Allow research task selection from sparse checkouts

[Arham-Tahir64]

Change and mathematical scope

The task picker currently rejects a sparse checkout when any card's first-step script, starting point or priority log is absent, even if that card is unrelated to the requested task. It now accepts an absent reference only when it is a regular file in HEAD and has a matching, stage-zero skip-worktree index blob. Sparse omissions are explicitly listed in the JSON receipt; the seed, track, methods and scope keys are unchanged.

This is a tooling repair. It validates catalogue references, not replay dependencies; users must materialize their chosen task's inputs before execution. Scientific sources, allocations, selected records and source pins are unchanged. No supplier admission or new kappa is claimed.

Task and overlap

• Base and refreshed canonical main: 18aacb215fc3717b06be8ceb9fca62aeca0d3e99, priority revision 6.
• Task/scope: portable-replay / chooser-sparse-catalogue-regular-head-files; changed scope from the adversarial-audit draw, nonce 6c43ccec08828911d1cc13913048461f.
• The live check showed the inspected scatter and corrected-census follow-ups already covered by Enforce simple scatter supports at opt-in checker and exporter boundaries #2992/Audit: exact scatter parity contract at the checker/exporter boundary (no new kappa) #3004/Audit fractional label coercion at the bit checker/exporter boundary #3100 and Independently corroborate the corrected seven-label word exclusion #3005. Those retain their results and credit. This change instead fixes the independently reproduced onboarding failure encountered while using a sparse checkout after disk exhaustion.
• Domain and stop rule: catalogue path availability and task-selection receipts only, using real temporary Git repositories. No scientific replay provisioning or catalogue reallocation.
• Canonical scope/coordination. Final overlap at 2026-10-17 00:45 UTC covered 253 open PRs and 1,861 discussion/comment/reply records, complete pagination and no warnings. Sparse local-note coverage and no code-diff search remain limits. No competing chooser fix was found.

Validation

Passed locally on macOS arm64, Git 2.53.0, CPython 3.12.14:

python3.12 -B -m unittest discover -s tests -p 'test_research_assignment.py' -v
python3.12 -B -m unittest discover -s tests -p 'test_research_coordination.py'
python3.12 -B -m unittest discover -s tests -p 'test_research_onboarding.py'
git diff --check

All 50 tests pass without skips (16 assignment, 8 coordination, 26 onboarding). New controls cover non-cone sparse checkout, cone sparse index, unchanged full/sparse CLI receipts, ordinary deletions, unknown paths, staged replacements, staged-only entries, omitted symlink entries, directory/external alias substitutions and Git failure. An original-source comparison rejects the same sparse fixture that the changed helper accepts. Existing full-checkout behavior remains available without Git path lookup when all references are present.

The actual sparse checkout also successfully runs:

python3 -B scripts/choose_research_task.py --seed 6c43ccec08828911d1cc13913048461f

It reports 34 omitted catalogue paths and retains all three original adversarial-audit scope keys and the unchanged selected-record SHA256 bb2eaa58647b353f88d0aec2eafb9d30df1be53589231b23e12ceef3e4bad129. The assignment tests also passed on CPython 3.14.7 before the final added cone-index control; the full 50-test result above is the canonical Python 3.12 run. Hosted CI, Windows execution and the selected mathematical cold replay are not claimed.

Credit and remaining work

Credit Douglas Colkitt and the repository onboarding contributors for the picker and standing procedure. Arham Tahir prepared this focused fix with substantial OpenAI Codex assistance; Apache-2.0. This is author-checked, not independent human review.

Scientific dependency hydration remains a separate task; the new receipt explicitly says the catalogue check cannot establish replay readiness. No additional research package or mathematical obligation is introduced. Workflow feedback and scope-change rationale.