chore: stop tracking runtime and build artifacts

[nadeem4]

Task 1 of the 14-task implementation plan: repo hygiene — stop tracking build artifacts.

What changed

• Untracked data/manufacturing.db and last_reasoning.json via git rm --cached (both remain on disk).
• Tidied .gitignore: grouped the loose trailing entries under existing comment headings, switched directory patterns to trailing-slash form, added .pytest_cache/, and added a trailing newline (the file previously had none).

Most of the plan's paths were already untracked

The plan was written against an older repo state. git ls-files shows chroma_db/, site/, logs/, artifacts/, __pycache__/ and .pytest_cache/ have zero tracked files already. Only two artifact paths were actually tracked, and those are the two removed here. The .gitignore entries for the rest are retained/tidied so they stay ignored.

Deviations from the plan text

(a) artifacts/ is kept anchored as /artifacts/ — the plan's bare artifacts/ is not used.

A bare artifacts/ matches a directory of that name at any depth, including the source package packages/core/src/nl2sql/execution/artifacts/, which would silently make new files there unaddable. This was a real bug fixed in an earlier PR. Verified both forms in isolation:

$ printf 'artifacts/\n' > /tmp/t
$ git -c core.excludesFile=/tmp/t check-ignore -v --no-index \
packages/core/src/nl2sql/execution/artifacts/store.py
/tmp/t:1:artifacts/ packages/core/src/nl2sql/execution/artifacts/store.py # MATCHES - hazard
$ printf '/artifacts/\n' > /tmp/t2
$ git -c core.excludesFile=/tmp/t2 check-ignore -v --no-index \
packages/core/src/nl2sql/execution/artifacts/store.py
(no output, exit 1) # safe

With the anchored form as committed, git check-ignore reports nothing for store.py, parquet.py, __init__.py, or a hypothetical new file in that directory.

(b) The bare data line is replaced with specific generated paths.

The plan says "do not ignore data/ wholesale", but .gitignore already contained a bare data line doing exactly that (git check-ignore attributed data/schema_store.db and data/demo_lite/ to .gitignore:27:data). It is replaced with the specific generated paths — data/demo_lite/, data/vector_store_demo/ — alongside the pre-existing data/*.db, which already covers manufacturing.db and schema_store.db. Files added under data/ are now visible to git again.

Discovery: no test loads a committed fixture

$ grep -rn "chroma_db\|last_reasoning" packages/*/tests packages/*/src conftest.py pytest.ini
packages/cli/src/nl2sql_cli/commands/run.py:91,97 # WRITES last_reasoning.json at runtime
packages/core/src/nl2sql/common/settings.py:17 # "./chroma_db" default setting value

Both are runtime paths, not committed fixtures.

$ grep -rn "manufacturing.db" packages conftest.py
(zero hits)

The plan said to keep data/manufacturing.db tracked only if tests use it. Nothing references it, so it is untracked. data/schema_store.db was never tracked; every schema_store.db reference is tmp_dir / "schema_store.db" in a runtime temp directory.

Verification

• git ls-files | wc -l: 411 before → 409 after — exactly the two intended removals.
• No previously tracked file became ignored (piped all of git ls-files through git check-ignore; only the two targets matched).
• Both files confirmed still present on disk after git rm --cached.
• git status clean after commit.

Test results

Command: OPENAI_API_KEY= pytest -m "not integration" -q (pytest-randomly active, so ordering varies).

Before (baseline, index and .gitignore temporarily reverted):

200 passed, 1 skipped, 48 deselected, 1 xfailed, 27 warnings in 11.16s

After (run twice for randomised ordering):

200 passed, 1 skipped, 48 deselected, 1 xfailed, 27 warnings in 11.34s
200 passed, 1 skipped, 48 deselected, 1 xfailed, 27 warnings in 5.02s

Identical before and after. The pre-existing skip (pyodbc driver not installed) and xfail (PhysicalValidator not wired into the default SQL subgraph) are unrelated to this change.

No CHANGELOG.md or version changes. No generated demo data committed.

[nadeem4]

Task 1 of the 14-task implementation plan: repo hygiene — stop tracking build artifacts.

What changed

• Untracked data/manufacturing.db and last_reasoning.json via git rm --cached (both remain on disk).
• Tidied .gitignore: grouped the loose trailing entries under existing comment headings, switched directory patterns to trailing-slash form, added .pytest_cache/, and added a trailing newline (the file previously had none).

Most of the plan's paths were already untracked

The plan was written against an older repo state. git ls-files shows chroma_db/, site/, logs/, artifacts/, __pycache__/ and .pytest_cache/ have zero tracked files already. Only two artifact paths were actually tracked, and those are the two removed here. The .gitignore entries for the rest are retained/tidied so they stay ignored.

Deviations from the plan text

(a) artifacts/ is kept anchored as /artifacts/ — the plan's bare artifacts/ is not used.

A bare artifacts/ matches a directory of that name at any depth, including the source package packages/core/src/nl2sql/execution/artifacts/, which would silently make new files there unaddable. This was a real bug fixed in an earlier PR. Verified both forms in isolation:

$ printf 'artifacts/\n' > /tmp/t
$ git -c core.excludesFile=/tmp/t check-ignore -v --no-index \
packages/core/src/nl2sql/execution/artifacts/store.py
/tmp/t:1:artifacts/ packages/core/src/nl2sql/execution/artifacts/store.py # MATCHES - hazard
$ printf '/artifacts/\n' > /tmp/t2
$ git -c core.excludesFile=/tmp/t2 check-ignore -v --no-index \
packages/core/src/nl2sql/execution/artifacts/store.py
(no output, exit 1) # safe

With the anchored form as committed, git check-ignore reports nothing for store.py, parquet.py, __init__.py, or a hypothetical new file in that directory.

(b) The bare data line is replaced with specific generated paths.

The plan says "do not ignore data/ wholesale", but .gitignore already contained a bare data line doing exactly that (git check-ignore attributed data/schema_store.db and data/demo_lite/ to .gitignore:27:data). It is replaced with the specific generated paths — data/demo_lite/, data/vector_store_demo/ — alongside the pre-existing data/*.db, which already covers manufacturing.db and schema_store.db. Files added under data/ are now visible to git again.

Discovery: no test loads a committed fixture

$ grep -rn "chroma_db\|last_reasoning" packages/*/tests packages/*/src conftest.py pytest.ini
packages/cli/src/nl2sql_cli/commands/run.py:91,97 # WRITES last_reasoning.json at runtime
packages/core/src/nl2sql/common/settings.py:17 # "./chroma_db" default setting value

Both are runtime paths, not committed fixtures.

$ grep -rn "manufacturing.db" packages conftest.py
(zero hits)

The plan said to keep data/manufacturing.db tracked only if tests use it. Nothing references it, so it is untracked. data/schema_store.db was never tracked; every schema_store.db reference is tmp_dir / "schema_store.db" in a runtime temp directory.

Verification

• git ls-files | wc -l: 411 before → 409 after — exactly the two intended removals.
• No previously tracked file became ignored (piped all of git ls-files through git check-ignore; only the two targets matched).
• Both files confirmed still present on disk after git rm --cached.
• git status clean after commit.

Test results

Command: OPENAI_API_KEY= pytest -m "not integration" -q (pytest-randomly active, so ordering varies).

Before (baseline, index and .gitignore temporarily reverted):

200 passed, 1 skipped, 48 deselected, 1 xfailed, 27 warnings in 11.16s

After (run twice for randomised ordering):

200 passed, 1 skipped, 48 deselected, 1 xfailed, 27 warnings in 11.34s
200 passed, 1 skipped, 48 deselected, 1 xfailed, 27 warnings in 5.02s

Identical before and after. The pre-existing skip (pyodbc driver not installed) and xfail (PhysicalValidator not wired into the default SQL subgraph) are unrelated to this change.

No CHANGELOG.md or version changes. No generated demo data committed.

[nadeem4]

Task 1 of the 14-task implementation plan: repo hygiene — stop tracking build artifacts.

What changed

• Untracked data/manufacturing.db and last_reasoning.json via git rm --cached (both remain on disk).
• Tidied .gitignore: grouped the loose trailing entries under existing comment headings, switched directory patterns to trailing-slash form, added .pytest_cache/, and added a trailing newline (the file previously had none).

Most of the plan's paths were already untracked

The plan was written against an older repo state. git ls-files shows chroma_db/, site/, logs/, artifacts/, __pycache__/ and .pytest_cache/ have zero tracked files already. Only two artifact paths were actually tracked, and those are the two removed here. The .gitignore entries for the rest are retained/tidied so they stay ignored.

Deviations from the plan text

(a) artifacts/ is kept anchored as /artifacts/ — the plan's bare artifacts/ is not used.

A bare artifacts/ matches a directory of that name at any depth, including the source package packages/core/src/nl2sql/execution/artifacts/, which would silently make new files there unaddable. This was a real bug fixed in an earlier PR. Verified both forms in isolation:

$ printf 'artifacts/\n' > /tmp/t
$ git -c core.excludesFile=/tmp/t check-ignore -v --no-index \
packages/core/src/nl2sql/execution/artifacts/store.py
/tmp/t:1:artifacts/ packages/core/src/nl2sql/execution/artifacts/store.py # MATCHES - hazard
$ printf '/artifacts/\n' > /tmp/t2
$ git -c core.excludesFile=/tmp/t2 check-ignore -v --no-index \
packages/core/src/nl2sql/execution/artifacts/store.py
(no output, exit 1) # safe

With the anchored form as committed, git check-ignore reports nothing for store.py, parquet.py, __init__.py, or a hypothetical new file in that directory.

(b) The bare data line is replaced with specific generated paths.

The plan says "do not ignore data/ wholesale", but .gitignore already contained a bare data line doing exactly that (git check-ignore attributed data/schema_store.db and data/demo_lite/ to .gitignore:27:data). It is replaced with the specific generated paths — data/demo_lite/, data/vector_store_demo/ — alongside the pre-existing data/*.db, which already covers manufacturing.db and schema_store.db. Files added under data/ are now visible to git again.

Discovery: no test loads a committed fixture

$ grep -rn "chroma_db\|last_reasoning" packages/*/tests packages/*/src conftest.py pytest.ini
packages/cli/src/nl2sql_cli/commands/run.py:91,97 # WRITES last_reasoning.json at runtime
packages/core/src/nl2sql/common/settings.py:17 # "./chroma_db" default setting value

Both are runtime paths, not committed fixtures.

$ grep -rn "manufacturing.db" packages conftest.py
(zero hits)

The plan said to keep data/manufacturing.db tracked only if tests use it. Nothing references it, so it is untracked. data/schema_store.db was never tracked; every schema_store.db reference is tmp_dir / "schema_store.db" in a runtime temp directory.

Verification

• git ls-files | wc -l: 411 before → 409 after — exactly the two intended removals.
• No previously tracked file became ignored (piped all of git ls-files through git check-ignore; only the two targets matched).
• Both files confirmed still present on disk after git rm --cached.
• git status clean after commit.

Test results

Command: OPENAI_API_KEY= pytest -m "not integration" -q (pytest-randomly active, so ordering varies).

Before (baseline, index and .gitignore temporarily reverted):

200 passed, 1 skipped, 48 deselected, 1 xfailed, 27 warnings in 11.16s

After (run twice for randomised ordering):

200 passed, 1 skipped, 48 deselected, 1 xfailed, 27 warnings in 11.34s
200 passed, 1 skipped, 48 deselected, 1 xfailed, 27 warnings in 5.02s

Identical before and after. The pre-existing skip (pyodbc driver not installed) and xfail (PhysicalValidator not wired into the default SQL subgraph) are unrelated to this change.

No CHANGELOG.md or version changes. No generated demo data committed.

[nadeem4]

Task 1 of the 14-task implementation plan: repo hygiene — stop tracking build artifacts.

What changed

• Untracked data/manufacturing.db and last_reasoning.json via git rm --cached (both remain on disk).
• Tidied .gitignore: grouped the loose trailing entries under existing comment headings, switched directory patterns to trailing-slash form, added .pytest_cache/, and added a trailing newline (the file previously had none).

Most of the plan's paths were already untracked

The plan was written against an older repo state. git ls-files shows chroma_db/, site/, logs/, artifacts/, __pycache__/ and .pytest_cache/ have zero tracked files already. Only two artifact paths were actually tracked, and those are the two removed here. The .gitignore entries for the rest are retained/tidied so they stay ignored.

Deviations from the plan text

(a) artifacts/ is kept anchored as /artifacts/ — the plan's bare artifacts/ is not used.

A bare artifacts/ matches a directory of that name at any depth, including the source package packages/core/src/nl2sql/execution/artifacts/, which would silently make new files there unaddable. This was a real bug fixed in an earlier PR. Verified both forms in isolation:

$ printf 'artifacts/\n' > /tmp/t
$ git -c core.excludesFile=/tmp/t check-ignore -v --no-index \
packages/core/src/nl2sql/execution/artifacts/store.py
/tmp/t:1:artifacts/ packages/core/src/nl2sql/execution/artifacts/store.py # MATCHES - hazard
$ printf '/artifacts/\n' > /tmp/t2
$ git -c core.excludesFile=/tmp/t2 check-ignore -v --no-index \
packages/core/src/nl2sql/execution/artifacts/store.py
(no output, exit 1) # safe

With the anchored form as committed, git check-ignore reports nothing for store.py, parquet.py, __init__.py, or a hypothetical new file in that directory.

(b) The bare data line is replaced with specific generated paths.

The plan says "do not ignore data/ wholesale", but .gitignore already contained a bare data line doing exactly that (git check-ignore attributed data/schema_store.db and data/demo_lite/ to .gitignore:27:data). It is replaced with the specific generated paths — data/demo_lite/, data/vector_store_demo/ — alongside the pre-existing data/*.db, which already covers manufacturing.db and schema_store.db. Files added under data/ are now visible to git again.

Discovery: no test loads a committed fixture

$ grep -rn "chroma_db\|last_reasoning" packages/*/tests packages/*/src conftest.py pytest.ini
packages/cli/src/nl2sql_cli/commands/run.py:91,97 # WRITES last_reasoning.json at runtime
packages/core/src/nl2sql/common/settings.py:17 # "./chroma_db" default setting value

Both are runtime paths, not committed fixtures.

$ grep -rn "manufacturing.db" packages conftest.py
(zero hits)

The plan said to keep data/manufacturing.db tracked only if tests use it. Nothing references it, so it is untracked. data/schema_store.db was never tracked; every schema_store.db reference is tmp_dir / "schema_store.db" in a runtime temp directory.

Verification

• git ls-files | wc -l: 411 before → 409 after — exactly the two intended removals.
• No previously tracked file became ignored (piped all of git ls-files through git check-ignore; only the two targets matched).
• Both files confirmed still present on disk after git rm --cached.
• git status clean after commit.

Test results

Command: OPENAI_API_KEY= pytest -m "not integration" -q (pytest-randomly active, so ordering varies).

Before (baseline, index and .gitignore temporarily reverted):

200 passed, 1 skipped, 48 deselected, 1 xfailed, 27 warnings in 11.16s

After (run twice for randomised ordering):

200 passed, 1 skipped, 48 deselected, 1 xfailed, 27 warnings in 11.34s
200 passed, 1 skipped, 48 deselected, 1 xfailed, 27 warnings in 5.02s

Identical before and after. The pre-existing skip (pyodbc driver not installed) and xfail (PhysicalValidator not wired into the default SQL subgraph) are unrelated to this change.

No CHANGELOG.md or version changes. No generated demo data committed.

[nadeem4]

Task 1 of the 14-task implementation plan: repo hygiene — stop tracking build artifacts.

What changed

• Untracked data/manufacturing.db and last_reasoning.json via git rm --cached (both remain on disk).
• Tidied .gitignore: grouped the loose trailing entries under existing comment headings, switched directory patterns to trailing-slash form, added .pytest_cache/, and added a trailing newline (the file previously had none).

Most of the plan's paths were already untracked

The plan was written against an older repo state. git ls-files shows chroma_db/, site/, logs/, artifacts/, __pycache__/ and .pytest_cache/ have zero tracked files already. Only two artifact paths were actually tracked, and those are the two removed here. The .gitignore entries for the rest are retained/tidied so they stay ignored.

Deviations from the plan text

(a) artifacts/ is kept anchored as /artifacts/ — the plan's bare artifacts/ is not used.

A bare artifacts/ matches a directory of that name at any depth, including the source package packages/core/src/nl2sql/execution/artifacts/, which would silently make new files there unaddable. This was a real bug fixed in an earlier PR. Verified both forms in isolation:

$ printf 'artifacts/\n' > /tmp/t
$ git -c core.excludesFile=/tmp/t check-ignore -v --no-index \
packages/core/src/nl2sql/execution/artifacts/store.py
/tmp/t:1:artifacts/ packages/core/src/nl2sql/execution/artifacts/store.py # MATCHES - hazard
$ printf '/artifacts/\n' > /tmp/t2
$ git -c core.excludesFile=/tmp/t2 check-ignore -v --no-index \
packages/core/src/nl2sql/execution/artifacts/store.py
(no output, exit 1) # safe

With the anchored form as committed, git check-ignore reports nothing for store.py, parquet.py, __init__.py, or a hypothetical new file in that directory.

(b) The bare data line is replaced with specific generated paths.

The plan says "do not ignore data/ wholesale", but .gitignore already contained a bare data line doing exactly that (git check-ignore attributed data/schema_store.db and data/demo_lite/ to .gitignore:27:data). It is replaced with the specific generated paths — data/demo_lite/, data/vector_store_demo/ — alongside the pre-existing data/*.db, which already covers manufacturing.db and schema_store.db. Files added under data/ are now visible to git again.

Discovery: no test loads a committed fixture

$ grep -rn "chroma_db\|last_reasoning" packages/*/tests packages/*/src conftest.py pytest.ini
packages/cli/src/nl2sql_cli/commands/run.py:91,97 # WRITES last_reasoning.json at runtime
packages/core/src/nl2sql/common/settings.py:17 # "./chroma_db" default setting value

Both are runtime paths, not committed fixtures.

$ grep -rn "manufacturing.db" packages conftest.py
(zero hits)

The plan said to keep data/manufacturing.db tracked only if tests use it. Nothing references it, so it is untracked. data/schema_store.db was never tracked; every schema_store.db reference is tmp_dir / "schema_store.db" in a runtime temp directory.

Verification

• git ls-files | wc -l: 411 before → 409 after — exactly the two intended removals.
• No previously tracked file became ignored (piped all of git ls-files through git check-ignore; only the two targets matched).
• Both files confirmed still present on disk after git rm --cached.
• git status clean after commit.

Test results

Command: OPENAI_API_KEY= pytest -m "not integration" -q (pytest-randomly active, so ordering varies).

Before (baseline, index and .gitignore temporarily reverted):

200 passed, 1 skipped, 48 deselected, 1 xfailed, 27 warnings in 11.16s

After (run twice for randomised ordering):

200 passed, 1 skipped, 48 deselected, 1 xfailed, 27 warnings in 11.34s
200 passed, 1 skipped, 48 deselected, 1 xfailed, 27 warnings in 5.02s

Identical before and after. The pre-existing skip (pyodbc driver not installed) and xfail (PhysicalValidator not wired into the default SQL subgraph) are unrelated to this change.

No CHANGELOG.md or version changes. No generated demo data committed.

[nadeem4]

Task 1 of the 14-task implementation plan: repo hygiene — stop tracking build artifacts.

What changed

• Untracked data/manufacturing.db and last_reasoning.json via git rm --cached (both remain on disk).
• Tidied .gitignore: grouped the loose trailing entries under existing comment headings, switched directory patterns to trailing-slash form, added .pytest_cache/, and added a trailing newline (the file previously had none).

Most of the plan's paths were already untracked

The plan was written against an older repo state. git ls-files shows chroma_db/, site/, logs/, artifacts/, __pycache__/ and .pytest_cache/ have zero tracked files already. Only two artifact paths were actually tracked, and those are the two removed here. The .gitignore entries for the rest are retained/tidied so they stay ignored.

Deviations from the plan text

(a) artifacts/ is kept anchored as /artifacts/ — the plan's bare artifacts/ is not used.

A bare artifacts/ matches a directory of that name at any depth, including the source package packages/core/src/nl2sql/execution/artifacts/, which would silently make new files there unaddable. This was a real bug fixed in an earlier PR. Verified both forms in isolation:

$ printf 'artifacts/\n' > /tmp/t
$ git -c core.excludesFile=/tmp/t check-ignore -v --no-index \
packages/core/src/nl2sql/execution/artifacts/store.py
/tmp/t:1:artifacts/ packages/core/src/nl2sql/execution/artifacts/store.py # MATCHES - hazard
$ printf '/artifacts/\n' > /tmp/t2
$ git -c core.excludesFile=/tmp/t2 check-ignore -v --no-index \
packages/core/src/nl2sql/execution/artifacts/store.py
(no output, exit 1) # safe

With the anchored form as committed, git check-ignore reports nothing for store.py, parquet.py, __init__.py, or a hypothetical new file in that directory.

(b) The bare data line is replaced with specific generated paths.

The plan says "do not ignore data/ wholesale", but .gitignore already contained a bare data line doing exactly that (git check-ignore attributed data/schema_store.db and data/demo_lite/ to .gitignore:27:data). It is replaced with the specific generated paths — data/demo_lite/, data/vector_store_demo/ — alongside the pre-existing data/*.db, which already covers manufacturing.db and schema_store.db. Files added under data/ are now visible to git again.

Discovery: no test loads a committed fixture

$ grep -rn "chroma_db\|last_reasoning" packages/*/tests packages/*/src conftest.py pytest.ini
packages/cli/src/nl2sql_cli/commands/run.py:91,97 # WRITES last_reasoning.json at runtime
packages/core/src/nl2sql/common/settings.py:17 # "./chroma_db" default setting value

Both are runtime paths, not committed fixtures.

$ grep -rn "manufacturing.db" packages conftest.py
(zero hits)

The plan said to keep data/manufacturing.db tracked only if tests use it. Nothing references it, so it is untracked. data/schema_store.db was never tracked; every schema_store.db reference is tmp_dir / "schema_store.db" in a runtime temp directory.

Verification

• git ls-files | wc -l: 411 before → 409 after — exactly the two intended removals.
• No previously tracked file became ignored (piped all of git ls-files through git check-ignore; only the two targets matched).
• Both files confirmed still present on disk after git rm --cached.
• git status clean after commit.

Test results

Command: OPENAI_API_KEY= pytest -m "not integration" -q (pytest-randomly active, so ordering varies).

Before (baseline, index and .gitignore temporarily reverted):

200 passed, 1 skipped, 48 deselected, 1 xfailed, 27 warnings in 11.16s

After (run twice for randomised ordering):

200 passed, 1 skipped, 48 deselected, 1 xfailed, 27 warnings in 11.34s
200 passed, 1 skipped, 48 deselected, 1 xfailed, 27 warnings in 5.02s

Identical before and after. The pre-existing skip (pyodbc driver not installed) and xfail (PhysicalValidator not wired into the default SQL subgraph) are unrelated to this change.

No CHANGELOG.md or version changes. No generated demo data committed.

[nadeem4]

Task 1 of the 14-task implementation plan: repo hygiene — stop tracking build artifacts.

What changed

• Untracked data/manufacturing.db and last_reasoning.json via git rm --cached (both remain on disk).
• Tidied .gitignore: grouped the loose trailing entries under existing comment headings, switched directory patterns to trailing-slash form, added .pytest_cache/, and added a trailing newline (the file previously had none).

Most of the plan's paths were already untracked

The plan was written against an older repo state. git ls-files shows chroma_db/, site/, logs/, artifacts/, __pycache__/ and .pytest_cache/ have zero tracked files already. Only two artifact paths were actually tracked, and those are the two removed here. The .gitignore entries for the rest are retained/tidied so they stay ignored.

Deviations from the plan text

(a) artifacts/ is kept anchored as /artifacts/ — the plan's bare artifacts/ is not used.

A bare artifacts/ matches a directory of that name at any depth, including the source package packages/core/src/nl2sql/execution/artifacts/, which would silently make new files there unaddable. This was a real bug fixed in an earlier PR. Verified both forms in isolation:

$ printf 'artifacts/\n' > /tmp/t
$ git -c core.excludesFile=/tmp/t check-ignore -v --no-index \
packages/core/src/nl2sql/execution/artifacts/store.py
/tmp/t:1:artifacts/ packages/core/src/nl2sql/execution/artifacts/store.py # MATCHES - hazard
$ printf '/artifacts/\n' > /tmp/t2
$ git -c core.excludesFile=/tmp/t2 check-ignore -v --no-index \
packages/core/src/nl2sql/execution/artifacts/store.py
(no output, exit 1) # safe

With the anchored form as committed, git check-ignore reports nothing for store.py, parquet.py, __init__.py, or a hypothetical new file in that directory.

(b) The bare data line is replaced with specific generated paths.

The plan says "do not ignore data/ wholesale", but .gitignore already contained a bare data line doing exactly that (git check-ignore attributed data/schema_store.db and data/demo_lite/ to .gitignore:27:data). It is replaced with the specific generated paths — data/demo_lite/, data/vector_store_demo/ — alongside the pre-existing data/*.db, which already covers manufacturing.db and schema_store.db. Files added under data/ are now visible to git again.

Discovery: no test loads a committed fixture

$ grep -rn "chroma_db\|last_reasoning" packages/*/tests packages/*/src conftest.py pytest.ini
packages/cli/src/nl2sql_cli/commands/run.py:91,97 # WRITES last_reasoning.json at runtime
packages/core/src/nl2sql/common/settings.py:17 # "./chroma_db" default setting value

Both are runtime paths, not committed fixtures.

$ grep -rn "manufacturing.db" packages conftest.py
(zero hits)

The plan said to keep data/manufacturing.db tracked only if tests use it. Nothing references it, so it is untracked. data/schema_store.db was never tracked; every schema_store.db reference is tmp_dir / "schema_store.db" in a runtime temp directory.

Verification

• git ls-files | wc -l: 411 before → 409 after — exactly the two intended removals.
• No previously tracked file became ignored (piped all of git ls-files through git check-ignore; only the two targets matched).
• Both files confirmed still present on disk after git rm --cached.
• git status clean after commit.

Test results

Command: OPENAI_API_KEY= pytest -m "not integration" -q (pytest-randomly active, so ordering varies).

Before (baseline, index and .gitignore temporarily reverted):

200 passed, 1 skipped, 48 deselected, 1 xfailed, 27 warnings in 11.16s

After (run twice for randomised ordering):

200 passed, 1 skipped, 48 deselected, 1 xfailed, 27 warnings in 11.34s
200 passed, 1 skipped, 48 deselected, 1 xfailed, 27 warnings in 5.02s

Identical before and after. The pre-existing skip (pyodbc driver not installed) and xfail (PhysicalValidator not wired into the default SQL subgraph) are unrelated to this change.

No CHANGELOG.md or version changes. No generated demo data committed.

[nadeem4]

Task 1 of the 14-task implementation plan: repo hygiene — stop tracking build artifacts.

What changed

• Untracked data/manufacturing.db and last_reasoning.json via git rm --cached (both remain on disk).
• Tidied .gitignore: grouped the loose trailing entries under existing comment headings, switched directory patterns to trailing-slash form, added .pytest_cache/, and added a trailing newline (the file previously had none).

Most of the plan's paths were already untracked

The plan was written against an older repo state. git ls-files shows chroma_db/, site/, logs/, artifacts/, __pycache__/ and .pytest_cache/ have zero tracked files already. Only two artifact paths were actually tracked, and those are the two removed here. The .gitignore entries for the rest are retained/tidied so they stay ignored.

Deviations from the plan text

(a) artifacts/ is kept anchored as /artifacts/ — the plan's bare artifacts/ is not used.

A bare artifacts/ matches a directory of that name at any depth, including the source package packages/core/src/nl2sql/execution/artifacts/, which would silently make new files there unaddable. This was a real bug fixed in an earlier PR. Verified both forms in isolation:

$ printf 'artifacts/\n' > /tmp/t
$ git -c core.excludesFile=/tmp/t check-ignore -v --no-index \
packages/core/src/nl2sql/execution/artifacts/store.py
/tmp/t:1:artifacts/ packages/core/src/nl2sql/execution/artifacts/store.py # MATCHES - hazard
$ printf '/artifacts/\n' > /tmp/t2
$ git -c core.excludesFile=/tmp/t2 check-ignore -v --no-index \
packages/core/src/nl2sql/execution/artifacts/store.py
(no output, exit 1) # safe

With the anchored form as committed, git check-ignore reports nothing for store.py, parquet.py, __init__.py, or a hypothetical new file in that directory.

(b) The bare data line is replaced with specific generated paths.

The plan says "do not ignore data/ wholesale", but .gitignore already contained a bare data line doing exactly that (git check-ignore attributed data/schema_store.db and data/demo_lite/ to .gitignore:27:data). It is replaced with the specific generated paths — data/demo_lite/, data/vector_store_demo/ — alongside the pre-existing data/*.db, which already covers manufacturing.db and schema_store.db. Files added under data/ are now visible to git again.

Discovery: no test loads a committed fixture

$ grep -rn "chroma_db\|last_reasoning" packages/*/tests packages/*/src conftest.py pytest.ini
packages/cli/src/nl2sql_cli/commands/run.py:91,97 # WRITES last_reasoning.json at runtime
packages/core/src/nl2sql/common/settings.py:17 # "./chroma_db" default setting value

Both are runtime paths, not committed fixtures.

$ grep -rn "manufacturing.db" packages conftest.py
(zero hits)

The plan said to keep data/manufacturing.db tracked only if tests use it. Nothing references it, so it is untracked. data/schema_store.db was never tracked; every schema_store.db reference is tmp_dir / "schema_store.db" in a runtime temp directory.

Verification

• git ls-files | wc -l: 411 before → 409 after — exactly the two intended removals.
• No previously tracked file became ignored (piped all of git ls-files through git check-ignore; only the two targets matched).
• Both files confirmed still present on disk after git rm --cached.
• git status clean after commit.

Test results

Command: OPENAI_API_KEY= pytest -m "not integration" -q (pytest-randomly active, so ordering varies).

Before (baseline, index and .gitignore temporarily reverted):

200 passed, 1 skipped, 48 deselected, 1 xfailed, 27 warnings in 11.16s

After (run twice for randomised ordering):

200 passed, 1 skipped, 48 deselected, 1 xfailed, 27 warnings in 11.34s
200 passed, 1 skipped, 48 deselected, 1 xfailed, 27 warnings in 5.02s

Identical before and after. The pre-existing skip (pyodbc driver not installed) and xfail (PhysicalValidator not wired into the default SQL subgraph) are unrelated to this change.

No CHANGELOG.md or version changes. No generated demo data committed.