[maximusunc]

Fixes #192.

Problem

shared_count_query() decoded shared_counts.lmdb values with int.from_bytes(raw, "little"), but the deployed artifact stores each count as ASCII decimal text carried over from the upstream Redis strings. Every emitted num_publications was therefore inflated, e.g.:

Stored	Decoded before this PR
b"1"	49
b"80"	12,344
b"10584" (Metformin × Diabetes Mellitus)	224,281,309,233

This also saturated the literature weight in scoring, so it affects ranking.

Confirming the stored format

Read directly from the deployed LMDB files:

metformin: {'index': 25657, 'pmc': 27317}  diabetes: {'index': 267, 'pmc': 614535}
pair raw: b'10584'
first 10 entries:
   b'0_1' b'80'
   b'0_10' b'2293'
   b'0_100' b'27082'
   ...

The per-curie counts match the values reported from CI/TEST/PROD in #192, and every sampled shared-count value is decimal text. The stored data is correct; the reader was wrong.

Changes

• workers/aragorn_omnicorp/worker.py: parse values with int(raw); anything that isn't a plain run of ASCII digits (binary bytes, sign, whitespace, underscores, empty) raises ValueError rather than being guessed at. Docstrings updated to describe the real format.
• tests/unit/aragorn/test_aragorn_omnicorp.py: the fixture now writes ASCII decimal (the old one wrote little-endian bytes, so it only tested the reader against its own assumption). Adds a test using raw key/value bytes from the production artifact, including 267_25657 → b"10584", and a parametrized test that malformed values are rejected.

A ValueError is caught by the overlay's existing outer try/except, which logs it and returns the message without Omnicorp co-occurrence edges rather than emitting wrong counts.

Testing

• tests/unit/aragorn/test_aragorn_omnicorp.py: 20 passed. Against the old decoder, 9 of these fail.
• black --check passes.
• Full unit suite: 10 failures in tests/unit/arax/test_query_parity.py, identical with and without this change, so they're unrelated (likely the local Python 3.13 vs the targeted 3.12).

After deploy, the #192 reproduction request should return num_publications = 10584. Expect rankings to shift, since the inflated counts were feeding the scorer.

---

Generated by Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 53.94%. Comparing base (1d5ed6c) to head (d3fe9a3).
⚠️ Report is 1 commits behind head on devel.

Files with missing lines	Coverage Δ
workers/aragorn_omnicorp/worker.py	76.10% <100.00%> (+0.17%)	⬆️

---

Continue to review full report in Codecov by Harness.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 010f7ef...d3fe9a3. Read the comment docs.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.