analysis

analysis/ — study reports

This folder holds one report per study, an index, the curator's decisions and the open questions. It is the place to start when investigating a result.

• INDEX.md — every study, grouped by theme, with status and key numbers.
• decisions.md — curator rulings (J. Stajich) in date order, with evidence.
• open-questions.md — pending items and what each waits on.
• <date>_<topic>.md — one report per study (related runs are grouped).

Paths

results/<folder>/ paths refer to the main checkout (/bigdata/stajichlab/jstajich/projects/MATPredict/results/). Summaries and small tables of most folders are also committed on the PR #9 branch (polish-scope-cuts) or on a curation branch; a report names the branch when a folder is committed only there. Per-genome TSVs open in any spreadsheet.

Report format (use for every new study)

# <Title>
Status: decided | open | superseded | running

## Question
One or two sentences.

## Data and code version
- Code: commit <sha> (frozen worktree .claude/worktrees/run-<sha>)
- Database: same tree as the code unless stated
- Inputs: genome list path, panel size

## Method
Short numbered steps. Name the rule, threshold and tool.

## Results
Numbers and small tables. Every number cites its source file.

## What changed in detection
Commits that implement a change, if any.

## Limits
What the data cannot show; sample sizes; circularity.

## Curator decisions
Made (date, ruling) / open (what is needed).

## Files
Paths to NOTE.md, TSVs, PDFs.

Rules

• Plain short sentences. No inflated claims. Say what was verified and what was inferred. Mark a number "unverified" if no source file shows it.
• Mark superseded conclusions explicitly, with what replaced them.
• Unpublished manuscript content (resource/MBE_202608/, git-excluded) must never be copied here. Cite the preprint DOI 10.1101/2025.09.11.675505 only for facts it states, and only aggregate comparisons made by this project.

Pre-sign-off regression check (curator ruling 2026-09-30)

Every new or changed curated record, classifier rebuild, paralog class, roster/scope change or detection-rule change must attach a regression check before sign-off. Run scripts/run_regression_panel.sh on baseline and candidate frozen worktrees (or scripts/regression_check.py diff on existing runs), and link diff/summary.md plus each panel's regression_summary.md in the study report. The curator reviews every call_lost, call_gained, idiomorph_changed, confidence_changed and core_model_changed line. The panel is testset/regression_panel.tsv (163 genomes + Zygo 23 both inputs). See docs/regression-check.md.