scientific(coverage): bind the exact normal critical value into the prospective design

[seonghobae]

Finding

CoverageCalibrationDesign::tepp_nominal_95_v1() prospectively fingerprints the nominal coverage target, practical band, attempt count, and Monte Carlo precision, but the actual interval-construction critical value is still caller-local in analysis_engine as NORMAL_95_PERCENT_CRITICAL_VALUE = 1.959963984540054.

That leaves the persisted prospective design under-specified: the evidence can claim tepp.coverage.nominal95.v1 while the executor's exact normal-marginal interval rule is represented only indirectly by source_head. A later executor drift in the critical value would change scientific coverage while retaining the same validation-design identity/fingerprint. This is the remaining criterion/construction split after #732/#737 made the prospective validation design durable before execution.

No 10,000-DGP acceptance outcome has been executed yet, so this prospective contract can still be completed without post-outcome threshold tuning.

Required RED / repair

• make the exact two-sided normal critical value an owned field of CoverageCalibrationDesign::tepp_nominal_95_v1();
• expose it through the validation owner and include its exact IEEE-754 binary64 bits in the canonical design fingerprint under a new fingerprint-domain version;
• make the coverage-calibration executor consume the owner-issued critical value rather than a local constant;
• update the pinned fingerprint contract and shard/evidence expectations that derive it, without changing the nominal 0.95 target, practical [0.91, 0.98] band, 10,000 attempted DGPs, or MCSE <= 0.005;
• preserve normal_marginal_interval_bounds(...) as generic numerical arithmetic; do not move estimator/covariance arithmetic into the design owner;
• document that this completes prospective interval-construction identity before the first acceptance run and does not itself provide acceptance evidence.

Boundary

This is prospective design completeness. It does not execute #680, define numerical-failure acceptability, alter the simulation DGP/seed schedule, authenticate the build, or promote a scientific/release claim.

Refs #639 #680 #721 #725 #732 #737.

[seonghobae]

Implemented ordinary-forward on #639 through exact head aeee7cdd2a9ddd6e1f6b24e43f9a580afaf1bb82.

Lineage: source RED 81935e4518bbf9da945d2180cb806dab6828b32e → validation-owner critical value badc4abd0ad3a33e4abf082e05392dd5837e6fd1 → fingerprint-v2 binding beab0f4cdf640e5e5b8e1d8eaaeeab8ee1cd0862 → pinned complete-design fingerprint ba7ec18fe15528434b95fe182c99f87e4bb3217e → analysis executor consumes the owner-issued value a8672cca7d6329c10ccdd293312c03a49a3a896a → CHANGELOG/currentization af83aaa30bc0ee332bbc24750f3904bb8d25205b / 0ff588e6ea4e85d8232732d809d61b8d60af5bc8 → rustfmt-compatible owner-consumer shape aeee7cdd2a9ddd6e1f6b24e43f9a580afaf1bb82.

The exact normal critical value is now 1.959963984540054; the complete tepp.coverage.nominal95.v1 design fingerprint is 4f937a115d273f5e1171de03eeeea92e30681772849a590365f4933668e8e557. Nominal 0.95, practical [0.91, 0.98], 10,000 attempts, and MCSE <= 0.005 are unchanged. The generic interval arithmetic remains in validation_core; only the prospective rule identity moved into the design owner.

This is still Draft source work. No 10,000-DGP acceptance run, numerical-failure acceptance threshold, release, or hosted GREEN is claimed here. Keep #740 open until the delta is normally inherited/landed with exact-head gates and review.