feat(scoring): model upstream review-collateral and non-code caps (Fixes #809)

apps/gittensory-ui/public/openapi.json

@@ -4296,16 +4296,20 @@
                       },
                       "credibilityFloor": {
                         "type": "number"
                       },
                       "credibilityObserved": {
                         "type": "number"
+                      },
+                      "reviewCollateralMultiplier": {
+                        "type": "number"
                       }
                     },
                     "required": [
                       "baseTokenGatePassed",
                       "openPrThreshold",
                       "openPrCount",
                       "collateralFraction",
+                      "reviewCollateralMultiplier",
                       "credibilityFloor",
                       "credibilityObserved"
                     ]
@@ -4551,13 +4555,17 @@
                       },
                       "credibilityObserved": {
                         "type": "number"
+                      },
+                      "reviewCollateralMultiplier": {
+                        "type": "number"
                       }
                     },
                     "required": [
                       "baseTokenGatePassed",
                       "openPrThreshold",
                       "openPrCount",
                       "collateralFraction",
+                      "reviewCollateralMultiplier",
                       "credibilityFloor",
                       "credibilityObserved"
                     ]
@@ -4803,13 +4811,17 @@
                       },
                       "credibilityObserved": {
                         "type": "number"
+                      },
+                      "reviewCollateralMultiplier": {
+                        "type": "number"
                       }
                     },
                     "required": [
                       "baseTokenGatePassed",
                       "openPrThreshold",
                       "openPrCount",
                       "collateralFraction",
+                      "reviewCollateralMultiplier",
                       "credibilityFloor",
                       "credibilityObserved"
                     ]
@@ -5055,13 +5067,17 @@
                       },
                       "credibilityObserved": {
                         "type": "number"
+                      },
+                      "reviewCollateralMultiplier": {
+                        "type": "number"
                       }
                     },
                     "required": [
                       "baseTokenGatePassed",
                       "openPrThreshold",
                       "openPrCount",
                       "collateralFraction",
+                      "reviewCollateralMultiplier",
                       "credibilityFloor",
                       "credibilityObserved"
                     ]
@@ -5307,13 +5323,17 @@
                       },
                       "credibilityObserved": {
                         "type": "number"
+                      },
+                      "reviewCollateralMultiplier": {
+                        "type": "number"
                       }
                     },
                     "required": [
                       "baseTokenGatePassed",
                       "openPrThreshold",
                       "openPrCount",
                       "collateralFraction",
+                      "reviewCollateralMultiplier",
                       "credibilityFloor",
                       "credibilityObserved"
                     ]
@@ -6207,13 +6227,17 @@
               },
               "credibilityObserved": {
                 "type": "number"
+              },
+              "reviewCollateralMultiplier": {
+                "type": "number"
               }
             },
             "required": [
               "baseTokenGatePassed",
               "openPrThreshold",
               "openPrCount",
               "collateralFraction",
+              "reviewCollateralMultiplier",
               "credibilityFloor",
               "credibilityObserved"
             ]
@@ -6455,13 +6479,17 @@
                     },
                     "credibilityObserved": {
                       "type": "number"
+                    },
+                    "reviewCollateralMultiplier": {
+                      "type": "number"
                     }
                   },
                   "required": [
                     "baseTokenGatePassed",
                     "openPrThreshold",
                     "openPrCount",
                     "collateralFraction",
+                    "reviewCollateralMultiplier",
                     "credibilityFloor",
                     "credibilityObserved"
                   ]

src/api/routes.ts

@@ -443,7 +443,7 @@
     status: z.enum(["passed", "failed", "not_run", "skipped", "focused", "unknown"]),
     summary: z.string().max(MAX_LOCAL_BRANCH_TEXT_CHARS).optional(),
     durationMs: z.number().int().min(0).optional(),
     exitCode: z.number().int().min(0).optional(),
   })
   .strict();
 
@@ -456,6 +456,7 @@
     sourceLines: z.number().min(0).optional(),
     testTokenScore: z.number().min(0).optional(),
     nonCodeTokenScore: z.number().min(0).optional(),
+  nonCodeLines: z.number().min(0).optional(),
     warnings: z.array(z.string().max(MAX_LOCAL_SCORER_WARNING_CHARS)).max(MAX_LOCAL_SCORER_WARNING_COUNT).optional(),
   })
   .strict();
@@ -533,6 +534,7 @@
   sourceLines: z.number().min(0).optional(),
   testTokenScore: z.number().min(0).optional(),
   nonCodeTokenScore: z.number().min(0).optional(),
+  nonCodeLines: z.number().min(0).optional(),
   existingContributorTokenScore: z.number().min(0).optional(),
   prAgeHours: z.number().min(0).optional(),
   openPrCount: z.number().int().min(0).optional(),

src/openapi/schemas.ts

@@ -1228,9 +1228,10 @@
 
 const ScoreGatesSchema = z.object({
   baseTokenGatePassed: z.boolean(),
   openPrThreshold: z.number(),
   openPrCount: z.number(),
   collateralFraction: z.number(),
+  reviewCollateralMultiplier: z.number(),
   credibilityFloor: z.number(),
   credibilityObserved: z.number(),
 });

src/registry/normalize.ts

@@ -1,9 +1,8 @@
+import { DEFAULT_ISSUE_DISCOVERY_SHARE } from "../scoring/model";
 import type { JsonValue, RegistryRepoConfig, RegistrySnapshot, RepoTimeDecayOverrides } from "../types";
 
 type RawRepoConfig = Record<string, JsonValue>;
 
-const DEFAULT_ISSUE_DISCOVERY_SHARE = 0.5;
-
 export function normalizeRegistryPayload(payload: unknown, source: RegistrySnapshot["source"], fetchedAt: string): RegistrySnapshot {
   const repos = extractRepoEntries(payload).map(([repo, config]) => normalizeRepo(repo, config));
   const totalEmissionShare = repos.reduce((sum, repo) => sum + repo.emissionShare, 0);

src/scoring/model.ts

@@ -4,9 +4,11 @@
 } from "../db/repositories";
 import { getLatestRegistrySnapshot } from "../registry/sync";
 import { syncUnmodeledScoringConstantDrift } from "../upstream/unmodeled-scoring-drift";
 import type { JsonValue, ScoringModelSnapshotRecord } from "../types";
 import { errorMessage, nowIso } from "../utils/json";
 
+export const DEFAULT_ISSUE_DISCOVERY_SHARE = 0.5;
+
 export const DEFAULT_SCORING_CONSTANTS: Record<string, number> = {
   OSS_EMISSION_SHARE: 0.9,
   // Upstream name is ISSUES_TREASURY_EMISSION_SHARE (plural). The prior singular spelling never matched
@@ -28,6 +30,9 @@
   OPEN_ISSUE_SPAM_TOKEN_SCORE_PER_SLOT: 300,
   MAX_OPEN_ISSUE_THRESHOLD: 30,
   OPEN_PR_COLLATERAL_PERCENT: 0.2,
+  MAX_OPEN_PR_REVIEW_COLLATERAL_MULTIPLIER: 2.0,
+  MAX_LINES_SCORED_FOR_NON_CODE_EXT: 300,
+  DEFAULT_ISSUE_DISCOVERY_SHARE,
   REVIEW_PENALTY_RATE: 0.15,
   STANDARD_ISSUE_MULTIPLIER: 1.33,
   MAINTAINER_ISSUE_MULTIPLIER: 1.66,

src/scoring/preview.ts

@@ -11,9 +11,11 @@
   linkedIssueContext?: LinkedIssueMultiplierContext | undefined;
   sourceTokenScore?: number | undefined;
   totalTokenScore?: number | undefined;
   sourceLines?: number | undefined;
   testTokenScore?: number | undefined;
   nonCodeTokenScore?: number | undefined;
+  /** Raw non-code line count before upstream's MAX_LINES_SCORED_FOR_NON_CODE_EXT cap. */
+  nonCodeLines?: number | undefined;
   existingContributorTokenScore?: number | undefined;
   openPrCount?: number | undefined;
   credibility?: number | undefined;
@@ -167,7 +169,10 @@
     baseTokenGatePassed: boolean;
     openPrThreshold: number;
     openPrCount: number;
+    /** Effective open-PR collateral fraction (OPEN_PR_COLLATERAL_PERCENT × reviewCollateralMultiplier). */
     collateralFraction: number;
+    /** Upstream open-PR review-collateral multiplier from CHANGES_REQUESTED reviews (≥ 1, capped). */
+    reviewCollateralMultiplier: number;
     credibilityFloor: number;
     credibilityObserved: number;
   };
@@ -288,7 +293,10 @@
   const directPrSlice = repoSlice * (1 - issueDiscoveryShare);
   const issueDiscoverySlice = repoSlice * issueDiscoveryShare;
   const sourceTokenScore = nonNegative(input.sourceTokenScore);
-  const totalTokenScore = nonNegative(input.totalTokenScore ?? sourceTokenScore + nonNegative(input.testTokenScore) + nonNegative(input.nonCodeTokenScore));
+  const cappedNonCodeTokenScore = applyNonCodeLineCap(input, constants);
+  const totalTokenScore = nonNegative(
+    input.totalTokenScore ?? sourceTokenScore + nonNegative(input.testTokenScore) + cappedNonCodeTokenScore,
+  );
   const sourceLines = Math.max(1, nonNegative(input.sourceLines ?? sourceTokenScore));
   const fixedBaseScore = input.fixedBaseScore ?? config?.fixedBaseScore ?? undefined;
   const rawDensity = sourceTokenScore / sourceLines;
@@ -315,7 +323,13 @@
   const credibilityFloor = constant(constants, "MIN_CREDIBILITY", 0.8);
   const credibilityMultiplier = credibilityObserved >= credibilityFloor ? 1 : credibilityObserved / credibilityFloor;
   const changesRequestedCount = nonNegative(input.changesRequestedCount);
-  const reviewPenaltyMultiplier = clamp(1 - changesRequestedCount * constant(constants, "REVIEW_PENALTY_RATE", 0.15), 0, 1);
+  const reviewPenaltyRate = constant(constants, "REVIEW_PENALTY_RATE", 0.15);
+  const reviewPenaltyMultiplier = clamp(1 - changesRequestedCount * reviewPenaltyRate, 0, 1);
+  const reviewCollateralMultiplier = Math.min(
+    constant(constants, "MAX_OPEN_PR_REVIEW_COLLATERAL_MULTIPLIER", 2.0),
+    1 + changesRequestedCount * reviewPenaltyRate,
+  );
+  const openPrCollateralPercent = constant(constants, "OPEN_PR_COLLATERAL_PERCENT", 0.2);
   const openPrCount = nonNegative(input.openPrCount);
   // The concurrency allowance is earned from the contributor's established merged-history token
   // score; the planned PR's own tokens (totalTokenScore) must not inflate its own open-PR threshold.
@@ -361,7 +375,8 @@
       baseTokenGatePassed,
       openPrThreshold,
       openPrCount,
-      collateralFraction: constant(constants, "OPEN_PR_COLLATERAL_PERCENT", 0.2),
+      reviewCollateralMultiplier: roundScore(reviewCollateralMultiplier),
+      collateralFraction: roundScore(openPrCollateralPercent * reviewCollateralMultiplier),
       credibilityFloor,
       credibilityObserved,
     },
@@ -911,6 +926,14 @@
   return clamp(0.75 + merged * 0.04 - stale * 0.03 - unlinked * 0.02, 0.25, 1);
 }
 
+function applyNonCodeLineCap(input: Pick<ScorePreviewInput, "nonCodeTokenScore" | "nonCodeLines">, constants: Record<string, number>): number {
+  const score = nonNegative(input.nonCodeTokenScore);
+  const lines = nonNegative(input.nonCodeLines);
+  if (score <= 0 || lines <= 0) return score;
+  const maxLines = constant(constants, "MAX_LINES_SCORED_FOR_NON_CODE_EXT", 300);
+  return lines <= maxLines ? score : score * (maxLines / lines);
+}
+
 function constant(constants: Record<string, number>, key: string, fallback: number): number {
   const value = constants[key];
   return typeof value === "number" && Number.isFinite(value) ? value : fallback;

src/signals/local-branch.ts

@@ -452,7 +452,7 @@
   observedPullRequestScenarios: ObservedPullRequestScenarios;
   duplicateRiskCount: number;
 }): ScorePreviewInput {
   const scorer = args.input.localScorer;
   const testLineCount = args.changedFiles.filter((file) => isTestFile(file.path)).reduce((sum, file) => sum + nonNegative(file.additions) + nonNegative(file.deletions), 0);
   const sourceLineCount = args.changedFiles
     .filter((file) => isCodeFile(file.path))
@@ -471,6 +471,7 @@
     sourceLines: scorer?.sourceLines ?? Math.max(1, sourceLineCount || args.changedLineCount || 1),
     testTokenScore: scorer?.testTokenScore ?? testLineCount,
     nonCodeTokenScore: scorer?.nonCodeTokenScore ?? nonCodeLineCount,
+    nonCodeLines: nonCodeLineCount,
     openPrCount: args.outcomeHistory.totals.openPullRequests,
     credibility: args.repoOutcome?.credibility ?? args.outcomeHistory.totals.credibility,
     metadataOnly: scorer?.mode !== "gittensor_root" && scorer?.mode !== "external_command",

test/unit/registry.test.ts

@@ -1,6 +1,7 @@
 import { afterEach, describe, expect, it, vi } from "vitest";
 import { getRepository, upsertRepositoryFromGitHub } from "../../src/db/repositories";
 import { normalizeRegistryPayload } from "../../src/registry/normalize";
+import { DEFAULT_ISSUE_DISCOVERY_SHARE } from "../../src/scoring/model";
 import { getLatestRegistrySnapshot, persistRegistrySnapshot, refreshRegistry } from "../../src/registry/sync";
 import { createTestEnv } from "../helpers/d1";
 
@@ -122,8 +123,15 @@
       trustedLabelPipeline: true,
     });
     expect(fromObjectMap.repositories.map((repo) => repo.repo)).toEqual(["JSONbored/gittensory"]);
+    expect(fromObjectMap.repositories[0]).toMatchObject({
+      repo: "JSONbored/gittensory",
+      issueDiscoveryShare: DEFAULT_ISSUE_DISCOVERY_SHARE,
+    });
     expect(fromArray.repositories.map((repo) => repo.repo)).toEqual(["JSONbored/gittensory", "bad/numbers"]);
-    expect(fromArray.repositories.find((repo) => repo.repo === "bad/numbers")).toMatchObject({ emissionShare: 0, issueDiscoveryShare: 0.5 });
+    expect(fromArray.repositories.find((repo) => repo.repo === "bad/numbers")).toMatchObject({
+      emissionShare: 0,
+      issueDiscoveryShare: DEFAULT_ISSUE_DISCOVERY_SHARE,
+    });
     expect(empty.repoCount).toBe(0);
   });