fix(scoring): parse underscores in fractional upstream constants

[bohdansolovie]

Closes #992

Summary

#969 added underscore support to parsePythonNumberConstants for integer parts (1_500_000) and scientific notation (1e-9), but the fractional digit classes still rejected _. Per PEP 515, underscores are valid between digits in any part of a numeric literal.

A valid upstream constant like RATE = 0.000_001 was truncated at the first fractional underscore and read as 0, silently skewing score previews and breakdowns that depend on the fetched scoring snapshot.

Root cause

// src/scoring/model.ts — before
const match = line.match(/^([A-Z][A-Z0-9_]+)\s*=\s*([-+]?(?:\d[\d_]*\.?\d*|\.\d+)(?:[eE][-+]?\d+)?)/);
//                                                          ^ int allows _     ^ frac: no _   ^ bare decimal: no _

Python literal (PEP 515)	True value	Before (bug)	After (fix)
RATE = 0.000_001	0.000001	0	0.000001
SCALE = 3.14_15	3.1415	3.14	3.1415
VAL = 1_000.000_5	1000.0005	1000	1000.0005
BARE = .5_0	0.5	0.5 (unchanged)	0.5

Integer grouping (1_500_000), plain floats (0.90), and exponents (5.8e1, 1e-9) continue to parse as before.

Changes

• src/scoring/model.ts — extend fractional digit classes to allow _ in both 123.456_789 and .5_0 forms.
• test/unit/scoring.test.ts — regression test for fractional underscore literals (#992).

Number(raw.replace(/_/g, "")) already strips underscores before parsing; no other logic changes.

Reachability

refreshScoringModelSnapshot fetches upstream gittensor/constants.py and runs it through parsePythonNumberConstants, merging results into scoring constants used by buildScorePreview, score breakdown, and unmodeled-constant detection. Any upstream maintainer using fractional digit-grouping for readability hits this path.

API / OpenAPI / MCP contract

No schema changes. Parsed upstream constant values may change when fractional underscores are present in constants.py.

Migration / deploy / secrets

None.

Security / privacy

None.

Validation

Run from repo root (Node >= 22):

npm run test:ci

All gates passed locally on branch fix/fractional-underscore-upstream-constants.

Distinct from #994

Unrelated to label multiplier selection (#994). This completes the fractional-underscore gap left by #969.

[ghost]

🟨🟨🟨🟨🟨🟨🟨🟨🟨🟨🟨🟨🟨🟨🟨🟨🟨🟨

🛠️ Maintainer review — Held for maintainer review

Reviewed 2 changed file(s) — two independent AI reviewers, synthesized. ✅ Safe to merge — both reviewers found no blocking issues.

ℹ️ Correct and all CI green — held for maintainer review.

Merge readiness: ✅ Mergeable · all CI checks green/neutral/skipped.

Summary

The change updates the regex in parsePythonNumberConstants to correctly handle underscore separators in both integer and fractional parts of Python numeric literals, fixing a parsing bug. Added unit test verifies the new behavior. No security concerns, build passes, and existing tests remain green. The modification is limited to parsing logic and does not introduce regressions.

💡 Nits — non-blocking, optional

• Regex could be refined to more strictly follow Python numeric literal rules.
• Consider tightening the regex to reject trailing underscores (e.g., 123_) which are not valid Python literals.
• Add a test case for an invalid constant (e.g., BAD = 1_) to ensure it is ignored.
• Consider adding a comment explaining why the regex change is necessary and what specific issues it addresses.

[ghost]

Note

🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦

💡 Gittensory review — advisory only

0 files · no blockers · readiness 0/100

💡 Advisory only — no action taken

Signal	Result	Evidence
Code review	✅ No blockers	synthesized
Gate result	⚠️ Skipped	#1039 is no longer open.

---

💰 Earn for open-source contributions like this. Gittensor lets GitHub contributors earn for the work they already do — register to start earning →.

Checked by Gittensory, a quiet PR intelligence layer for OSS maintainers.

[superagent-security]

Superagent didn't find any vulnerabilities or security issues in this PR.

[ghost]

Held for maintainer (guardrail_area) — correct + CI green

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 96.85%. Comparing base (3425026) to head (f9074aa).

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #1039   +/-   ##
=======================================
  Coverage   96.85%   96.85%           
=======================================
  Files         141      141           
  Lines       17501    17501           
  Branches     6334     6334           
=======================================
  Hits        16951    16951           
  Misses        104      104           
  Partials      446      446           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.