files column sums to 118 — the tracked .sh count. Summing catches an under-count in any single row without pinning every number exactly, which would make the table tedious to maintain for no extra safety.

Mutation-tested, not assumed:

Mutation	Result
Delete the `doubles/` row	2 assertions red
`learn/` files 14 → 5	1 assertion red
Rename the ADR away	all 3 red

That last row is why the file-exists assertion exists — without it a rename would leave the greps matching nothing and the whole file passing while checking nothing, exactly how a layering contract rotted unnoticed in #946.

✅ Verification

make sa · make lint · bash build.sh bin -v → ✅ Build verified ✅. 1633 sequential / 1592 parallel-simple-strict — baseline + the 3 new tests.

ADR-011's module table was wrong in two places, and both errors came from counting with a one-level glob instead of from the tree. `doubles/` was missing outright. The table is ordered by load order, so it was generated from the entrypoint's `source` lines -- and `doubles/` is the one module the entrypoint does not source, `assert/index.sh` does. It fell through silently while the prose directly above said "Seventeen", a number that came from counting directories. The two disagreed and nothing noticed. The module is now row 12, where it actually loads, and the row says where it is sourced from so the next reader does not have to rediscover the discrepancy. `learn/` was listed at 5 files / 240 lines. It is 14 files / 1296. The difference is `learn/lessons/`, which `src/learn/*.sh` does not descend into -- the same one-level-glob mistake `Makefile:67` made with nested tests. Every other row's line count had drifted by two or three from recent comment edits; those are refreshed too. `docs/project-overview.md` listed 15 of the 17 modules, missing `doubles` and `dev`. Spies and mocks are a headline feature to be absent from the overview of the source layout. The new test asserts every `src/` module has a row, and that the Files column sums to the number of tracked `.sh` files in `src/` -- 118. Summing catches an under-count in any single row without pinning every number to an exact value, which would make the table annoying to maintain for no extra safety. Mutation-tested rather than assumed: deleting the `doubles/` row turns two assertions red, changing `learn/` from 14 to 5 turns one red, and renaming the ADR turns all three red. That last one is why the file-exists assertion is there -- without it a rename would leave the greps matching nothing and the whole file passing while checking nothing, which is exactly how a layering contract rotted unnoticed in #946. 1633 sequential / 1592 parallel; both are baseline + the 3 new tests.

… the ADR contract The Bash 3.0 jobs run in a container where git refuses to read the repo (dubious-ownership), so `git ls-files` exited non-zero and produced nothing. Both count-based assertions then compared against an empty listing and failed for a reason unrelated to the ADR. `find` needs no git and no index.

Chemaclass · 2026-08-02T09:24:07Z

🤔 Background

ADR-011's module table was wrong in two places. Both errors came from counting with a one-level glob instead of from the tree.

💡 Changes

doubles/ was missing outright. The table is ordered by load order, so it was generated from the entrypoint's source lines — and doubles/ is the one module the entrypoint doesn't source (assert/index.sh does). It fell through silently while the prose directly above said "Seventeen", a number that came from counting directories. The two disagreed and nothing noticed. It's now row 12, where it actually loads, and the row records where it's sourced from.

learn/ was listed at 5 files / 240 lines. It's 14 / 1296. The difference is learn/lessons/, which src/learn/*.sh doesn't descend into — the same one-level-glob mistake Makefile:67 made with nested tests. Other rows had drifted 2-3 lines from recent comment edits; refreshed too.

docs/project-overview.md listed 15 of 17 modules, missing doubles and dev. Spies and mocks are a headline feature to be absent from the source-layout overview.

🧪 The contract test

Asserts every src/ module has a row, and that the Files column sums to 118 — the tracked .sh count. Summing catches an under-count in any single row without pinning every number exactly, which would make the table tedious to maintain for no extra safety.

Mutation-tested, not assumed:

Mutation	Result
Delete the `doubles/` row	2 assertions red
`learn/` files 14 → 5	1 assertion red
Rename the ADR away	all 3 red

That last row is why the file-exists assertion exists — without it a rename would leave the greps matching nothing and the whole file passing while checking nothing, exactly how a layering contract rotted unnoticed in #946.

✅ Verification

make sa · make lint · bash build.sh bin -v → ✅ Build verified ✅. 1633 sequential / 1592 parallel-simple-strict — baseline + the 3 new tests.

ADR-011's module table was wrong in two places, and both errors came from counting with a one-level glob instead of from the tree. `doubles/` was missing outright. The table is ordered by load order, so it was generated from the entrypoint's `source` lines -- and `doubles/` is the one module the entrypoint does not source, `assert/index.sh` does. It fell through silently while the prose directly above said "Seventeen", a number that came from counting directories. The two disagreed and nothing noticed. The module is now row 12, where it actually loads, and the row says where it is sourced from so the next reader does not have to rediscover the discrepancy. `learn/` was listed at 5 files / 240 lines. It is 14 files / 1296. The difference is `learn/lessons/`, which `src/learn/*.sh` does not descend into -- the same one-level-glob mistake `Makefile:67` made with nested tests. Every other row's line count had drifted by two or three from recent comment edits; those are refreshed too. `docs/project-overview.md` listed 15 of the 17 modules, missing `doubles` and `dev`. Spies and mocks are a headline feature to be absent from the overview of the source layout. The new test asserts every `src/` module has a row, and that the Files column sums to the number of tracked `.sh` files in `src/` -- 118. Summing catches an under-count in any single row without pinning every number to an exact value, which would make the table annoying to maintain for no extra safety. Mutation-tested rather than assumed: deleting the `doubles/` row turns two assertions red, changing `learn/` from 14 to 5 turns one red, and renaming the ADR turns all three red. That last one is why the file-exists assertion is there -- without it a rename would leave the greps matching nothing and the whole file passing while checking nothing, which is exactly how a layering contract rotted unnoticed in #946. 1633 sequential / 1592 parallel; both are baseline + the 3 new tests.

… the ADR contract The Bash 3.0 jobs run in a container where git refuses to read the repo (dubious-ownership), so `git ls-files` exited non-zero and produced nothing. Both count-based assertions then compared against an empty listing and failed for a reason unrelated to the ADR. `find` needs no git and no index.

Chemaclass · 2026-08-02T09:24:07Z

🤔 Background

ADR-011's module table was wrong in two places. Both errors came from counting with a one-level glob instead of from the tree.

💡 Changes

doubles/ was missing outright. The table is ordered by load order, so it was generated from the entrypoint's source lines — and doubles/ is the one module the entrypoint doesn't source (assert/index.sh does). It fell through silently while the prose directly above said "Seventeen", a number that came from counting directories. The two disagreed and nothing noticed. It's now row 12, where it actually loads, and the row records where it's sourced from.

learn/ was listed at 5 files / 240 lines. It's 14 / 1296. The difference is learn/lessons/, which src/learn/*.sh doesn't descend into — the same one-level-glob mistake Makefile:67 made with nested tests. Other rows had drifted 2-3 lines from recent comment edits; refreshed too.

docs/project-overview.md listed 15 of 17 modules, missing doubles and dev. Spies and mocks are a headline feature to be absent from the source-layout overview.

🧪 The contract test

Asserts every src/ module has a row, and that the Files column sums to 118 — the tracked .sh count. Summing catches an under-count in any single row without pinning every number exactly, which would make the table tedious to maintain for no extra safety.

Mutation-tested, not assumed:

Mutation	Result
Delete the `doubles/` row	2 assertions red
`learn/` files 14 → 5	1 assertion red
Rename the ADR away	all 3 red

That last row is why the file-exists assertion exists — without it a rename would leave the greps matching nothing and the whole file passing while checking nothing, exactly how a layering contract rotted unnoticed in #946.

✅ Verification

make sa · make lint · bash build.sh bin -v → ✅ Build verified ✅. 1633 sequential / 1592 parallel-simple-strict — baseline + the 3 new tests.

ADR-011's module table was wrong in two places, and both errors came from counting with a one-level glob instead of from the tree. `doubles/` was missing outright. The table is ordered by load order, so it was generated from the entrypoint's `source` lines -- and `doubles/` is the one module the entrypoint does not source, `assert/index.sh` does. It fell through silently while the prose directly above said "Seventeen", a number that came from counting directories. The two disagreed and nothing noticed. The module is now row 12, where it actually loads, and the row says where it is sourced from so the next reader does not have to rediscover the discrepancy. `learn/` was listed at 5 files / 240 lines. It is 14 files / 1296. The difference is `learn/lessons/`, which `src/learn/*.sh` does not descend into -- the same one-level-glob mistake `Makefile:67` made with nested tests. Every other row's line count had drifted by two or three from recent comment edits; those are refreshed too. `docs/project-overview.md` listed 15 of the 17 modules, missing `doubles` and `dev`. Spies and mocks are a headline feature to be absent from the overview of the source layout. The new test asserts every `src/` module has a row, and that the Files column sums to the number of tracked `.sh` files in `src/` -- 118. Summing catches an under-count in any single row without pinning every number to an exact value, which would make the table annoying to maintain for no extra safety. Mutation-tested rather than assumed: deleting the `doubles/` row turns two assertions red, changing `learn/` from 14 to 5 turns one red, and renaming the ADR turns all three red. That last one is why the file-exists assertion is there -- without it a rename would leave the greps matching nothing and the whole file passing while checking nothing, which is exactly how a layering contract rotted unnoticed in #946. 1633 sequential / 1592 parallel; both are baseline + the 3 new tests.

… the ADR contract The Bash 3.0 jobs run in a container where git refuses to read the repo (dubious-ownership), so `git ls-files` exited non-zero and produced nothing. Both count-based assertions then compared against an empty listing and failed for a reason unrelated to the ADR. `find` needs no git and no index.

Chemaclass · 2026-08-02T09:24:07Z

🤔 Background

ADR-011's module table was wrong in two places. Both errors came from counting with a one-level glob instead of from the tree.

💡 Changes

doubles/ was missing outright. The table is ordered by load order, so it was generated from the entrypoint's source lines — and doubles/ is the one module the entrypoint doesn't source (assert/index.sh does). It fell through silently while the prose directly above said "Seventeen", a number that came from counting directories. The two disagreed and nothing noticed. It's now row 12, where it actually loads, and the row records where it's sourced from.

learn/ was listed at 5 files / 240 lines. It's 14 / 1296. The difference is learn/lessons/, which src/learn/*.sh doesn't descend into — the same one-level-glob mistake Makefile:67 made with nested tests. Other rows had drifted 2-3 lines from recent comment edits; refreshed too.

docs/project-overview.md listed 15 of 17 modules, missing doubles and dev. Spies and mocks are a headline feature to be absent from the source-layout overview.

🧪 The contract test

Asserts every src/ module has a row, and that the Files column sums to 118 — the tracked .sh count. Summing catches an under-count in any single row without pinning every number exactly, which would make the table tedious to maintain for no extra safety.

Mutation-tested, not assumed:

Mutation	Result
Delete the `doubles/` row	2 assertions red
`learn/` files 14 → 5	1 assertion red
Rename the ADR away	all 3 red

That last row is why the file-exists assertion exists — without it a rename would leave the greps matching nothing and the whole file passing while checking nothing, exactly how a layering contract rotted unnoticed in #946.

✅ Verification

make sa · make lint · bash build.sh bin -v → ✅ Build verified ✅. 1633 sequential / 1592 parallel-simple-strict — baseline + the 3 new tests.

ADR-011's module table was wrong in two places, and both errors came from counting with a one-level glob instead of from the tree. `doubles/` was missing outright. The table is ordered by load order, so it was generated from the entrypoint's `source` lines -- and `doubles/` is the one module the entrypoint does not source, `assert/index.sh` does. It fell through silently while the prose directly above said "Seventeen", a number that came from counting directories. The two disagreed and nothing noticed. The module is now row 12, where it actually loads, and the row says where it is sourced from so the next reader does not have to rediscover the discrepancy. `learn/` was listed at 5 files / 240 lines. It is 14 files / 1296. The difference is `learn/lessons/`, which `src/learn/*.sh` does not descend into -- the same one-level-glob mistake `Makefile:67` made with nested tests. Every other row's line count had drifted by two or three from recent comment edits; those are refreshed too. `docs/project-overview.md` listed 15 of the 17 modules, missing `doubles` and `dev`. Spies and mocks are a headline feature to be absent from the overview of the source layout. The new test asserts every `src/` module has a row, and that the Files column sums to the number of tracked `.sh` files in `src/` -- 118. Summing catches an under-count in any single row without pinning every number to an exact value, which would make the table annoying to maintain for no extra safety. Mutation-tested rather than assumed: deleting the `doubles/` row turns two assertions red, changing `learn/` from 14 to 5 turns one red, and renaming the ADR turns all three red. That last one is why the file-exists assertion is there -- without it a rename would leave the greps matching nothing and the whole file passing while checking nothing, which is exactly how a layering contract rotted unnoticed in #946. 1633 sequential / 1592 parallel; both are baseline + the 3 new tests.

… the ADR contract The Bash 3.0 jobs run in a container where git refuses to read the repo (dubious-ownership), so `git ls-files` exited non-zero and produced nothing. Both count-based assertions then compared against an empty listing and failed for a reason unrelated to the ADR. `find` needs no git and no index.

Chemaclass added the documentation Improvements or additions to documentation label Aug 2, 2026

Chemaclass self-assigned this Aug 2, 2026

Chemaclass merged commit bf34ea7 into mainAug 2, 2026
37 checks passed

Chemaclass deleted the docs/adr-module-table-drift branch August 2, 2026 09:38

Chemaclass added the documentation Improvements or additions to documentation label Aug 2, 2026

Chemaclass self-assigned this Aug 2, 2026

Chemaclass merged commit bf34ea7 into mainAug 2, 2026
37 checks passed

Chemaclass deleted the docs/adr-module-table-drift branch August 2, 2026 09:38

Chemaclass added the documentation Improvements or additions to documentation label Aug 2, 2026

Chemaclass self-assigned this Aug 2, 2026

Chemaclass merged commit bf34ea7 into mainAug 2, 2026
37 checks passed

Chemaclass deleted the docs/adr-module-table-drift branch August 2, 2026 09:38

Chemaclass added the documentation Improvements or additions to documentation label Aug 2, 2026

Chemaclass self-assigned this Aug 2, 2026

Chemaclass merged commit bf34ea7 into mainAug 2, 2026
37 checks passed

Chemaclass deleted the docs/adr-module-table-drift branch August 2, 2026 09:38

Chemaclass added the documentation Improvements or additions to documentation label Aug 2, 2026

Chemaclass self-assigned this Aug 2, 2026

Chemaclass merged commit bf34ea7 into mainAug 2, 2026
37 checks passed

Chemaclass deleted the docs/adr-module-table-drift branch August 2, 2026 09:38

Chemaclass added the documentation Improvements or additions to documentation label Aug 2, 2026

Chemaclass self-assigned this Aug 2, 2026

Chemaclass merged commit bf34ea7 into mainAug 2, 2026
37 checks passed

Chemaclass deleted the docs/adr-module-table-drift branch August 2, 2026 09:38

Chemaclass added the documentation Improvements or additions to documentation label Aug 2, 2026

Chemaclass self-assigned this Aug 2, 2026

Chemaclass merged commit bf34ea7 into mainAug 2, 2026
37 checks passed

Chemaclass deleted the docs/adr-module-table-drift branch August 2, 2026 09:38

Chemaclass added the documentation Improvements or additions to documentation label Aug 2, 2026

Chemaclass self-assigned this Aug 2, 2026

Chemaclass merged commit bf34ea7 into mainAug 2, 2026
37 checks passed

Chemaclass deleted the docs/adr-module-table-drift branch August 2, 2026 09:38

Uh oh!

Uh oh!

Conversation

Chemaclass commented Aug 2, 2026

🤔 Background

💡 Changes

🧪 The contract test

✅ Verification

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Uh oh!

Uh oh!

Conversation

Chemaclass commented Aug 2, 2026

🤔 Background

💡 Changes

🧪 The contract test

✅ Verification

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Uh oh!

Uh oh!

Conversation

Chemaclass commented Aug 2, 2026

🤔 Background

💡 Changes

🧪 The contract test

✅ Verification

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Uh oh!

Uh oh!

Conversation

Chemaclass commented Aug 2, 2026

🤔 Background

💡 Changes

🧪 The contract test

✅ Verification

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Uh oh!

Uh oh!

Conversation

Chemaclass commented Aug 2, 2026

🤔 Background

💡 Changes

🧪 The contract test

✅ Verification