@wanshi-kg/wanshi 0.1.0

package/dist/evaluation/classifier/labeledSamples.js

@@ -0,0 +1,379 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.LABELED_SAMPLES = void 0;
+exports.LABELED_SAMPLES = [
+    // ── code ───────────────────────────────────────────────────────────────
+    {
+        id: "code-1",
+        path: "src/services/AuthService.ts",
+        expected: "code",
+        content: `import { TokenStore } from "./TokenStore";
+
+export class AuthService extends BaseService {
+  async login(user: string, password: string): Promise<Session> {
+    const hash = await this.hasher.hash(password);
+    if (!this.store.verify(user, hash)) {
+      throw new Error("invalid credentials");
+    }
+    return this.store.createSession(user);
+  }
+}`,
+    },
+    {
+        id: "code-2",
+        path: "scripts/clean_data.py",
+        expected: "code",
+        content: `import os
+import json
+
+def clean(records):
+    for r in records:
+        if r.get("value") is None:
+            continue
+        yield {"id": r["id"], "value": float(r["value"])}
+
+def main():
+    with open("data.json") as f:
+        records = json.load(f)
+    print(list(clean(records)))`,
+    },
+    // ── financial ──────────────────────────────────────────────────────────
+    {
+        id: "financial-1",
+        path: "reports/q3-earnings.md",
+        expected: "financial",
+        content: `Q3 revenue rose to $4.2 billion, beating analyst estimates of $3.9 billion.
+EBITDA margin improved to 28%. The board declared a dividend of $0.45 per share.
+NASDAQ: ACME closed up 6% after the earnings call. Full-year guidance was raised.`,
+    },
+    {
+        id: "financial-2",
+        path: "finance/portfolio-review.txt",
+        expected: "financial",
+        content: `Portfolio allocation: 60% equity, 30% bonds, 10% cash. The S&P 500 holdings
+returned $1,250,000 in unrealized gains. P/E ratio of the tech sleeve sits at 24.
+Dividend yield across the portfolio is 2.1%. SEC 10-K filings reviewed for each holding.`,
+    },
+    // ── medical ────────────────────────────────────────────────────────────
+    {
+        id: "medical-1",
+        path: "records/patient-note.txt",
+        expected: "medical",
+        content: `Patient presents with hypertension. Prescribed lisinopril 10 mg PO daily and
+metformin 500 mg BID for type 2 diabetes. Blood pressure 148/92 mmHg, heart rate 84 bpm.
+Diagnosis confirmed; follow-up in 4 weeks to reassess medication and symptoms.`,
+    },
+    {
+        id: "medical-2",
+        path: "trials/study-protocol.txt",
+        expected: "medical",
+        content: `This randomized controlled, double-blind, placebo-controlled trial evaluates
+drug efficacy in Phase II. FDA approval is pending. Adverse events were recorded per
+protocol. IRB-approved informed consent obtained from all participants. 200 mg IV dose.`,
+    },
+    // ── legal ──────────────────────────────────────────────────────────────
+    {
+        id: "legal-1",
+        path: "legal/nda.txt",
+        expected: "legal",
+        content: `WHEREAS the parties agree to the following terms. Effective Date: 2023-01-01.
+Termination: either party may terminate upon 30 days notice. The receiving party shall
+not disclose Confidential Information. This agreement is governed by the jurisdiction of Delaware.`,
+    },
+    {
+        id: "legal-2",
+        path: "contracts/master-services-agreement.md",
+        expected: "legal",
+        content: `The Contractor shall indemnify the Client for any breach of this Agreement.
+Liability for damages is capped per Section 7.2. Counsel for both parties reviewed the
+provisions. The plaintiff waives any claim arising under this clause.`,
+    },
+    // ── research ───────────────────────────────────────────────────────────
+    {
+        id: "research-1",
+        path: "papers/sparse-attention.md",
+        expected: "research",
+        content: `## Abstract
+We propose a sparse-attention method and demonstrate that it reduces memory cost.
+Our experimental results on the PG-19 dataset improve perplexity over the baseline.
+We build on Beltagy et al. (2020). doi:10.1145/1234567. p < 0.01 across 4 seeds.`,
+    },
+    {
+        id: "research-2",
+        path: "research/ablation-findings.md",
+        expected: "research",
+        content: `## Methodology
+We investigated whether dropout improves generalization. The hypothesis was tested on a
+held-out benchmark. Experimental evaluation reports accuracy: 0.91 and f1-score: 0.88.
+The confidence interval excludes the baseline. Smith et al. (2019) reported similar findings.`,
+    },
+    // ── transcript ─────────────────────────────────────────────────────────
+    {
+        id: "transcript-1",
+        path: "meetings/q4-standup.txt",
+        expected: "transcript",
+        content: `Meeting Minutes - Q4 Standup
+Attendees: John, Sarah, Mike
+Sarah: we agreed to ship the beta on Friday.
+Mike: I'll follow up with the vendor.
+Action Items:
+- John: finalize the budget by next week`,
+    },
+    {
+        id: "transcript-2",
+        path: "transcripts/customer-interview.txt",
+        expected: "transcript",
+        content: `Interviewer: what's your biggest pain point today?
+Participant: onboarding takes too long.
+Interviewer: and how would you fix it?
+Participant: a guided setup. We decided to prototype that next sprint.
+Speaker 2: agreed, let's schedule a review.`,
+    },
+    // ── tabular ────────────────────────────────────────────────────────────
+    {
+        id: "tabular-1",
+        path: "data/sales-2023.csv",
+        expected: "tabular",
+        content: `product,region,units_sold,revenue,margin
+Widget A,EMEA,1500,15000,0.32
+Widget B,APAC,2200,25000,0.28
+Widget C,AMER,900,9000,0.41
+Widget D,EMEA,1750,19500,0.30`,
+    },
+    {
+        id: "tabular-2",
+        path: "exports/inventory.tsv",
+        expected: "tabular",
+        content: `sku\twarehouse\ton_hand\treorder_point\tsupplier
+A-100\tDAL\t420\t100\tAcme
+B-200\tSEA\t85\t120\tGlobex
+C-300\tNYC\t310\t150\tInitech
+D-400\tDAL\t12\t50\tAcme`,
+    },
+    // ── communication ──────────────────────────────────────────────────────
+    {
+        id: "communication-1",
+        path: "mail/budget-update.eml",
+        expected: "communication",
+        content: `From: john@acme.com
+To: sarah@acme.com
+Cc: mike@acme.com
+Subject: Q4 budget draft
+
+Hi Sarah, attached is the budget draft. Please review the travel section.
+Best regards,
+John`,
+    },
+    {
+        id: "communication-2",
+        path: "messages/reply-thread.txt",
+        expected: "communication",
+        content: `On Mon, Nov 3, Sarah Lee wrote:
+> Can you send the updated numbers?
+
+Hi Sarah, sure — forwarded them just now. Thanks for the reminder.
+Regards, John (cc: mike@acme.com)`,
+    },
+    // ── documentation ──────────────────────────────────────────────────────
+    {
+        id: "documentation-1",
+        path: "README.md",
+        expected: "documentation",
+        content: `# Wanshi
+
+Turns files into knowledge graphs.
+
+## Installation
+\`\`\`bash
+npm install -g wanshi
+\`\`\`
+
+## Getting Started
+To get started, run the CLI. You can configure the output format in config.yaml.`,
+    },
+    {
+        id: "documentation-2",
+        path: "docs/api-guide.md",
+        expected: "documentation",
+        content: `## API Reference
+This guide shows how to use the endpoints. Follow these steps to authenticate.
+
+\`\`\`bash
+GET /graph
+\`\`\`
+
+You need to set the API key first. See the configuration section for parameters.`,
+    },
+    // ── technical ──────────────────────────────────────────────────────────
+    {
+        id: "technical-1",
+        path: "logs/app.log",
+        expected: "technical",
+        content: `2023-11-03T09:12:44.512Z INFO  server started on port 8080
+2023-11-03T09:12:45.001Z WARN  cache miss for key user:42
+2023-11-03T09:12:46.220Z ERROR connection timeout to database after 5000ms
+2023-11-03T09:12:47.330Z INFO  retrying connection (attempt 2)`,
+    },
+    {
+        id: "technical-2",
+        path: "config/server.yaml",
+        expected: "technical",
+        content: `host: 0.0.0.0
+port: 8080
+timeout: 30
+database:
+  url: postgres://localhost/app
+  pool: 10
+cache:
+  ttl: 300`,
+    },
+    // ── narrative ──────────────────────────────────────────────────────────
+    {
+        id: "narrative-1",
+        path: "articles/ai-and-society.md",
+        expected: "narrative",
+        content: `The future of work is being reshaped by automation. However, the picture is more
+nuanced than the headlines suggest. Furthermore, many roles will evolve rather than vanish.
+According to recent analysis, the transition will be gradual. The story explores both sides.`,
+    },
+    {
+        id: "narrative-2",
+        path: "posts/remote-work-essay.txt",
+        expected: "narrative",
+        content: `This essay examines how remote work changed team culture. It discusses the trade-offs
+between flexibility and connection. Therefore, companies must rethink their norms.
+Meanwhile, employees report higher satisfaction. The article argues for a hybrid approach.`,
+    },
+    // ── reference ──────────────────────────────────────────────────────────
+    {
+        id: "reference-1",
+        path: "reference/glossary.md",
+        expected: "reference",
+        content: `Entity: a uniquely named node in the knowledge graph.
+Observation: a provenance-stamped fact attached to an entity.
+Relation: a typed directed edge between two entities.
+See also: cross-reference the schema definition for each term.`,
+    },
+    {
+        id: "reference-2",
+        path: "specs/abbreviations.md",
+        expected: "reference",
+        content: `API: Application Programming Interface.
+DOI: Digital Object Identifier.
+KG: Knowledge Graph.
+This catalog lists common acronyms alphabetically. Each entry is indexed by its symbol.`,
+    },
+    // ── hard / ambiguous (neutral paths, cross-cutting signals) ─────────────
+    // These are the cases that matter for S2/S3: confusable content and paths
+    // that carry no telltale extension, so the classifier must decide on content
+    // alone. The gold label is the most defensible single class, but the current
+    // heuristic is expected to confuse several of these — that is the point.
+    {
+        id: "hard-research-prose",
+        path: "notes/sparse-idea.md",
+        expected: "research",
+        content: `We hypothesize that sparse attention preserves accuracy while cutting cost.
+Our analysis on a held-out benchmark suggests the approach generalizes. The methodology
+extends prior work, and the experimental results beat the baseline on most metrics.`,
+    },
+    {
+        id: "hard-narrative-sciencey",
+        path: "articles/science-column.md",
+        expected: "narrative",
+        content: `The promise of fusion energy has captivated scientists for decades. According to
+researchers, recent breakthroughs bring it closer. However, skeptics argue the timeline
+remains uncertain. This article explores what the latest experiments mean for the public.`,
+    },
+    {
+        id: "hard-doc-plain",
+        path: "docs/overview.txt",
+        expected: "documentation",
+        content: `This page explains how to set up the tool. You can install it with the package
+manager, then run the command. To get started, follow these steps and configure the
+options. See the usage section for more, and refer to the tutorial for a walkthrough.`,
+    },
+    {
+        id: "hard-reference-defs",
+        path: "notes/terms.txt",
+        expected: "reference",
+        content: `Idempotent: an operation that yields the same result when applied repeatedly.
+Latency: the time between a request and its response.
+Throughput: the number of operations completed per unit time.
+See also: the performance glossary for related definitions.`,
+    },
+    {
+        id: "hard-technical-plain",
+        path: "infra/setup.txt",
+        expected: "technical",
+        content: `host: api.internal
+port: 9090
+timeout: 60
+The service connects to the database and the cache. On startup it loads the configuration
+and binds to the port. Restart the daemon after changing any setting.`,
+    },
+    {
+        id: "hard-code-in-md",
+        path: "notes/snippet.md",
+        expected: "code",
+        content: `Here's the helper I wrote:
+
+\`\`\`js
+function debounce(fn, ms) {
+  let t;
+  return (...args) => { clearTimeout(t); t = setTimeout(() => fn(...args), ms); };
+}
+\`\`\``,
+    },
+    {
+        id: "hard-communication-chat",
+        path: "chat/dm.md",
+        expected: "communication",
+        content: `hey, did you get a chance to look at the proposal?
+yeah, sent you my notes a minute ago. let me know if the budget section makes sense.
+thanks! I'll reply after the call. cc'ing Dana so she's in the loop.`,
+    },
+    {
+        id: "hard-transcript-plain",
+        path: "audio/recording.txt",
+        expected: "transcript",
+        content: `Anna: so where did we land on the launch date?
+Ben: I think the 15th is realistic if QA signs off.
+Anna: okay, let's plan for that and revisit Thursday.
+Ben: works for me. I'll update the schedule.`,
+    },
+    {
+        id: "hard-medical-prose",
+        path: "health/summary.txt",
+        expected: "medical",
+        content: `The patient has a chronic condition managed with ongoing treatment. Symptoms include
+fatigue and elevated blood pressure. The care team adjusted the therapeutic plan and will
+monitor the diagnosis. A follow-up visit is scheduled to review the prescription.`,
+    },
+    {
+        id: "hard-financial-prose",
+        path: "memos/quarter.txt",
+        expected: "financial",
+        content: `Revenue grew this quarter and earnings exceeded our internal forecast. The investment
+in the new line is paying off, and the board expects the dividend to hold. Analysts remain
+bullish on the stock despite a softer outlook for the sector.`,
+    },
+    {
+        id: "hard-legal-light",
+        path: "policy/terms.txt",
+        expected: "legal",
+        content: `By using this service you agree to these terms. The provider may terminate access for
+breach of the conditions. Each party retains liability only as set out in this agreement.
+Disputes are subject to the courts of the stated jurisdiction.`,
+    },
+    {
+        id: "hard-tabular-markdown",
+        path: "reports/summary.md",
+        expected: "tabular",
+        content: `| product | region | units | revenue |
+| ------- | ------ | ----- | ------- |
+| Widget A | EMEA | 1500 | 15000 |
+| Widget B | APAC | 2200 | 25000 |
+| Widget C | AMER | 900 | 9000 |`,
+    },
+];
+//# sourceMappingURL=labeledSamples.js.map

package/dist/evaluation/classifier/labeledSamples.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"labeledSamples.js","sourceRoot":"","sources":["../../../src/evaluation/classifier/labeledSamples.ts"],"names":[],"mappings":";;;AAoBa,QAAA,eAAe,GAAoB;IAC9C,0EAA0E;IAC1E;QACE,EAAE,EAAE,QAAQ;QACZ,IAAI,EAAE,6BAA6B;QACnC,QAAQ,EAAE,MAAM;QAChB,OAAO,EAAE;;;;;;;;;;EAUX;KACC;IACD;QACE,EAAE,EAAE,QAAQ;QACZ,IAAI,EAAE,uBAAuB;QAC7B,QAAQ,EAAE,MAAM;QAChB,OAAO,EAAE;;;;;;;;;;;;gCAYmB;KAC7B;IAED,0EAA0E;IAC1E;QACE,EAAE,EAAE,aAAa;QACjB,IAAI,EAAE,wBAAwB;QAC9B,QAAQ,EAAE,WAAW;QACrB,OAAO,EAAE;;kFAEqE;KAC/E;IACD;QACE,EAAE,EAAE,aAAa;QACjB,IAAI,EAAE,8BAA8B;QACpC,QAAQ,EAAE,WAAW;QACrB,OAAO,EAAE;;yFAE4E;KACtF;IAED,0EAA0E;IAC1E;QACE,EAAE,EAAE,WAAW;QACf,IAAI,EAAE,0BAA0B;QAChC,QAAQ,EAAE,SAAS;QACnB,OAAO,EAAE;;+EAEkE;KAC5E;IACD;QACE,EAAE,EAAE,WAAW;QACf,IAAI,EAAE,2BAA2B;QACjC,QAAQ,EAAE,SAAS;QACnB,OAAO,EAAE;;wFAE2E;KACrF;IAED,0EAA0E;IAC1E;QACE,EAAE,EAAE,SAAS;QACb,IAAI,EAAE,eAAe;QACrB,QAAQ,EAAE,OAAO;QACjB,OAAO,EAAE;;mGAEsF;KAChG;IACD;QACE,EAAE,EAAE,SAAS;QACb,IAAI,EAAE,wCAAwC;QAC9C,QAAQ,EAAE,OAAO;QACjB,OAAO,EAAE;;sEAEyD;KACnE;IAED,0EAA0E;IAC1E;QACE,EAAE,EAAE,YAAY;QAChB,IAAI,EAAE,4BAA4B;QAClC,QAAQ,EAAE,UAAU;QACpB,OAAO,EAAE;;;iFAGoE;KAC9E;IACD;QACE,EAAE,EAAE,YAAY;QAChB,IAAI,EAAE,+BAA+B;QACrC,QAAQ,EAAE,UAAU;QACpB,OAAO,EAAE;;;8FAGiF;KAC3F;IAED,0EAA0E;IAC1E;QACE,EAAE,EAAE,cAAc;QAClB,IAAI,EAAE,yBAAyB;QAC/B,QAAQ,EAAE,YAAY;QACtB,OAAO,EAAE;;;;;yCAK4B;KACtC;IACD;QACE,EAAE,EAAE,cAAc;QAClB,IAAI,EAAE,oCAAoC;QAC1C,QAAQ,EAAE,YAAY;QACtB,OAAO,EAAE;;;;4CAI+B;KACzC;IAED,0EAA0E;IAC1E;QACE,EAAE,EAAE,WAAW;QACf,IAAI,EAAE,qBAAqB;QAC3B,QAAQ,EAAE,SAAS;QACnB,OAAO,EAAE;;;;8BAIiB;KAC3B;IACD;QACE,EAAE,EAAE,WAAW;QACf,IAAI,EAAE,uBAAuB;QAC7B,QAAQ,EAAE,SAAS;QACnB,OAAO,EAAE;;;;yBAIY;KACtB;IAED,0EAA0E;IAC1E;QACE,EAAE,EAAE,iBAAiB;QACrB,IAAI,EAAE,wBAAwB;QAC9B,QAAQ,EAAE,eAAe;QACzB,OAAO,EAAE;;;;;;;KAOR;KACF;IACD;QACE,EAAE,EAAE,iBAAiB;QACrB,IAAI,EAAE,2BAA2B;QACjC,QAAQ,EAAE,eAAe;QACzB,OAAO,EAAE;;;;kCAIqB;KAC/B;IAED,0EAA0E;IAC1E;QACE,EAAE,EAAE,iBAAiB;QACrB,IAAI,EAAE,WAAW;QACjB,QAAQ,EAAE,eAAe;QACzB,OAAO,EAAE;;;;;;;;;;iFAUoE;KAC9E;IACD;QACE,EAAE,EAAE,iBAAiB;QACrB,IAAI,EAAE,mBAAmB;QACzB,QAAQ,EAAE,eAAe;QACzB,OAAO,EAAE;;;;;;;iFAOoE;KAC9E;IAED,0EAA0E;IAC1E;QACE,EAAE,EAAE,aAAa;QACjB,IAAI,EAAE,cAAc;QACpB,QAAQ,EAAE,WAAW;QACrB,OAAO,EAAE;;;+DAGkD;KAC5D;IACD;QACE,EAAE,EAAE,aAAa;QACjB,IAAI,EAAE,oBAAoB;QAC1B,QAAQ,EAAE,WAAW;QACrB,OAAO,EAAE;;;;;;;WAOF;KACR;IAED,0EAA0E;IAC1E;QACE,EAAE,EAAE,aAAa;QACjB,IAAI,EAAE,4BAA4B;QAClC,QAAQ,EAAE,WAAW;QACrB,OAAO,EAAE;;6FAEgF;KAC1F;IACD;QACE,EAAE,EAAE,aAAa;QACjB,IAAI,EAAE,6BAA6B;QACnC,QAAQ,EAAE,WAAW;QACrB,OAAO,EAAE;;2FAE8E;KACxF;IAED,0EAA0E;IAC1E;QACE,EAAE,EAAE,aAAa;QACjB,IAAI,EAAE,uBAAuB;QAC7B,QAAQ,EAAE,WAAW;QACrB,OAAO,EAAE;;;+DAGkD;KAC5D;IACD;QACE,EAAE,EAAE,aAAa;QACjB,IAAI,EAAE,wBAAwB;QAC9B,QAAQ,EAAE,WAAW;QACrB,OAAO,EAAE;;;wFAG2E;KACrF;IAED,2EAA2E;IAC3E,0EAA0E;IAC1E,6EAA6E;IAC7E,6EAA6E;IAC7E,yEAAyE;IACzE;QACE,EAAE,EAAE,qBAAqB;QACzB,IAAI,EAAE,sBAAsB;QAC5B,QAAQ,EAAE,UAAU;QACpB,OAAO,EAAE;;oFAEuE;KACjF;IACD;QACE,EAAE,EAAE,yBAAyB;QAC7B,IAAI,EAAE,4BAA4B;QAClC,QAAQ,EAAE,WAAW;QACrB,OAAO,EAAE;;0FAE6E;KACvF;IACD;QACE,EAAE,EAAE,gBAAgB;QACpB,IAAI,EAAE,mBAAmB;QACzB,QAAQ,EAAE,eAAe;QACzB,OAAO,EAAE;;sFAEyE;KACnF;IACD;QACE,EAAE,EAAE,qBAAqB;QACzB,IAAI,EAAE,iBAAiB;QACvB,QAAQ,EAAE,WAAW;QACrB,OAAO,EAAE;;;4DAG+C;KACzD;IACD;QACE,EAAE,EAAE,sBAAsB;QAC1B,IAAI,EAAE,iBAAiB;QACvB,QAAQ,EAAE,WAAW;QACrB,OAAO,EAAE;;;;sEAIyD;KACnE;IACD;QACE,EAAE,EAAE,iBAAiB;QACrB,IAAI,EAAE,kBAAkB;QACxB,QAAQ,EAAE,MAAM;QAChB,OAAO,EAAE;;;;;;;OAON;KACJ;IACD;QACE,EAAE,EAAE,yBAAyB;QAC7B,IAAI,EAAE,YAAY;QAClB,QAAQ,EAAE,eAAe;QACzB,OAAO,EAAE;;qEAEwD;KAClE;IACD;QACE,EAAE,EAAE,uBAAuB;QAC3B,IAAI,EAAE,qBAAqB;QAC3B,QAAQ,EAAE,YAAY;QACtB,OAAO,EAAE;;;6CAGgC;KAC1C;IACD;QACE,EAAE,EAAE,oBAAoB;QACxB,IAAI,EAAE,oBAAoB;QAC1B,QAAQ,EAAE,SAAS;QACnB,OAAO,EAAE;;kFAEqE;KAC/E;IACD;QACE,EAAE,EAAE,sBAAsB;QAC1B,IAAI,EAAE,mBAAmB;QACzB,QAAQ,EAAE,WAAW;QACrB,OAAO,EAAE;;8DAEiD;KAC3D;IACD;QACE,EAAE,EAAE,kBAAkB;QACtB,IAAI,EAAE,kBAAkB;QACxB,QAAQ,EAAE,OAAO;QACjB,OAAO,EAAE;;+DAEkD;KAC5D;IACD;QACE,EAAE,EAAE,uBAAuB;QAC3B,IAAI,EAAE,oBAAoB;QAC1B,QAAQ,EAAE,SAAS;QACnB,OAAO,EAAE;;;;iCAIoB;KAC9B;CACF,CAAC"}

package/dist/evaluation/compare/goldCompare.js

@@ -0,0 +1,126 @@
+"use strict";
+// Shared scoring core for the gold-labeled two-way comparisons (wanshi vs KGGen).
+//
+// Dataset-agnostic: CrossRE (sentence-level, 6 domains), SemEval (sentence-level,
+// no domains), Re-DocRED (document-level, Wikidata schema + Ign-F1). The CLI
+// (scripts/gold-compare.ts) handles loading/extraction/caching; this module owns
+// the metric math so there is ONE source of truth across datasets.
+//
+// HEADLINE = node entity-capture (semantic): every graph node → a self-triplet →
+// matched against gold entities. Fair across open-predicate tools whose free
+// predicates won't string/embed-match a dataset's abstract gold predicate vocab.
+// Triplet-level entity/relation/triple F1 are also reported (understate uniformly).
+//
+// Optional per-call extras:
+//   - domainById  → per-domain node-capture F1 (CrossRE).
+//   - ignoreKeys  → Re-DocRED Ign-F1: drop train-seen (subj|pred|obj) triples from
+//                   BOTH gold and predictions before the triplet-level metrics, so a
+//                   tool gets no credit for memorized training facts.
+var __awaiter = (this && this.__awaiter) || function (thisArg, _arguments, P, generator) {
+    function adopt(value) { return value instanceof P ? value : new P(function (resolve) { resolve(value); }); }
+    return new (P || (P = Promise))(function (resolve, reject) {
+        function fulfilled(value) { try { step(generator.next(value)); } catch (e) { reject(e); } }
+        function rejected(value) { try { step(generator["throw"](value)); } catch (e) { reject(e); } }
+        function step(result) { result.done ? resolve(result.value) : adopt(result.value).then(fulfilled, rejected); }
+        step((generator = generator.apply(thisArg, _arguments || [])).next());
+    });
+};
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.tripleKey = tripleKey;
+exports.scoreGraph = scoreGraph;
+exports.loadJsonl = loadJsonl;
+exports.appendJsonl = appendJsonl;
+const TripleMetrics_1 = require("../metrics/TripleMetrics");
+const compareScoring_1 = require("../crossre/compareScoring");
+const addTally = (a, b) => {
+    a.tp += b.tp;
+    a.fp += b.fp;
+    a.fn += b.fn;
+};
+/** Normalized (subject|predicate|object) key for train-seen-triple exclusion (Ign-F1). */
+function tripleKey(t) {
+    return `${t.subject.trim().toLowerCase()}|${t.predicate.trim().toLowerCase()}|${t.object.trim().toLowerCase()}`;
+}
+function scoreGraph(ids, graphById, goldById, exact, semantic, opts) {
+    return __awaiter(this, void 0, void 0, function* () {
+        var _a, _b, _c;
+        const domainById = opts === null || opts === void 0 ? void 0 : opts.domainById;
+        const ignoreKeys = opts === null || opts === void 0 ? void 0 : opts.ignoreKeys;
+        const exactTrip = [];
+        const semTrip = [];
+        const ignExactTrip = [];
+        const ignSemTrip = [];
+        const nodeSem = { tp: 0, fp: 0, fn: 0 }; // micro-averaged across samples
+        const nodeExact = { tp: 0, fp: 0, fn: 0 };
+        const nodeByDomain = new Map();
+        let triples = 0, ents = 0;
+        for (const id of ids) {
+            const kg = (_a = graphById.get(id)) !== null && _a !== void 0 ? _a : { entities: [], relations: [] };
+            const gold = (_b = goldById.get(id)) !== null && _b !== void 0 ? _b : [];
+            const trip = (0, compareScoring_1.kgToTriplets)(kg);
+            const nodes = (0, compareScoring_1.nodeTriplets)(kg);
+            triples += trip.length;
+            ents += kg.entities.length;
+            exactTrip.push((0, TripleMetrics_1.computeExactMetrics)(trip, gold, exact));
+            semTrip.push(yield (0, TripleMetrics_1.computeSemanticMetrics)(trip, gold, semantic));
+            // Ign-F1: drop train-seen triples from both sides before scoring.
+            if (ignoreKeys) {
+                const tripF = trip.filter((t) => !ignoreKeys.has(tripleKey(t)));
+                const goldF = gold.filter((t) => !ignoreKeys.has(tripleKey(t)));
+                ignExactTrip.push((0, TripleMetrics_1.computeExactMetrics)(tripF, goldF, exact));
+                ignSemTrip.push(yield (0, TripleMetrics_1.computeSemanticMetrics)(tripF, goldF, semantic));
+            }
+            // Node entity-capture: match the full node set against gold entities.
+            const ns = yield semantic.matchEntities(nodes, gold);
+            addTally(nodeSem, ns);
+            addTally(nodeExact, exact.matchEntities(nodes, gold));
+            if (domainById) {
+                const d = (_c = domainById.get(id)) !== null && _c !== void 0 ? _c : 'unknown';
+                if (!nodeByDomain.has(d))
+                    nodeByDomain.set(d, { tp: 0, fp: 0, fn: 0 });
+                addTally(nodeByDomain.get(d), ns);
+            }
+        }
+        let perDomainNode;
+        if (domainById) {
+            perDomainNode = new Map();
+            for (const [d, t] of nodeByDomain)
+                perDomainNode.set(d, (0, TripleMetrics_1.computeMetrics)(t.tp, t.fp, t.fn));
+        }
+        return {
+            nodeEntitySem: (0, TripleMetrics_1.computeMetrics)(nodeSem.tp, nodeSem.fp, nodeSem.fn),
+            nodeEntityExact: (0, TripleMetrics_1.computeMetrics)(nodeExact.tp, nodeExact.fp, nodeExact.fn),
+            tripletSem: (0, TripleMetrics_1.microAverage)(semTrip),
+            tripletExact: (0, TripleMetrics_1.microAverage)(exactTrip),
+            perDomainNode,
+            ignTripletSem: ignoreKeys ? (0, TripleMetrics_1.microAverage)(ignSemTrip) : undefined,
+            ignTripletExact: ignoreKeys ? (0, TripleMetrics_1.microAverage)(ignExactTrip) : undefined,
+            triplesPer: ids.length ? triples / ids.length : 0,
+            entsPer: ids.length ? ents / ids.length : 0,
+        };
+    });
+}
+// ─── JSONL cache (append + load + truncation-tolerant; the CheckpointService idiom) ──
+function loadJsonl(file, fs) {
+    const map = new Map();
+    if (!fs.existsSync(file))
+        return map;
+    for (const line of fs.readFileSync(file, 'utf-8').split('\n')) {
+        const t = line.trim();
+        if (!t)
+            continue;
+        try {
+            const rec = JSON.parse(t);
+            if (rec && rec.id !== undefined)
+                map.set(rec.id, rec);
+        }
+        catch (_a) {
+            /* tolerate a truncated final line from an interrupted write */
+        }
+    }
+    return map;
+}
+function appendJsonl(file, rec, fs) {
+    fs.appendFileSync(file, JSON.stringify(rec) + '\n', 'utf-8');
+}
+//# sourceMappingURL=goldCompare.js.map

package/dist/evaluation/compare/goldCompare.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"goldCompare.js","sourceRoot":"","sources":["../../../src/evaluation/compare/goldCompare.ts"],"names":[],"mappings":";AAAA,kFAAkF;AAClF,EAAE;AACF,kFAAkF;AAClF,6EAA6E;AAC7E,iFAAiF;AACjF,mEAAmE;AACnE,EAAE;AACF,iFAAiF;AACjF,6EAA6E;AAC7E,iFAAiF;AACjF,oFAAoF;AACpF,EAAE;AACF,4BAA4B;AAC5B,0DAA0D;AAC1D,mFAAmF;AACnF,qFAAqF;AACrF,sEAAsE;;;;;;;;;;;AA+BtE,8BAEC;AAED,gCAoEC;AAGD,8BAcC;AAED,kCAEC;AAvHD,4DAAqH;AAErH,8DAAuE;AAmBvE,MAAM,QAAQ,GAAG,CAAC,CAAQ,EAAE,CAAyC,EAAE,EAAE;IACvE,CAAC,CAAC,EAAE,IAAI,CAAC,CAAC,EAAE,CAAC;IAAC,CAAC,CAAC,EAAE,IAAI,CAAC,CAAC,EAAE,CAAC;IAAC,CAAC,CAAC,EAAE,IAAI,CAAC,CAAC,EAAE,CAAC;AAC3C,CAAC,CAAC;AAEF,0FAA0F;AAC1F,SAAgB,SAAS,CAAC,CAAU;IAClC,OAAO,GAAG,CAAC,CAAC,OAAO,CAAC,IAAI,EAAE,CAAC,WAAW,EAAE,IAAI,CAAC,CAAC,SAAS,CAAC,IAAI,EAAE,CAAC,WAAW,EAAE,IAAI,CAAC,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC,WAAW,EAAE,EAAE,CAAC;AAClH,CAAC;AAED,SAAsB,UAAU,CAC9B,GAAa,EACb,SAAsC,EACtC,QAAgC,EAChC,KAAmB,EACnB,QAAyB,EACzB,IAAqE;;;QAErE,MAAM,UAAU,GAAG,IAAI,aAAJ,IAAI,uBAAJ,IAAI,CAAE,UAAU,CAAC;QACpC,MAAM,UAAU,GAAG,IAAI,aAAJ,IAAI,uBAAJ,IAAI,CAAE,UAAU,CAAC;QAEpC,MAAM,SAAS,GAAmB,EAAE,CAAC;QACrC,MAAM,OAAO,GAAmB,EAAE,CAAC;QACnC,MAAM,YAAY,GAAmB,EAAE,CAAC;QACxC,MAAM,UAAU,GAAmB,EAAE,CAAC;QACtC,MAAM,OAAO,GAAU,EAAE,EAAE,EAAE,CAAC,EAAE,EAAE,EAAE,CAAC,EAAE,EAAE,EAAE,CAAC,EAAE,CAAC,CAAG,gCAAgC;QAClF,MAAM,SAAS,GAAU,EAAE,EAAE,EAAE,CAAC,EAAE,EAAE,EAAE,CAAC,EAAE,EAAE,EAAE,CAAC,EAAE,CAAC;QACjD,MAAM,YAAY,GAAG,IAAI,GAAG,EAAiB,CAAC;QAC9C,IAAI,OAAO,GAAG,CAAC,EAAE,IAAI,GAAG,CAAC,CAAC;QAE1B,KAAK,MAAM,EAAE,IAAI,GAAG,EAAE,CAAC;YACrB,MAAM,EAAE,GAAG,MAAA,SAAS,CAAC,GAAG,CAAC,EAAE,CAAC,mCAAI,EAAE,QAAQ,EAAE,EAAE,EAAE,SAAS,EAAE,EAAE,EAAE,CAAC;YAChE,MAAM,IAAI,GAAG,MAAA,QAAQ,CAAC,GAAG,CAAC,EAAE,CAAC,mCAAI,EAAE,CAAC;YACpC,MAAM,IAAI,GAAG,IAAA,6BAAY,EAAC,EAAE,CAAC,CAAC;YAC9B,MAAM,KAAK,GAAG,IAAA,6BAAY,EAAC,EAAE,CAAC,CAAC;YAC/B,OAAO,IAAI,IAAI,CAAC,MAAM,CAAC;YACvB,IAAI,IAAI,EAAE,CAAC,QAAQ,CAAC,MAAM,CAAC;YAE3B,SAAS,CAAC,IAAI,CAAC,IAAA,mCAAmB,EAAC,IAAI,EAAE,IAAI,EAAE,KAAK,CAAC,CAAC,CAAC;YACvD,OAAO,CAAC,IAAI,CAAC,MAAM,IAAA,sCAAsB,EAAC,IAAI,EAAE,IAAI,EAAE,QAAQ,CAAC,CAAC,CAAC;YAEjE,kEAAkE;YAClE,IAAI,UAAU,EAAE,CAAC;gBACf,MAAM,KAAK,GAAG,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,UAAU,CAAC,GAAG,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;gBAChE,MAAM,KAAK,GAAG,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,UAAU,CAAC,GAAG,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;gBAChE,YAAY,CAAC,IAAI,CAAC,IAAA,mCAAmB,EAAC,KAAK,EAAE,KAAK,EAAE,KAAK,CAAC,CAAC,CAAC;gBAC5D,UAAU,CAAC,IAAI,CAAC,MAAM,IAAA,sCAAsB,EAAC,KAAK,EAAE,KAAK,EAAE,QAAQ,CAAC,CAAC,CAAC;YACxE,CAAC;YAED,sEAAsE;YACtE,MAAM,EAAE,GAAG,MAAM,QAAQ,CAAC,aAAa,CAAC,KAAK,EAAE,IAAI,CAAC,CAAC;YACrD,QAAQ,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC;YACtB,QAAQ,CAAC,SAAS,EAAE,KAAK,CAAC,aAAa,CAAC,KAAK,EAAE,IAAI,CAAC,CAAC,CAAC;YAEtD,IAAI,UAAU,EAAE,CAAC;gBACf,MAAM,CAAC,GAAG,MAAA,UAAU,CAAC,GAAG,CAAC,EAAE,CAAC,mCAAI,SAAS,CAAC;gBAC1C,IAAI,CAAC,YAAY,CAAC,GAAG,CAAC,CAAC,CAAC;oBAAE,YAAY,CAAC,GAAG,CAAC,CAAC,EAAE,EAAE,EAAE,EAAE,CAAC,EAAE,EAAE,EAAE,CAAC,EAAE,EAAE,EAAE,CAAC,EAAE,CAAC,CAAC;gBACvE,QAAQ,CAAC,YAAY,CAAC,GAAG,CAAC,CAAC,CAAE,EAAE,EAAE,CAAC,CAAC;YACrC,CAAC;QACH,CAAC;QAED,IAAI,aAAmD,CAAC;QACxD,IAAI,UAAU,EAAE,CAAC;YACf,aAAa,GAAG,IAAI,GAAG,EAAE,CAAC;YAC1B,KAAK,MAAM,CAAC,CAAC,EAAE,CAAC,CAAC,IAAI,YAAY;gBAAE,aAAa,CAAC,GAAG,CAAC,CAAC,EAAE,IAAA,8BAAc,EAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;QAC5F,CAAC;QAED,OAAO;YACL,aAAa,EAAE,IAAA,8BAAc,EAAC,OAAO,CAAC,EAAE,EAAE,OAAO,CAAC,EAAE,EAAE,OAAO,CAAC,EAAE,CAAC;YACjE,eAAe,EAAE,IAAA,8BAAc,EAAC,SAAS,CAAC,EAAE,EAAE,SAAS,CAAC,EAAE,EAAE,SAAS,CAAC,EAAE,CAAC;YACzE,UAAU,EAAE,IAAA,4BAAY,EAAC,OAAO,CAAC;YACjC,YAAY,EAAE,IAAA,4BAAY,EAAC,SAAS,CAAC;YACrC,aAAa;YACb,aAAa,EAAE,UAAU,CAAC,CAAC,CAAC,IAAA,4BAAY,EAAC,UAAU,CAAC,CAAC,CAAC,CAAC,SAAS;YAChE,eAAe,EAAE,UAAU,CAAC,CAAC,CAAC,IAAA,4BAAY,EAAC,YAAY,CAAC,CAAC,CAAC,CAAC,SAAS;YACpE,UAAU,EAAE,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,OAAO,GAAG,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC;YACjD,OAAO,EAAE,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,IAAI,GAAG,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC;SAC5C,CAAC;IACJ,CAAC;CAAA;AAED,wFAAwF;AACxF,SAAgB,SAAS,CAAU,IAAY,EAAE,EAAuB;IACtE,MAAM,GAAG,GAAG,IAAI,GAAG,EAAa,CAAC;IACjC,IAAI,CAAC,EAAE,CAAC,UAAU,CAAC,IAAI,CAAC;QAAE,OAAO,GAAG,CAAC;IACrC,KAAK,MAAM,IAAI,IAAI,EAAE,CAAC,YAAY,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC;QAC9D,MAAM,CAAC,GAAG,IAAI,CAAC,IAAI,EAAE,CAAC;QACtB,IAAI,CAAC,CAAC;YAAE,SAAS;QACjB,IAAI,CAAC;YACH,MAAM,GAAG,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;YAC1B,IAAI,GAAG,IAAI,GAAG,CAAC,EAAE,KAAK,SAAS;gBAAE,GAAG,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,EAAE,GAAG,CAAC,CAAC;QACxD,CAAC;QAAC,WAAM,CAAC;YACP,+DAA+D;QACjE,CAAC;IACH,CAAC;IACD,OAAO,GAAG,CAAC;AACb,CAAC;AAED,SAAgB,WAAW,CAAC,IAAY,EAAE,GAAY,EAAE,EAAuB;IAC7E,EAAE,CAAC,cAAc,CAAC,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,GAAG,IAAI,EAAE,OAAO,CAAC,CAAC;AAC/D,CAAC"}

package/dist/evaluation/crossre/compareScoring.js

@@ -0,0 +1,30 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.kgToTriplets = kgToTriplets;
+exports.nodeTriplets = nodeTriplets;
+/**
+ * Flatten a KnowledgeGraph's relations into triplets — one per relationType label.
+ * An empty/blank relationType falls back to `related to`, mirroring
+ * `MineDataset.toGraph` (which gives KGGen's empty edges the same label). Without
+ * this, wanshi relations that came back with `relationType: []` would silently
+ * vanish from the triplet metrics while KGGen's empty edges survive — an asymmetry
+ * that would understate wanshi on the relation/triple levels.
+ */
+function kgToTriplets(kg) {
+    return kg.relations.flatMap((r) => {
+        const labels = r.relationType.length ? r.relationType : ['related to'];
+        return labels.map((rel) => ({ subject: r.from, predicate: rel || 'related to', object: r.to }));
+    });
+}
+/**
+ * Represent each graph NODE as a self-triplet so the existing
+ * Exact/SemanticMatcher.matchEntities (which reads subjects ∪ objects) measures
+ * entity-capture over the full node set — NOT just relation endpoints. This is the
+ * fair cross-tool headline: "did the tool recover the gold entities at all",
+ * independent of whether it also drew an edge between them (wanshi is edge-sparse,
+ * so scoring entities only via relation endpoints would understate it).
+ */
+function nodeTriplets(kg) {
+    return kg.entities.map((e) => ({ subject: e.name, predicate: '', object: e.name }));
+}
+//# sourceMappingURL=compareScoring.js.map

package/dist/evaluation/crossre/compareScoring.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"compareScoring.js","sourceRoot":"","sources":["../../../src/evaluation/crossre/compareScoring.ts"],"names":[],"mappings":";;AAWA,oCAKC;AAUD,oCAEC;AAzBD;;;;;;;GAOG;AACH,SAAgB,YAAY,CAAC,EAAkB;IAC7C,OAAO,EAAE,CAAC,SAAS,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,EAAE;QAChC,MAAM,MAAM,GAAG,CAAC,CAAC,YAAY,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC,YAAY,CAAC,CAAC;QACvE,OAAO,MAAM,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,CAAC,EAAE,OAAO,EAAE,CAAC,CAAC,IAAI,EAAE,SAAS,EAAE,GAAG,IAAI,YAAY,EAAE,MAAM,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC;IAClG,CAAC,CAAC,CAAC;AACL,CAAC;AAED;;;;;;;GAOG;AACH,SAAgB,YAAY,CAAC,EAAkB;IAC7C,OAAO,EAAE,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,EAAE,OAAO,EAAE,CAAC,CAAC,IAAI,EAAE,SAAS,EAAE,EAAE,EAAE,MAAM,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC;AACtF,CAAC"}