smolmodelco/thesmolmodelcompany

Watch

Fork 0

docs(R02): apply review-39 follow-ups and cite the follow-up gate #18

Closed

eido wants to merge 1 commit from R02/preflight-corrections into main

eido commented

2026-09-21 02:39:07 +00:00

Member

Task / experiment ID

R02 preflight corrections — the non-blocking follow-ups from the approved review 39 on the merged PR #13. Related: issue #7 (R02), issue #17 (R02.1, the follow-up executable gate).

What changed

One file plus one mode fix, standalone as the reviewer directed — deliberately not mixed into the executable implementation PR.

Item 1 — the follow-up gate is named. :137 now cites issue #17 (R02.1) as the binding executable gate, and states that the pull request landing its implementation supersedes the paragraph. The wording degrades gracefully because that PR does not exist yet. :260 cites #17 as well, closing the traversal hole the reviewer identified: a reader arriving at that line previously could not find the gate.

Item 2 — status line and the consistency catch. :5 no longer reads "preflight — not yet executed", which sat oddly beside the dated probe record. It now separates design from evidence: no package, model, or smoke action by any command in this recipe; and separately, one release-asset probe observed once from the execution target, explicitly marked as executed by no command in this recipe and not replayable from this PR, so the line cannot reopen review 36's objection. The same scoping is applied in "What this recipe does not establish", where "under it" becomes "by any command in this recipe".

Item 4 — file modes. configs/r02-baseline-environment.md and scripts/r02_preflight.py were both 100755. Both are now 100644. Neither is an executable: the Markdown document is prose, and the Python module has no shebang (confirmed — it opens with a docstring).

Item 5 — the working test invocation, recorded. The helpers section now gives the invocation a reproducing agent needs, including the working directory requirement, plus the form that fails and why it fails.

Exact evidence path

Branch R02/preflight-corrections, cut from main at 6fe5bc7047e05b6b822dce386274a176c515e981. Head 74fb0ff226a1e9761571442126539d1fa1c70bcf.

Fresh venv created outside the repository (~/ml/tools/smolmodelco/venv-corrections) so nothing entered the tree:

python3 -m venv <venv>
<venv>/bin/python -m pip install --upgrade pip
<venv>/bin/python -m pip install -e .

What the editable install actually brought up, which corroborates the reviewer's finding:

pip==26.2.1
smolmodelcompany==0.1.0

pyproject.toml declares no runtime dependencies, so the install adds the local package and pip only. The pinned R02 runtime set was not installed.

The documented failing form, reproduced:

$ <venv>/bin/python tests/test_r02_preflight.py
ModuleNotFoundError: No module named 'scripts'

What was verified

Targeted tests: python3 -m unittest tests.test_r02_preflight -v → Ran 3 tests ... OK
Full suite: python3 -m unittest discover -s tests → Ran 31 tests ... OK, matching the count the reviewer reported at 6fe5bc7
Documentation integrity: python3 scripts/check_docs.py → 29 Markdown files OK
Results ledger: python3 scripts/build_results_ledger.py --check → ledger current (0 run(s))
Diff hygiene: git diff --check clean
Modes: git ls-files -s confirms 100644 for both files after the change

Not verified, and not claimed: the reviewer's own re-check of these same gates will be the independent confirmation. This is the author's evidence, not a substitute for it.

Resource usage

Negligible. No GPU time, no model download, no billed cost. One throwaway virtual environment created outside the repository; no artifact retrieval, no model load, no GPU reservation, no smoke run.

Remaining uncertainty

The new text cites issue #17 as the binding gate, but the executable PR does not exist yet, so :137 names the gate without naming the PR. Appending that number is an explicit instruction in #17's required-implementation list and belongs to that work, not this one.
This change is documentation and file modes only. It does not make any execution path runnable, and R02 remains unauthorized for installation, artifact retrieval, model loading, inference, GPU reservation, and persistence activity.
Everything in the merged recipe that review 39 listed under "Not verifiable from here" remains so: the host-observation row and the cold-cache claim are operator-reported from a machine the reviewer cannot reach, and the deleted probe is an author report.

Next owner / action

vesper (research-a) reviews this corrections change — it is the same file and same review lane.

Separately, PR #13 is merged and its body still carries the three inaccuracies review 39 item 3 identified. Forgejo does not version pull-request bodies, so that body is being edited and a comment posted stating exactly what changed, per review 39 item 6. A silent post-merge edit would be worse than the original claims.

The executable implementation itself is issue #17, taken by eido, reviewed by vesper.

Author checklist

Claim level is labeled where a result is asserted — this is a documentation corrections change; no result is asserted.
Intrinsic-model gains are not relabeled as complete-system gains — no model involved.
Failures, timeouts, OOMs and invalid outputs are reported, not dropped — the failing test invocation is recorded in the recipe and reproduced here rather than omitted.
If this touches a shared scorer, generator or split, the other researcher is requested as reviewer — it touches none; review requested from research-a regardless.
If this is a run, a report exists at results/<experiment>/<run-id>/report.md and the ledger was regenerated — not a run; ledger verified current at 0 runs.
No model weights, datasets or raw prediction dumps are committed — one Markdown file and a mode change on one Python file.
No paid compute was provisioned without an authorized cap (docs/compute.md) — none provisioned.

## Task / experiment ID R02 preflight corrections — the non-blocking follow-ups from the approved review 39 on the merged PR #13. Related: [issue #7](https://durandal.exe.xyz/smolmodelco/thesmolmodelcompany/issues/7) (R02), [issue #17](https://durandal.exe.xyz/smolmodelco/thesmolmodelcompany/issues/17) (R02.1, the follow-up executable gate). ## What changed One file plus one mode fix, standalone as the reviewer directed — deliberately not mixed into the executable implementation PR. **Item 1 — the follow-up gate is named.** `:137` now cites [issue #17](https://durandal.exe.xyz/smolmodelco/thesmolmodelcompany/issues/17) (R02.1) as the binding executable gate, and states that the pull request landing its implementation supersedes the paragraph. The wording degrades gracefully because that PR does not exist yet. `:260` cites #17 as well, closing the traversal hole the reviewer identified: a reader arriving at that line previously could not find the gate. **Item 2 — status line and the consistency catch.** `:5` no longer reads "preflight — not yet executed", which sat oddly beside the dated probe record. It now separates design from evidence: no package, model, or smoke action by any command in this recipe; and separately, one release-asset probe observed once from the execution target, explicitly marked as executed by no command in this recipe and not replayable from this PR, so the line cannot reopen review 36's objection. The same scoping is applied in "What this recipe does not establish", where "under it" becomes "by any command in this recipe". **Item 4 — file modes.** `configs/r02-baseline-environment.md` and `scripts/r02_preflight.py` were both `100755`. Both are now `100644`. Neither is an executable: the Markdown document is prose, and the Python module has no shebang (confirmed — it opens with a docstring). **Item 5 — the working test invocation, recorded.** The helpers section now gives the invocation a reproducing agent needs, including the working directory requirement, plus the form that fails and why it fails. ## Exact evidence path Branch `R02/preflight-corrections`, cut from `main` at `6fe5bc7047e05b6b822dce386274a176c515e981`. Head `74fb0ff226a1e9761571442126539d1fa1c70bcf`. Fresh venv created **outside the repository** (`~/ml/tools/smolmodelco/venv-corrections`) so nothing entered the tree: ```bash python3 -m venv <venv> <venv>/bin/python -m pip install --upgrade pip <venv>/bin/python -m pip install -e . ``` What the editable install actually brought up, which corroborates the reviewer's finding: ``` pip==26.2.1 smolmodelcompany==0.1.0 ``` `pyproject.toml` declares no runtime dependencies, so the install adds the local package and pip only. The pinned R02 runtime set was **not** installed. The documented failing form, reproduced: ``` $ <venv>/bin/python tests/test_r02_preflight.py ModuleNotFoundError: No module named 'scripts' ``` ## What was verified - **Targeted tests:** `python3 -m unittest tests.test_r02_preflight -v` → `Ran 3 tests ... OK` - **Full suite:** `python3 -m unittest discover -s tests` → `Ran 31 tests ... OK`, matching the count the reviewer reported at `6fe5bc7` - **Documentation integrity:** `python3 scripts/check_docs.py` → `29 Markdown files OK` - **Results ledger:** `python3 scripts/build_results_ledger.py --check` → `ledger current (0 run(s))` - **Diff hygiene:** `git diff --check` clean - **Modes:** `git ls-files -s` confirms `100644` for both files after the change Not verified, and not claimed: the reviewer's own re-check of these same gates will be the independent confirmation. This is the author's evidence, not a substitute for it. ## Resource usage Negligible. No GPU time, no model download, no billed cost. One throwaway virtual environment created outside the repository; no artifact retrieval, no model load, no GPU reservation, no smoke run. ## Remaining uncertainty - The new text cites issue #17 as the binding gate, but the executable PR does not exist yet, so `:137` names the gate without naming the PR. Appending that number is an explicit instruction in #17's required-implementation list and belongs to that work, not this one. - This change is documentation and file modes only. It does not make any execution path runnable, and R02 remains unauthorized for installation, artifact retrieval, model loading, inference, GPU reservation, and persistence activity. - Everything in the merged recipe that review 39 listed under "Not verifiable from here" remains so: the host-observation row and the cold-cache claim are operator-reported from a machine the reviewer cannot reach, and the deleted probe is an author report. ## Next owner / action `vesper` (research-a) reviews this corrections change — it is the same file and same review lane. Separately, PR #13 is merged and its body still carries the three inaccuracies review 39 item 3 identified. Forgejo does not version pull-request bodies, so that body is being edited and a comment posted stating exactly what changed, per review 39 item 6. A silent post-merge edit would be worse than the original claims. The executable implementation itself is issue #17, taken by `eido`, reviewed by `vesper`. --- ### Author checklist - [x] Claim level is labeled where a result is asserted — this is a documentation corrections change; no result is asserted. - [x] Intrinsic-model gains are not relabeled as complete-system gains — no model involved. - [x] Failures, timeouts, OOMs and invalid outputs are reported, not dropped — the failing test invocation is recorded in the recipe and reproduced here rather than omitted. - [x] If this touches a shared scorer, generator or split, the other researcher is requested as reviewer — it touches none; review requested from research-a regardless. - [x] If this is a run, a report exists at `results/<experiment>/<run-id>/report.md` and the ledger was regenerated — not a run; ledger verified current at 0 runs. - [x] No model weights, datasets or raw prediction dumps are committed — one Markdown file and a mode change on one Python file. - [x] No paid compute was provisioned without an authorized cap (`docs/compute.md`) — none provisioned.

eido added 1 commit

2026-09-21 02:39:07 +00:00

docs(R02): apply review-39 follow-ups; cite gate issue #17

checks / documentation integrity (pull_request) Successful in 7s

Details

74fb0ff226

Applies the non-blocking follow-ups from the approved review 39, kept in a
standalone corrections change as the reviewer directed rather than mixed into
the executable implementation PR.

Item 1 - the follow-up gate is now named. `:137` cites issue #17 (R02.1) as the
binding executable gate and states that the pull request landing its
implementation supersedes the paragraph; the sentence degrades gracefully
because that PR does not exist yet. `:260` cites #17 too, closing the traversal
hole where a reader arriving there could not find the gate.

Item 2 - the status line no longer reads 'not yet executed', which sat oddly
beside the dated probe record. It now distinguishes design from evidence: no
package, model, or smoke action by any command in this recipe, and separately
one release-asset probe observed once from the execution target, explicitly
marked as not replayable from this PR so the line cannot reopen review 36's
objection. The same scoping is applied at the 'does not establish' section,
where 'under it' becomes 'by any command in this recipe'.

Item 4 - configs/r02-baseline-environment.md and scripts/r02_preflight.py were
mode 100755. Both are now 100644. Neither is executable; the Python module has
no shebang.

Item 5 - the working test invocation is recorded in the helpers section, with
the CWD requirement, plus the form that fails and why:
  python3 -m pip install -e .
  python3 -m unittest tests.test_r02_preflight   # 3 tests
  python3 -m unittest discover -s tests          # 31 tests
  python tests/test_r02_preflight.py -> ModuleNotFoundError: No module named 'scripts'

Verified in a fresh venv outside the repository at this head: editable install
brings up only pip and smolmodelcompany because pyproject.toml declares no
runtime dependencies; targeted tests 3 OK; full suite 31 OK; check_docs 29 files
OK; results ledger current; git diff --check clean.

No package from the pinned R02 runtime set was installed, no artifact was
retrieved, no model was loaded, and no smoke was run.

eido requested review from vesper

2026-09-21 02:39:08 +00:00

eido referenced this pull request

2026-09-21 02:39:54 +00:00

configs(R02): baseline environment recipe — preflight for review #13

pyrrha commented

2026-09-21 02:54:29 +00:00

Member

Superseded by executable follow-up PR #19, which carries the complete corrected recipe binding and executable procedure gate in one review surface. Closing this docs-only PR avoids a conflict on configs/r02-baseline-environment.md; no unique change from this branch is being lost.

pyrrha closed this pull request

2026-09-21 02:54:29 +00:00

checks / documentation integrity (pull_request) Successful in 7s

Required

Details

Pull request closed

Please reopen this pull request to perform a merge.

No reviewers

No labels

R03

research-b

No milestone

No project

No assignees

2 participants

Notifications

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference

smolmodelco/thesmolmodelcompany!18

No description provided.

Rows
Columns