Complete the Week 4 executable capstone

[skyzh]

Why

Week 4 needed one executable cumulative learner path and a capstone record whose identity does not change with the absolute spelling of the active Python interpreter.

Change

• refresh and run every supplied Week 4 test through the selected day;
• add learner-routed generation witnesses while keeping maintainer mutations outside copied tests;
• add the deterministic, no-download Days 1–9 capstone and its completed Week 4 status/command guide;
• record validation through a stable workspace-relative interpreter identity while executing the active interpreter;
• bind two distinct absolute interpreter spellings to byte-identical capstone records and a fixed output SHA;
• remove the unused evaluator command, script, and orphan task packages.

Final exact-head review

The reviewed PR head is e2b409ae82520416c7173333463c0b5298be1e48 / tree c00b63288e61577c5001344ec452d02594137ab8 on base b852b2087eb8821a7fc57b5108e793ff4242675b. The direct successor from 6a51d9925c9287e5a97932635e2a2f1c2cb0c583 changes only week4-capstone.py and tests_refsol/test_week_4_day_9.py; the cumulative base-to-head scope is exactly 30 paths, including all eleven accepted prose paths. The other 313 records reproduce preservation manifest 8344a306a53fbd27d87cc8ae05c58cb9af46ab5ae934d57b136e5c5cfb4d5fd5.

Two distinct absolute interpreter spellings produce the same newline-terminated 2,995-byte capstone record, receipt IDs, 7207 -> 5203 compaction accounting, selected branch, and E42 range, with no environment path leakage. The fixed SHA-256 is e906b7977917b44eacc7e59932ddf393828614dd3a737a929f077f0d84730901.

The cumulative learner commands pass at 16/24/33/40/46/58/80/86/112; 171/171 focused tests and the full 503 passed / 8 intentional skips are green, as are native build/install, lint/compile/format, mdBook build, sitemap, exact scope/preservation, and hosted run 33393503902. The inherited mdbook test failures remain confined to unchanged setup/Weeks 1–3 snippets; every Week 4 chapter passes.

Wholly fresh independent reviews are GO on this exact head:

• learner contract and walkability: Scholar
• factual consistency and preservation: Oracle
• correctness and causal coverage: Sage
• accounting and methodology: Tuner

This merge does not authorize model download, benchmark, Phase C, publication, release, tag, or deployment.

AI-Assisted: GPT-5.6 Sol + Forge + Sentinel