Refresh README, ROADMAP and methodology against what actually shipped

[emrefbulut]

Documentation only — no source file is touched. Stacked on #18; the base retargets to main once that merges.

1. README (af5b513)

Four problems, one of them actively misleading.

measure-leakage was documented as running "at split seed 42 and train seed 0". That is the Phase 5 regression described as if it were the design. The command measures 15 seed pairs; the single pair was the bug this project has spent the last stretch removing. A README that documents the defect as the feature is worse than one merely out of date.

also fixed	was	now
--force	no limits stated	applies to categories 2–5; 1 and 6 refused with a reason
command tour	no iqforge train	added — the rest of the README and CONTRIBUTING both use it
roadmap boxes	no --group-by, no CUDA, no frequency-aware labelling	all three listed, ticked per reality

On the status block: it still says 0.4.0, deliberately. __version__, CITATION.cff and the newest tag all agree, and docs/release-notes/v0.5.0.md says on its first line that it is an unpublished draft. Changing it to 0.5.0 would make the README claim a release that does not exist. What it now adds is that main carries unreleased work, so a reader can tell which parts are ahead of the wheel they installed.

2. ROADMAP (baa2602)

"Repeat the leakage measurement on a real recording" sat unchecked as though nothing had happened. Half of it shipped: the stride sweep ran on DASH7 and LoRaIQ, both tables are in artifacts/, and LoRaIQ at 7/8 overlap is the first individually significant real-data result here (+9.6 pp ± 2.7, t = 3.5).

Now two entries — the overlap half ticked with its numbers and its remaining limit (intermediate overlaps unresolved at 15 seed pairs), and the open item narrowed to the SNR half, pointing at docs/leakage-real-snr.md where the seed count and SNR list are locked before the run. The parity-gate re-measurement is added as its own finished entry.

The versioning footnote named 0.3.x / 0.4.0 as things to cut; 0.4.0 is out.

3. methodology reader's guide (5957658)

1100+ lines, nine sections, no contents. A question → section table, because "Measurement 1 — accuracy inflation against SNR" does not tell someone who wants to know how much a leaky split invents that this is their section. It also flags §7 (mistakes shipped) and §9 (what the numbers do not support) as worth reading regardless.

Anchors verified against GitHub's slug rule rather than assumed — my first checker collapsed consecutive spaces and flagged two correct anchors as dead. The em dash in those headings leaves a double space, and GitHub makes one hyphen per space without collapsing.

4. Repo sweep (24b0db3)

• No dead relative links anywhere in *.md, docs/, .github/ — checked mechanically.
• Version strings are consistent at 0.4.0 across README, CITATION.cff, ROADMAP and docs/publishing.md.
• One expired statement: docs/phase5-phase6-checklist.md reports readiness as of 2026-08-19 and says no next-version release note exists. It does now. The file is a dated record, so correcting the line in place would falsify it — a superseded note at the top says the phase has since been executed and re-verified, names the line most likely to mislead, and points at CHANGELOG [Unreleased] and ROADMAP for current state.
• README.md:304 "It does not train anything" was checked and is correct — it describes iqforge audit, which does not.

Verification

ruff check .            All checks passed!
ruff format --check .   82 files already formatted
pytest -q -rs           385 passed, 1 skipped
git diff --stat         README.md, ROADMAP.md, docs/methodology.md, docs/phase5-phase6-checklist.md

Reported, not changed

docs/leakage-real-snr.md still says the SNR grid "has not been run". A completed 180-run SNR sweep exists uncommitted in a sibling worktree, so from this repository's point of view the statement is still true. Committing that result is a decision for you, not a docs fix.

🤖 Generated with Claude Code