docs(ci): bind current replay guidance to the selected record

[petalas]

Summary

 README.md -> docs/ci-verification.md
- Current: #543 + #626 and its historical Linux run
+ Current: selected record, #589 + #1094, replay, workflow and source bindings
+ Historical: preserved #543 + #626 Linux evidence

This is a navigation correction with a regression check, not a construction or κ improvement. Only docs/ci-verification.md and tests/test_research_current_status.py change. The test resolves Markdown links against the selected record rather than pinning a numerical constant. The documentation separates workflow configuration, dated local evidence and exact-head hosted execution.

Task and overlap

Task scope-navigation, refined scope selected-ci-navigation/README-to-ci-verification/1094-vs-626, priority revision 4. Chooser nonce 5c1429e0cff797bb5636a2dfd147b412; consolidation track. Scope status is refined, not changed.

Base and refreshed canonical main: 0fdd7d82de03e43c9e5a1000dfbcd380d1b687d2. Contribution: 72deb11ddd9af6ebe0bbad7865905a7d7aca7650. No pending dependency or scientific source is changed. Sources are the selected record and its review, source manifest and integration receipt, plus the workflow at that base.

Canonical Discussion #2211. Pre-submission overlap refresh covered 450 open PR bodies, 905 discussion records and 451 reported PR file lists, ending at 2026-10-15 00:07:48 UTC. No open PR reported a change to this document. These searches do not prove novelty. Nearby navigation work addresses different questions: #2037 covers R439 semantic-D evidence, #2183 covers bootstrap obligations, and #2133 covers P104 correction scopes.

Evidence

Before, the new regression rejected the original introduction, which called #626 current and did not link the selected record. After, the following checks pass on Linux x86_64 with Python 3.13.5:

python3 -B -m unittest discover -s tests -p 'test_research_*.py' -v
# 66 tests: OK, with two existing Python-3.12-only skips.
python3 -O -B -m unittest discover -s tests -p test_research_current_status.py -v
# Six tests: OK.
python3 -W error -X dev -B -m unittest discover -s tests -p test_research_current_status.py -v
# Six tests: OK.
python3 -B scripts/verify_selected_result.py
# PASS selected record, exact kappa, original source inventory and proof supplement.
python3 -B scripts/research_baseline.py
# PASS_RESEARCH_BASELINE_BINDINGS_ONLY.
git diff --check

Five failure controls were rejected: the original introduction, a stale replay command with otherwise correct links, and individually wrong review, integration and source-manifest links. All seven local document links exist, and the documented cold command matches the workflow.

To reproduce the original-introduction rejection on this branch:

python3 -B - <<'PY'
from pathlib import Path
import runpy
import subprocess
import unittest
from unittest.mock import patch
module = runpy.run_path('tests/test_research_current_status.py')
old = subprocess.check_output(
    ['git', 'show', '72deb11d^:docs/ci-verification.md'], encoding='utf-8')
with patch.object(Path, 'read_text', return_value=old):
    suite = unittest.defaultTestLoader.loadTestsFromTestCase(module['CurrentCiNavigationTests'])
    result = unittest.TextTestRunner().run(suite)
if result.wasSuccessful() or result.errors:
    raise SystemExit('Expected a clean regression failure')
print('PASS_STALE_INTRODUCTION_REJECTED')
PY

No full supplier replay, Lean proof or hosted-CI source reconciliation was run for this navigation-only patch. The integration receipt's pending hosted status is explicitly described as dated evidence, not current run status. All mathematical and physical obligations remain unchanged. No separate independent reviewer was available.

Credit

The selected construction is Rohan Arun's #1094 on Jeremy Kothe's cumulative complex supplier and Hao Yan's controller, with eumemic's #589/#555 bit lineage. The #626 Linux reconciliation and Anas Khan's #90 historical-reproduction clarification remain predecessor work. Historical evidence, source pins, licenses and the selected frontier are unchanged.

Open next questions

A hosted-CI coverage claim still needs the exact tested head, input bindings and logs. This patch provides the navigation, not that reconciliation. There is no new supplier or proof obligation to hand off.

Merge danger

This is a reversible documentation and test change. Its effect is limited to CI navigation and research regression checks. No workflow, generated artifact or selected-result field changes.