Accept test values within 20% or within 5 x the combined error bar
(sqrt(2) x the test run's ERR, as no targets carry _ERR yet), to see how
the error-bar based acceptance behaves across all platforms.

Co-Authored-By: Claude Opus 5.5 <[email protected]>

Both spellings are accepted, so instruments can move to the new tags
gradually:
- mctest: %Test: lines are tests like %Example:, %TestScan: lines scans
  like %Scan: (incl. NAME_ERR targets and multi-line value lists)
- mccode-git-diff-code: %Test:/%TestScan: lines are test definitions too
- mcdoc (mccodelib parse_header/format_examples): the test section also
  starts at %Test:/%TestScan:, and the lines are labelled "Test:" and
  "TestScan:" (scan lines were labelled "Scan:" before); Tomography's
  README updated to the new label

Checked: all four tags parse in mctest, mcdoc and the diff tool; mcdoc
output for all 452 example instrument headers is unchanged apart from
the scan label.

Co-Authored-By: Claude Opus 5.5 <[email protected]>

A value from very few rays has a huge error bar, so S x ERR would accept
almost anything (e.g. ISIS_CRISP at -n100: 28% of target, but only -2.2
sigma). --sigma now only uses error bars with ERR/|I| <= 25%
(SIGMA_MAX_RELERR, ~1/sqrt(N) for N >~ 16 unweighted events), for the
test value and, when given, the target's _ERR; I = 0 or a missing ERR
also count as unusable. Otherwise the value is judged by the 20% rule.
The log says when this happens. mcviewtest applies the same rule.

Checked: basictest instruments at -n1e6 --mpi=auto unchanged (all pass
on "20% or 5 sigma"); at -n100 the starved values fall back to 20% and
fail.

Co-Authored-By: Claude Opus 5.5 <[email protected]>

- --sigma is renamed --nsigma: it is a maximum z-score, not a sigma
  value (no alias, it was not in use yet). CI testsuites use --nsigma 5.
- --pvalue P: the same threshold as a two-sided Gaussian p-value
  (0.0027 = 3 sigma, 5.7e-7 = 5 sigma), converted with
  statistics.NormalDist; cannot be combined with --nsigma.
- Error bars are only used from >= 100 effective events,
  N_eff = (I/ERR)^2 (was ERR/I <= 25%, ~16 events), matching the ~100
  event floor proposed for --statfactor; below that the 20% rule
  applies. Noisy tests like BNL_H8_simple (N_eff ~19 at 1e6) are thus
  judged by the 20% rule.

Co-Authored-By: Claude Opus 5.5 <[email protected]>

--statfactor F multiplies the ncount of every test by F: the general
ncount (-n, default 1e6) for single-value tests, and the per-point ncount
of each scan (the scan line's -n, else at most 1e5). Run timeouts grow
with F, and the factor is part of the result directory's name.

With --nsigma/--pvalue, a target without NAME_ERR is assumed to come from
the unscaled ncount, i.e. sigma_target = sigma_test * sqrt(F), so the
combined error bar is sqrt(1 + F) * ERR_test (sqrt(2) for F = 1, as
before). The test run's own ERR follows the scaled statistics by itself.
mcviewtest applies the same rule from the stored statfactor.

Checked on the McStas basic-test instruments + SE_example scans:
F = 10 all pass (incl. BNL_H8_simple, which fails 20% at 1e6); F = 0.1
gives 5 failures that fall below 100 effective events and fail 20%.
File-input tests (MCPL, KDSource) do not change with F, so for them only
the assumed target error changes.

Co-Authored-By: Claude Opus 5.5 <[email protected]>

Explains for statistics-savvy readers how mctest's --nsigma, --pvalue and
--statfactor judge test values against their targets: where I and ERR
come from (checked against mcestimate_error), the decision table, the
sqrt(1+f) assumption for unrecorded target errors, the N_eff >= 100 gate
(Kish effective sample size), the n-sigma / p-value equivalence and the
multiple-comparisons argument for 5 sigma in CI, known limitations
(SPLIT-correlated weights, systematic target drift, file-input tests, MPI
random streams, Gaussian approximation), and measured results on the CI
basic tests at statfactor 0.1, 1 and 10. Linked from mctest.md.

Co-Authored-By: Claude Opus 5.5 <[email protected]>

mctest.md:
- %Test:/%TestScan: are the documented test tags (older names %Example:
  and %Scan: still accepted)
- the scan example uses the {}-enclosed target list the parser requires
  (the old example without braces would not have been recognised)
- a value or scan point passes within 20%, or with --nsigma/--pvalue
  within the statistical tolerance; --nsigma row gives sqrt(1+F) with
  --statfactor F; --skipnontest/--noscans/--strict rows use the new tags
- mcviewtest: one row per test/scan, cells coloured by the rule mctest
  used, per-point tooltip on scan cells
mcrun.md: --scan_split is ignored (with a warning) for an explicit
multi-process --mpi.
README.md: link to the statistical acceptance page.

The .html twins are regenerated from the .md files with python-markdown
(tables, toc, fenced_code) and the existing page head, which reproduces
the committed mcplot.html exactly; mctest.html and mcrun.html had not
followed the earlier %Scan:/--scan_split doc changes. Adds
mctest-statistics.html.

Co-Authored-By: Claude Opus 5.5 <[email protected]>

Converts tools/Python/NAME.md to NAME.html with python-markdown (tables,
toc, fenced_code), keeping each page's existing <head>; mcdoc.md becomes
mcdoc_tool.html, and links to .md files are rewritten to the .html names.
With no arguments all tool docs are converted. Running it on the current
tree reproduces all 9 committed .html pages byte for byte.

Co-Authored-By: Claude Opus 5.5 <[email protected]>

mcstas-pygen turns test lines into McStasScript instr.add_test calls.
It now accepts %Test: as well as %Example:, in file order when both are
used. %TestScan: lines (parameter scans) have no add_test equivalent and
are left alone; "%Test:" does not match them. Warnings, the generated
comment, the --no-tests help text and codegen.md/.html name both tags.

Checked: BNL_H8 with %Test: instead of %Example: generates identical
code; a header mixing %Example:/%Test:/%TestScan: gives the three
single-value tests in file order and the generated script compiles; no
compiler warnings for pygen.c.

Co-Authored-By: Claude Opus 5.5 <[email protected]>

Co-Authored-By: Claude Opus 5.5 <[email protected]>

Co-Authored-By: Claude Opus 5.5 <[email protected]>