[wshlavacek]

Draft. Do not merge this PR. It tracks the maintainer follow-ups to #984, and runs CI on them, while #984 waits for the contributor's sign-off (request). When the sign-off arrives, these commits go onto #984's branch and #984 is the PR that merges, so the contributor keeps the credit.

Commits

Commit	Author	What
6c04ed2	srivarshni7	#984, unchanged: the starting_params entry names the sorted files instead of saying "every result file" is sorted (#932)
720609a	maintainer	Names all three sorted files (samples.txt, sorted_params*.txt, bootstrapped_parameter_sets.txt), says the sort is case-sensitive, and says start_point pins only a sampler's first chain. Adds tests/test_result_file_parameter_order.py
1d1f8d4	maintainer	Adds srivarshni7 to CONTRIBUTORS.md

When the sign-off lands

• Rebase 720609a and 1d1f8d4 onto Clarify alphabetical ordering in starting_params docs #984's signed-off commit
• Push them to Clarify alphabetical ordering in starting_params docs #984's branch, srivarshni7:patch-1 (maintainer edits are allowed)
• Merge Clarify alphabetical ordering in starting_params docs #984 with a merge commit, not a squash, so the contributor stays the sole author of their own commit
• Close this PR and delete contrib/984-followups

How it was checked

• Independent oracle. The new test's expected orders are written out by hand, not computed with sorted(). The three parameters have non-overlapping boxes, so a value under the wrong header falls outside the box that header names. Four deliberate breakages of the writers were each caught: a case-insensitive sort, values out of step with the header, starting_params applied in reverse, and starting_params winning over continue_run.
• Sibling paths. Only two writers sort parameter names: Trajectory, which writes sorted_params* and bootstrapped_parameter_sets.txt, and the sampler's samples.txt writer. Every other writer that lists parameters follows declaration order. That covers start_point.txt, credible*.txt, diagnostics.txt, the am files under A_MCMC/ and adaptive_files/, and the Liu–West filter tables. This was confirmed by real de, mh, am, continue_run and bootstrap runs with parameters declared kcat, Km, alpha.
• Option interactions. None: this is docs and a test, with no change under pybnf/, so no changelog fragment is needed.
• Independent review. A reviewer that did not write the change ran the paragraph's claims against the code and wrote the test. It also found three existing defects, now filed as starting_params is not checked at load: an out-of-box value leaves am stuck outside the box for the whole run, extra values are dropped, too few crash with IndexError #1013, start_point.txt labels values set by starting_params as 'sampled' #1014 and bootstrap never resamples an objective = expression fit's data: every replicate refits the full data set #1015.
• Local checks. The new test passes (4 passed), as do the neighbouring tests (107 passed) and test_adaptive_mcmc.py (58 passed). ruff@0.15.14 and sphinx-build -W are clean.

Refs #932, #984.