Update raw_score output for v2 scoring

docs/library.md

@@ -148,7 +148,7 @@ the following fields:
 |-------|------|-------------|
 | `user_login` | `str` | GitHub username that was scored |
 | `context_repo` | `str` | Repository used as scoring context |
-| `raw_score` | `float` | Raw graph score before normalization |
+| `raw_score` | `float` | Pre-normalization score: graph score (v1) or logit (v2) |
 | `normalized_score` | `float` | Normalized score (0.0 - 1.0) |
 | `trust_level` | `TrustLevel` | HIGH, MEDIUM, LOW, UNKNOWN, or BOT |
 | `percentile` | `float` | Percentile rank (0.0 - 1.0) |

docs/mcp-server.md

@@ -180,7 +180,7 @@ Returns an expanded breakdown with contributions, flags, and metadata.
   "context_repo": "octocat/Hello-World",
   "trust_level": "HIGH",
   "normalized_score": 0.82,
-  "raw_score": 0.0045,
+  "raw_score": 0.2871,
   "account_age_days": 3650,
   "total_merged_prs": 47,
   "unique_repos_contributed": 12,

docs/methodology.md

@@ -101,7 +101,7 @@ These raw weights are normalized to sum to 1.0, so actual values in the random w
 
 The directed graph is scored using personalized graph-based ranking with a damping factor (alpha) of 0.85. This produces a raw score for the user node.
 
-Normalization converts the raw score to a 0-1 range:
+**v1:** Normalization converts the raw graph score to a 0-1 range:
 
 ```
 baseline = 1 / num_nodes
@@ -111,6 +111,18 @@ normalized = ratio / (ratio + 1)
 
 This sigmoid-like mapping means a score equal to the uniform baseline maps to 0.5, with diminishing returns above.
 
+**v2:** The normalized graph score is combined with merge rate and account age into a logit, then passed through a sigmoid:
+
+```
+logit = intercept + graph_score_weight * graph_score
+      + merge_rate_weight * merge_rate
+      + account_age_weight * log(account_age_days + 1)
+
+normalized = 1 / (1 + e^(-logit))
+```
+
+`raw_score` in the v2 output contains the pre-sigmoid logit value.
+
 ### Classification
 
 | Level | Threshold | Meaning |

src/good_egg/scorer.py

@@ -206,7 +206,7 @@ def _score_v2(
         return TrustScore(
             user_login=login,
             context_repo=context_repo,
-            raw_score=raw_score,
+            raw_score=logit,
             normalized_score=normalized,
             trust_level=trust_level,
             account_age_days=user_data.profile.account_age_days,

tests/test_scorer.py

@@ -2,6 +2,7 @@
 
 from __future__ import annotations
 
+import math
 from datetime import UTC, datetime, timedelta
 
 from good_egg.config import GoodEggConfig
@@ -426,8 +427,8 @@ def test_v2_scoring_produces_nonzero_score(self) -> None:
         data = _make_contribution_data(merged_prs=prs, repos=repos, closed_pr_count=5)
         result = scorer.score(data, "my-org/my-elixir-app")
 
-        assert result.raw_score > 0.0
-        assert 0.0 <= result.normalized_score <= 1.0
+        expected = 1.0 / (1.0 + math.exp(-result.raw_score))
+        assert abs(result.normalized_score - expected) < 1e-9
         assert result.scoring_model == "v2"
 
     def test_v2_component_scores_populated(self) -> None: