# Rotation: operational readiness and complete-calendar evidence

The retained full sample does not support a paired performance ranking. All 8,640 registered comparison rows remain unavailable. Earlier snapshot diagnostics are retained under their original protocol; they do not replace the new complete-calendar checks.

## Scope and observed findings

The historical-prefix audit covers 2,829 sessions from 2015-01-05 through 2026-08-24, 43 distinct ETFs and 542 executable weekly signals. Core, candidate and industry panels overlap. Of 31,978 symbol-week eligibility records, 12,998 differ from the retained snapshot rules. Each reference baseline has 336 unavailable signal weeks, from 2016-01-15 to 2023-09-08. Missing amount windows and an ineligible defensive ETF can affect the same week, so reason counts overlap.

The frozen numerical rule requires 252 price observations, complete amounts for the preceding 252 sessions, and median traded amount of at least CNY 10 million. This recomputation does not establish historical delivery, original adjustment-factor vintages, historical membership, mapping validity or complete delisted-fund coverage. Evidence dimensions remain separate. A dimension with no records is explicitly unknown, rather than accepted.

## Conditional comparisons

The budget comparator uses broad-market ETF 510300 and bond ETF 511260 to follow the reference strategy's target risky weight. It measures selection conditional on that strategy's budget path; it cannot independently attribute market timing skill. Actual exposures can differ after price movement and execution.

The separate risk comparison applies an 8% estimated annual ceiling to both arms using the preceding 60 complete return sessions, sample covariance and 252-session annualization. Missing history makes a required target unavailable. Bond-only estimated risk above the ceiling is also unavailable. Equal estimated ceilings do not imply equal realized volatility.

The primary case uses 5 basis points and next-session-close execution. Predeclared sensitivities use 0 or 10 basis points, or a two-session delay. Both arms must use the identical full evaluation calendar. A missing required target or execution invalidates the full case; a shorter favorable interval is not substituted. Annual, trailing and overlapping rolling windows preserve those availability decisions. Rolling windows are descriptive, not independent experiments.

## Prospective collection

Preparation and independent input acceptance are distinct from weekly collection and settlement. A Sunday bootstrap can verify interfaces and retained inputs, but cannot prove that weekday data arrives before the proposed 19:00 Shanghai cutoff. The public release records an observed operational snapshot, not a live feed. Capture remains unarmed until real weekday readiness and a matching future-effective registration exist. Zero collected or settled weeks means no prospective performance is available.

## Publication and reproduction boundary

The downloadable reviewed release provides protocol and evidence digests, aggregate rows, availability reasons and review time. The website validates the recursive field allowlist, hashes and paired calendars before exposing additional interactive windows. CSV exports preserve unavailable values as empty fields with explicit status. Private source locators, holdings, vendor raw rows and credentials are excluded. Existing releases remain immutable. Comparisons across changed protocols or source-trust states do not produce numeric performance deltas.