feat(validation): deflated_expectancy, the number the monitor should anchor to

[mspinola]

docs/edge_monitor.md carried this at the top of its open list since the monitor shipped: EdgeBaseline took a deflated_expectancy float and nothing in the package produced one. deflated_sharpe corrects a Sharpe and returns a probability, which is the right output for a gate and useless to a monitor. A monitor needs a number in R, and without one the only anchor available was the sample mean, which is the number the parameters were optimized on.

The conversion

It rides on the bar deflated_sharpe already uses. SR0 is the expected maximum per-trade Sharpe of N noise trials; carry it back into R by the winner's own sigma and subtract:

deflated = mu - sigma * SR0

Both functions now call one _expected_max_sharpe, so two corrections for one search cannot disagree about how big the search was.

Three choices worth recording

• It takes trial LOGS, not trial Sharpes, unlike deflated_sharpe. The Sharpes are computed inside so their clock cannot be got wrong: multiplying a per-month Sharpe by a per-trade sigma gives a haircut in no units at all, silently, which is the v0.4.0 units bug in a new costume.
• It is a bias correction, NOT a significance test, and the docstring, the __str__ and the property name all say so. It removes the selection bias a search of this size is expected to produce, so a pure-noise winner still clears zero roughly half the time: measured 56% / 47% / 44% for N = 5 / 20 / 100, against deflated_sharpe correctly calling 0% of the same draws significant. I named the property survives first, which reads as a verdict it does not deliver; it is is_positive, and the measurement that caught this is pinned as its reproducer.
• EdgeBaseline.from_log accepts the result object as well as a float, because handing over the wrong field of a result you already computed anchors the monitor to the pre-correction number while reporting deflated=True.

Two findings from wiring the example to the real function

• The worked example faked this step with raw_mean * 0.8 under a comment beginning "Pretend". With a real 64-config search the haircut is 26% of the raw edge rather than 20%, and the naive baseline flatters by 36% rather than 25%. Every figure in tutorial §14 moved and is re-pinned.
• tests/test_edge_monitor_example.py rebuilt the baseline it exists to pin instead of importing it, so all five assertions passed unchanged while the example they guard printed entirely different numbers. It now imports promoted_book(). A guard that reconstructs what it guards is not a guard.

The payoff, as a test rather than a claim

Run one untouched, fully healthy live book against both baselines: the naive one returns DEGRADED where the deflated one returns HOLDING. An inflated reference does not merely mis-scale the ratios, it manufactures alarms on books that never decayed.

373 tests pass, ruff clean, mkdocs build --strict clean.

🤖 Generated with Claude Code

[mspinola]

Rebased onto #115, which landed while this was open. The two overlap more than the titles suggest and both changes are kept:

#115 moved the ARL budget to calendar years, so every figure in the worked example and tutorial §14 moved again on top of the ones this PR already moved. Recomputed with both applied and re-pinned: h 43.17 → 35.09 sigma, ARL0 3,752 trades (25.0 yr), ARL1 658 trades (4.4 yr), and the halved book now alarms at live trade 276 rather than 1,138. The measured in-control inflation is 1.09x and the shifted figure 0.96x, which I updated rather than keeping #115's 1.01x.

#115 also fixed the dead firing-rate channel from crucible's end, by falling back to exit_date in _trades_per_year. That is the same defect I fixed from the seam end earlier today in crucible-stack#17 and npf#184 (the seam built its TradeLog from bare floats, so no date column existed to read at all). The two fixes are complementary, not redundant: #115 makes the baseline side work for exit-dated logs, and the seam work makes the live side work at all. entry_date stays preferred, so npf continues to supply it.

The headline result is unchanged by the rebase: run one untouched, fully healthy live book against both baselines and the naive one still returns DEGRADED where the deflated one returns HOLDING.

378 tests pass, ruff clean, mkdocs build --strict clean.