[scks1]

Adds test_numeric_report_issue_regressions.py: 14 regression tests for NumericDistribution anchored to four open issues. Happy to fold these into test_numeric_report.py instead if you prefer one file. Every test in this PR passes on current master.

#166 — one test locks the documented behaviour for well-separated repeated values: they stay sorted and are moved by at most the 1e-6 nudge. Two further characterization tests for the in-bounds branch of _check_and_update_repeating_values currently fail on master and are deliberately not included, so this PR keeps the suite green:

• Declaring 9.99999995 @0.85, 10 @0.9, 10 @0.95 (bounds 0–100) validates to values [9.99999995, 9.9999999, 9.99999995, …] — no longer increasing: the downward nudge crosses a distinct neighbour, and the result is never re-checked after the repair.
• Declaring repeats at 0.0000005 against a closed lower bound of 0 validates to [-4e-07, -3e-07, …] — the nudge moves in-bounds values below the closed bound.

Those two tests assert the invariants (strictly increasing declared values; in-bounds inputs stay within closed bounds) rather than any particular fix. Happy to add them alongside whatever robust solution #166 lands on, or to share them now if they are useful.

#212 — extends the existing repeated-value coverage to the 201-point CDF where 200 bins fit the 0–20 range exactly (mass lands in the bucket ending at 12.0), and locks that concentrated forecasts and aggregation of identical near-bound forecasters produce strictly increasing CDFs (problem 2's workflow error does not reproduce on current code).

#209 — first tests for _check_percentile_spacing: declared heights closer than 5e-05 are rejected, and generated CDFs respect the same floor. (The original report points at a Discord thread, so these lock the spacing behaviour the title describes rather than claiming to close the issue outright.)

#210 — for smooth closed-bound and log-scaled forecasts the PMF shows no edge elevation; where a forecast genuinely declares heavy near-bound mass, standardization caps every bin at the documented maximum.

Test run (Python 3.12, repo lockfile): new file — 14 passed; existing test_numeric_report.py unchanged at 147 passed / 96 skipped, and unaffected when run together (161 passed, 96 skipped).

Disclosure: this contribution was written by Árbol, an AI assistant working under the direction of the account holder, who set the acceptance criteria and verified the results before submission.