Fix partition aggregation and wire median_neighbor_distance

[sanghoonio]

genom_dist currently fails all four checks: aggregation.py has a syntax error, so import bbconf fails and no tests are collected. This fixes that, plus one other bug found while looking into it.

Two commits, independent of each other — the first is just the CI unblock.

---

1. The module doesn't parse — f31557f

5edc86f (Copilot Autofix) removed agg_columns.extend( and its [ from _aggregate_partitions, leaving a dangling list:

agg_columns = []
for name, column in _PARTITION_COLUMNS:
            func.avg(column).label(f"{name}_mean"),   # ← extend([ was here

bedsets.py and bedfiles.py both import this module, so nothing in bbconf imports. a8de95d and fa444b7 were both green; 5edc86f is where it went red.

Restored the wrapper.

The same commit made partition values 100× too small

It also dropped the * 100, acting on a review comment saying the *_percentage columns "appear to already be stored as percentages (0–100)".

They're fractions:

where	what it stores
regionstat.R:212	Freq / length(query)
gtars_backend.py:227	count / total
bedbase-ui selection-stats.tsx:11	renders them as value * 100

Easy to get wrong from inside bbconf — the column name, the old docstring, and the test fixture all suggest otherwise, and all three sources above live in other repos.

What I did: the missing * 100 is only a bug because the keys are named _pct. So rather than putting the multipliers back, I renamed the keys — mean_pct/sd_pct become mean/sd, and partitions now returns fractions like everything else in BedSetDistributions.

Consumers scale for display. bedbase-ui does this at the plot layer (b8a2464 on master), next to the axis label and matching how the local comparison path already works. Nothing was deployed against the old keys.

The docstring now records where the fraction comes from, so the next reader has the evidence to hand.

2. median_neighbor_distance never had a value — 5b35653

_SCALAR_COLUMNS read it from BedStats.tssdist. Nothing writes that column — it appears twice in the whole tree (initial migration, ORM declaration), and BedStatsModel doesn't declare the field, so extra="ignore" drops the value bedboss computes before it can be stored. count() returned 0, and the key was silently omitted from every response.

Added a real column, plus the BedStatsModel field and a migration. Both edits are needed together: with only the model field every insert raises TypeError; with only the column, pydantic keeps dropping the value.

Dropped tssdist rather than renaming it. It's a leftover TSS-distance column superseded by median_tss_dist (both created by the initial migration). Neighbor distance is a different measurement — the gap between consecutive regions within a file, not the distance to the nearest annotated TSS — so a rename would relabel stale TSS values and average them into bedset stats.

Migration chains off c7f3a9e1b204, the only head on this branch.

---

Worth checking before merge

• SELECT count(*) FROM bed_stats WHERE tssdist IS NOT NULL; — should be 0, since no writer has ever existed. If it isn't, that's real data being discarded.
• mean_pct → mean is a response-shape change. bedbase-ui is the only consumer I could find, and it's updated.
• docs/schema.svg still shows tssdist and needs regenerating against a live DB.

Verified

Lint and format pass. Single alembic head; upgrade and downgrade both render correctly offline. ORM metadata matches the migration. Partition and scalar SQL compile against PostgreSQL with no reference to tssdist. median_neighbor_distance added to both test fixtures, so test_upload exercises the model↔ORM coupling.

Not verified: the DB-backed tests. No Postgres available where I was working, so CI is the first real run.

Left alone

Noticed but not touched — happy to open separate PRs:

• per-chrom n in _aggregate_region_distribution is taken from the first row only
• tss_histogram docstring says "summed"; the code computes AVG/SD
• get_distributions() fallback reports n_files=0 for bedsets that do have members
• get_batch docstring claims one round-trip; the two selectinloads make it three
• get(full=True) still returns the distributions blob

🤖 Generated with Claude Code

https://claude.ai/code/session_01PRGhWjGiUXFoygd35QsKQ7

[sanghoonio]

Superseded — pushed the fix directly to genom_dist as 2fd21d3, so it's part of #127 now.

Dropped the median_neighbor_distance schema commit from scope. That work is preserved on fix/median-neighbor-distance (5b35653) if it's wanted later.

🤖 Generated with Claude Code

https://claude.ai/code/session_01PRGhWjGiUXFoygd35QsKQ7