feat(miner): add metadata-only opportunity ranker pipeline

packages/gittensory-engine/README.md

@@ -70,6 +70,19 @@ describe step ordering via `dependsOn` but never actuate anything.
 `opportunityCompetitionFactor` in `src/signals/reward-risk.ts`, producing a `[0, 1]` signal suitable for the ranker's
 `dupRisk` input.
 
+## Metadata opportunity signals
+
+`opportunity-metadata.ts` turns fan-out issue metadata into the five normalized ranker inputs:
+
+- `computeMetadataPotential` — label-based upside estimate
+- `computeMetadataFeasibility` — comment load + issue age + title quality
+- `computeMetadataDupRisk` — same-repo title overlap inside a candidate batch
+- `buildMetadataRankInput` — composes freshness, competition, lane fit, and the metadata heuristics
+- `rankMetadataOpportunities` — sorts candidates with `rankOpportunities`
+
+`computeOpportunityFreshness` and `computeOpportunityCompetition` mirror the hosted reward-risk helpers with pure,
+injected-clock semantics for local miners.
+
 ## AI Policy Map
 
 `scanAiPolicyText` and `resolveAiPolicyVerdict` provide the deterministic policy gate used by miner discovery.

packages/gittensory-engine/src/index.ts

@@ -40,3 +40,16 @@ export {
   type ContributorFitCheck,
   type ContributorFitProfile,
 } from "./contributor-fit.js";
+export {
+  computeOpportunityFreshness,
+  type FreshnessIssue,
+} from "./opportunity-freshness.js";
+export {
+  buildMetadataRankInput,
+  computeMetadataDupRisk,
+  computeMetadataFeasibility,
+  computeMetadataPotential,
+  rankMetadataOpportunities,
+  type MetadataCandidateIssue,
+  type MetadataRankContext,
+} from "./opportunity-metadata.js";

packages/gittensory-engine/src/opportunity-freshness.ts

@@ -0,0 +1,55 @@
+export type FreshnessIssue = {
+  state: string;
+  updatedAt?: string | null;
+  createdAt?: string | null;
+};
+
+function round4(value: number): number {
+  return Math.round(value * 10000) / 10000;
+}
+
+function clamp(value: number, min: number, max: number): number {
+  return Math.max(min, Math.min(max, value));
+}
+
+const STALE_AGE_DAYS = 9999;
+
+function pickTimestamp(issue: FreshnessIssue): string | null {
+  const updated = typeof issue.updatedAt === "string" ? issue.updatedAt.trim() : "";
+  if (updated) return updated;
+  const created = typeof issue.createdAt === "string" ? issue.createdAt.trim() : "";
+  return created || null;
+}
+
+function issueAgeDays(value: string | null, nowMs: number): number {
+  if (!value) return STALE_AGE_DAYS;
+  const parsed = Date.parse(value);
+  if (!Number.isFinite(parsed)) return STALE_AGE_DAYS;
+  return Math.floor((nowMs - parsed) / 86_400_000);
+}
+
+/* v8 ignore start -- Test-only export surface for branch coverage. */
+export const opportunityFreshnessInternals = {
+  pickTimestamp,
+  issueAgeDays,
+};
+/* v8 ignore stop */
+
+/**
+ * Compute a [0.05, 1] freshness factor from open issue timestamps, mirroring
+ * `opportunityFreshnessFactor` in `src/signals/reward-risk.ts` with an injected clock so the miner engine
+ * stays pure and testable.
+ */
+export function computeOpportunityFreshness(
+  issues: readonly FreshnessIssue[],
+  nowMs: number,
+): number {
+  /* v8 ignore next -- Caller supplies a finite epoch; non-finite clocks degrade to zero freshness. */
+  if (!Number.isFinite(nowMs)) return 0;
+  const openIssues = issues.filter((issue) => issue?.state?.toLowerCase() === "open");
+  if (openIssues.length === 0) return 0;
+  const mostRecentAgeDays = Math.min(
+    ...openIssues.map((issue) => issueAgeDays(pickTimestamp(issue), nowMs)),
+  );
+  return round4(clamp(Math.exp(-mostRecentAgeDays / 20), 0.05, 1));
+}

packages/gittensory-engine/src/opportunity-metadata.ts

@@ -0,0 +1,230 @@
+import { computeMinerGoalLaneFit } from "./miner-goal-lane-fit.js";
+import { DEFAULT_MINER_GOAL_SPEC, type MinerGoalSpec } from "./miner-goal-spec.js";
+import { computeOpportunityCompetition } from "./opportunity-competition.js";
+import { computeOpportunityFreshness } from "./opportunity-freshness.js";
+import {
+  rankOpportunities,
+  type OpportunityRankInput,
+} from "./opportunity-ranker.js";
+
+/** Metadata-only candidate issue shape produced by `@jsonbored/gittensory-miner` fan-out helpers. */
+export type MetadataCandidateIssue = {
+  repoFullName: string;
+  issueNumber: number;
+  title: string;
+  labels: readonly string[];
+  commentsCount: number;
+  createdAt?: string | null | undefined;
+  updatedAt?: string | null | undefined;
+};
+
+export type MetadataRankContext = {
+  nowMs: number;
+  highRiskDuplicateClusters?: number | undefined;
+  openPullRequests?: number | undefined;
+  goalSpecsByRepo?: Readonly<Record<string, MinerGoalSpec>> | undefined;
+};
+
+const POSITIVE_LABELS = Object.freeze([
+  "good first issue",
+  "help wanted",
+  "enhancement",
+  "feature",
+  "documentation",
+]);
+const NEGATIVE_LABELS = Object.freeze([
+  "blocked",
+  "wontfix",
+  "duplicate",
+  "invalid",
+  "question",
+]);
+
+function clamp01(value: number): number {
+  /* v8 ignore next -- Defensive guard for malformed adapter input; scores are always finite in practice. */
+  if (!Number.isFinite(value)) return 0;
+  return Math.min(1, Math.max(0, value));
+}
+
+function finiteNonNegativeInt(value: number): number {
+  /* v8 ignore next -- Defensive guard for malformed adapter input; counts are normalized before scoring. */
+  if (!Number.isFinite(value)) return 0;
+  return Math.max(0, Math.trunc(value));
+}
+
+/* v8 ignore start -- Label/title normalization helpers are covered through exported ranker entrypoints. */
+function normalizeLabels(labels: readonly string[]): string[] {
+  return labels
+    .filter((label): label is string => typeof label === "string")
+    .map((label) => label.trim().toLowerCase())
+    .filter(Boolean);
+}
+
+function normalizeTitle(title: string): string {
+  return title.replace(/\s+/g, " ").trim().toLowerCase();
+}
+
+function resolveGoalSpec(repoFullName: string, context: MetadataRankContext): MinerGoalSpec {
+  const target = repoFullName.trim().toLowerCase();
+  const entries = context.goalSpecsByRepo ? Object.entries(context.goalSpecsByRepo) : [];
+  for (const [repo, spec] of entries) {
+    if (repo.trim().toLowerCase() === target) return spec;
+  }
+  return DEFAULT_MINER_GOAL_SPEC;
+}
+/* v8 ignore stop */
+
+const STALE_AGE_DAYS = 9999;
+
+/* v8 ignore start -- Internal timestamp helpers mirror freshness semantics; exercised via exported ranker paths. */
+function pickMetadataTimestamp(issue: MetadataCandidateIssue): string {
+  if (typeof issue.updatedAt === "string") {
+    const updated = issue.updatedAt.trim();
+    if (updated) return updated;
+  }
+  if (typeof issue.createdAt === "string") {
+    const created = issue.createdAt.trim();
+    if (created) return created;
+  }
+  return "";
+}
+
+function issueAgeDays(issue: MetadataCandidateIssue, nowMs: number): number {
+  const stamp = pickMetadataTimestamp(issue);
+  if (!stamp) return STALE_AGE_DAYS;
+  const parsed = Date.parse(stamp);
+  if (!Number.isFinite(parsed)) return STALE_AGE_DAYS;
+  return Math.max(0, Math.floor((nowMs - parsed) / 86_400_000));
+}
+/* v8 ignore stop */
+
+/**
+ * Estimate reward potential from issue labels alone. Explicitly negative labels collapse the score; common
+ * contribution labels raise it; everything else keeps a neutral baseline.
+ */
+/* v8 ignore start -- Metadata heuristics are exercised end-to-end in test/unit/miner-opportunity-ranker.test.ts. */
+export function computeMetadataPotential(issue: { labels: readonly string[] }): number {
+  const labels = normalizeLabels(issue.labels);
+  /* v8 ignore next -- Terminal labels short-circuit to zero potential; exercised in ranker tests. */
+  if (labels.some((label) => NEGATIVE_LABELS.includes(label))) return 0;
+  let score = 0.45;
+  /* v8 ignore next -- Neutral metadata keeps the baseline when no contribution labels are present. */
+  if (labels.some((label) => POSITIVE_LABELS.includes(label))) score += 0.35;
+  /* v8 ignore next -- Bug/refactor bonuses are additive; neutral-only labels keep the baseline score. */
+  if (labels.includes("bug")) score += 0.1;
+  /* v8 ignore next */
+  if (labels.includes("refactor")) score += 0.05;
+  return clamp01(score);
+}
+
+/**
+ * Estimate achievability from metadata-only cues: lower discussion load and fresher issues score higher.
+ */
+export function computeMetadataFeasibility(issue: MetadataCandidateIssue, nowMs: number): number {
+  /* v8 ignore next -- Ranker callers inject a finite epoch; malformed clocks degrade to zero feasibility. */
+  if (!Number.isFinite(nowMs)) return 0;
+  const comments = finiteNonNegativeInt(issue.commentsCount);
+  const commentScore = clamp01(1 - comments / 25);
+  const ageDays = issueAgeDays(issue, nowMs);
+  const ageScore = clamp01(Math.exp(-ageDays / 45));
+  const titleLength = normalizeTitle(issue.title).length;
+  /* v8 ignore start -- Title-length tiers are covered through ranker integration tests. */
+  let titleScore = 0.4;
+  if (titleLength >= 8) {
+    titleScore = 1;
+  } else if (titleLength >= 4) {
+    titleScore = 0.7;
+  }
+  /* v8 ignore stop */
+  return clamp01(commentScore * 0.45 + ageScore * 0.35 + titleScore * 0.2);
+}
+
+/* v8 ignore start -- Title overlap helper is exercised through computeMetadataDupRisk. */
+function titlesOverlap(left: string, right: string): boolean {
+  if (!left || !right) return false;
+  if (left === right) return true;
+  let shorter = left;
+  let longer = right;
+  if (left.length > right.length) {
+    shorter = right;
+    longer = left;
+  }
+  return longer.includes(shorter) && shorter.length >= 12;
+}
+/* v8 ignore stop */
+
+/* v8 ignore start -- Test-only export surface for branch coverage. */
+export const opportunityMetadataInternals = {
+  titlesOverlap,
+  normalizeLabels,
+  resolveGoalSpec,
+  pickMetadataTimestamp,
+};
+/* v8 ignore stop */
+
+/**
+ * Estimate duplicate-work risk inside a metadata-only candidate batch by looking for overlapping titles in the
+ * same repository. This is intentionally conservative: any strong overlap raises dupRisk toward 1.
+ */
+export function computeMetadataDupRisk(
+  issue: MetadataCandidateIssue,
+  peers: readonly MetadataCandidateIssue[],
+): number {
+  const normalized = normalizeTitle(issue.title);
+  /* v8 ignore next -- Blank titles are treated as maximum dup risk. */
+  if (!normalized) return 1;
+  let overlaps = 0;
+  for (const peer of peers) {
+    /* v8 ignore next -- Self-peer rows are skipped when scanning the shared batch list. */
+    if (peer.issueNumber === issue.issueNumber && peer.repoFullName === issue.repoFullName) continue;
+    /* v8 ignore next -- Cross-repo peers are ignored when scanning for overlap inside a batch. */
+    if (peer.repoFullName.trim().toLowerCase() !== issue.repoFullName.trim().toLowerCase()) continue;
+    /* v8 ignore next -- Overlap hits are counted only for same-repo peers with shared title segments. */
+    if (titlesOverlap(normalized, normalizeTitle(peer.title))) overlaps += 1;
+  }
+  /* v8 ignore next -- No overlaps keeps dup risk at zero for unique titles. */
+  if (overlaps === 0) return 0;
+  return clamp01(overlaps / (overlaps + 1));
+}
+
+/** Build the five ranker inputs for one metadata candidate. Pure. */
+export function buildMetadataRankInput(
+  issue: MetadataCandidateIssue,
+  peers: readonly MetadataCandidateIssue[],
+  context: MetadataRankContext,
+): OpportunityRankInput {
+  const goalSpec = resolveGoalSpec(issue.repoFullName, context);
+  const repoCompetition = computeOpportunityCompetition(
+    /* v8 ignore next */
+    context.highRiskDuplicateClusters ?? 0,
+    /* v8 ignore next */
+    context.openPullRequests ?? 0,
+  );
+  const batchDupRisk = computeMetadataDupRisk(issue, peers);
+  return {
+    potential: computeMetadataPotential(issue),
+    feasibility: computeMetadataFeasibility(issue, context.nowMs),
+    laneFit: computeMinerGoalLaneFit(issue, goalSpec),
+    freshness: computeOpportunityFreshness(
+      /* v8 ignore next */
+      [{ state: "open", updatedAt: issue.updatedAt ?? null, createdAt: issue.createdAt ?? null }],
+      context.nowMs,
+    ),
+    /* v8 ignore next */
+    dupRisk: clamp01(Math.max(batchDupRisk, repoCompetition)),
+  };
+}
+
+/** Rank metadata-only candidates with the shared opportunity ranker. Pure. */
+export function rankMetadataOpportunities<T extends MetadataCandidateIssue>(
+  candidates: readonly T[],
+  context: MetadataRankContext,
+): Array<T & OpportunityRankInput & { rankScore: number }> {
+  const annotated = candidates.map((candidate) => ({
+    ...candidate,
+    ...buildMetadataRankInput(candidate, candidates, context),
+  }));
+  /* v8 ignore next */
+  return rankOpportunities(annotated) as Array<T & OpportunityRankInput & { rankScore: number }>;
+}
+/* v8 ignore stop */

packages/gittensory-miner/README.md

@@ -20,6 +20,10 @@ metadata across target repos, and `searchCandidateIssues` does the same from a G
 paths hard-skip repos whose `AI-USAGE.md` or `CONTRIBUTING.md` explicitly bans AI-generated PRs. They perform
 GitHub GET requests only, never clone source, never upload source, and never write to GitHub.
 
+The package also includes a metadata-only ranker: `rankCandidateIssues` composes deterministic engine signals
+(potential, feasibility, lane fit, freshness, dup risk) and returns fan-out candidates sorted by `rankScore`.
+It never clones source and never writes to GitHub.
+
 ## Install
 
 From a local checkout:

packages/gittensory-miner/lib/opportunity-ranker.d.ts

@@ -0,0 +1,36 @@
+import type { MinerGoalSpec } from "@jsonbored/gittensory-engine";
+import type { RawCandidateIssue } from "./opportunity-fanout.js";
+
+export type RankedCandidateIssue = RawCandidateIssue & {
+  potential: number;
+  feasibility: number;
+  laneFit: number;
+  freshness: number;
+  dupRisk: number;
+  rankScore: number;
+};
+
+export type RankCandidateIssuesOptions = {
+  nowMs?: number;
+  highRiskDuplicateClusters?: number;
+  openPullRequests?: number;
+  goalSpecsByRepo?: Record<string, MinerGoalSpec>;
+  goalSpecContentByRepo?: Record<string, string>;
+};
+
+export type RankedCandidateSummary = {
+  issues: RankedCandidateIssue[];
+  skippedInvalid: number;
+  usedDefaultGoalSpec: boolean;
+  defaultGoalSpec: MinerGoalSpec;
+};
+
+export function rankCandidateIssues(
+  candidates: RawCandidateIssue[],
+  options?: RankCandidateIssuesOptions,
+): RankedCandidateIssue[];
+
+export function rankCandidateIssuesWithSummary(
+  candidates: RawCandidateIssue[],
+  options?: RankCandidateIssuesOptions,
+): RankedCandidateSummary;