[fluorine]

Change and mathematical scope

A non-cone sparse checkout that omits a selected root input such as LICENSE cannot complete baseline setup: the preflight emits git sparse-checkout add ., but . is a literal non-cone pattern and leaves the input absent. Repeating the command leaves the same omission.

Generate materialization patterns in the checkout's existing mode. Cone-mode directory commands stay unchanged. Non-cone commands anchor and escape root files and package directories, restoring required inputs without matching similarly named paths. The preflight stays read-only. Sparse/HEAD validation, deletion and staged-drift rejection, batched path validation, schema routing and baseline authentication remain intact; configuration-read failures propagate before a command is published. This is a tooling repair; no new κ or selected-frontier change.

Task and overlap

User-directed checking/process improvement; refined portable-replay suggestion, scope key 91a9a2afec282c926282, chooser seed 6a5ebbaaab445e684ab18ccc01d0aa90, priority revision 7. Concrete object: the emitted command for valid omitted root files in a non-cone checkout. Base/refreshed main: 32890bc5de034ab8fc93ac88ead7279a6a416d04; contribution: efd8edfcc20f8023a8832ba0efa1046143180261.

Targeted all-state PR searches for sparse/materialize, non-cone and root-level baseline work found #3725, #4631 and #4936. This restores the next step their helper promises while retaining their validation contracts. Recent handoffs were consulted; #5463's Windows stream-reader work is separate. Coverage is limited, not an exhaustive novelty certificate: the offline overlap helper had no usable remote snapshot and incomplete local inventory. No mathematical predecessor executable was changed.

Validation

All 16 sparse baseline regression tests pass on Python 3.12, normal and optimized, locally on Windows and in hosted Linux/Windows checks. The workflow lives separately on the contributor fork's main; its checkout is the contribution branch.

python -B -m unittest discover -s tests -p 'test_research_baseline_sparse*.py'
python -O -B -m unittest discover -s tests -p 'test_research_baseline_sparse*.py'
python -B scripts/research_selected_baseline.py

The new real-Git regression follows the emitted command, restores root LICENSE, a spaced/bracketed root filename and a spaced/bracketed package, reaches zero omissions, and keeps unrelated root files, wildcard decoys and nested same-name paths omitted. The old command's failure was reproduced before editing. An injected configuration failure is also rejected. Existing late-deletion, staged-drift and Git-failure controls remain.

Selected source/metadata verification passes locally and in hosted Linux verification, reporting PASS_RESEARCH_BASELINE_BINDINGS_ONLY for selected #5413, κ = 605457647/200000000000. Native Windows required repository-local Git long-path support and restoring missing inputs after an initial path-length failure. This authenticates source/metadata only, not native operators, supplier admission or the complete theorem. git diff --check passes. No full selected cold replay, Lean, Docker or unrelated historical replay was run.

Credit

Prepared for Joseph Anglada (@fluorine) with substantial OpenAI Codex assistance. Retains Douglas Colkitt's helper, Arham-Tahir64's sparse contract and the explicit-inventory/Windows batching predecessor repairs. Apache-2.0.

Open next questions

A separate review could bound the emitted materialization command itself on Windows; this repair does not add an executor or change sparse mode. Large directory-level fetches remain broader than an exact-file export. Neither question needs a scientific frontier change.