d365fo-mcp 1.0.0

package/dist/eval/oracle/actualArtifactResolution.js

@@ -0,0 +1,73 @@
+/**
+ * Multi-artifact (`--actual-dir`) actual-file resolution for src/eval/oracle/cli.ts.
+ * Pure logic + fs reads — no CLI argv/process side effects — split out so it can be
+ * unit-tested without triggering the CLI script's `main()` (docs/AGENT_EVAL_LOOP.md §6).
+ */
+import * as fs from 'fs';
+import * as path from 'path';
+import { canonicalizePrefix } from './normalize.js';
+/**
+ * Resolve the actual-dir file matching a golden artifact filename. Tries an
+ * exact filename match first (fast path, and the only path when golden and
+ * actual happen to share the same EXTENSION_PREFIX session); if that misses,
+ * falls back to matching on the PREFIX-CANONICALISED filename (the golden's
+ * filename is itself typically a prefixed object name, e.g.
+ * "ContosoMyContract.metadata.xml", produced under a different session than the
+ * one that generated the actual artifacts) so a whole L3/L4 multi-artifact
+ * case doesn't spuriously score every artifact as missing/extra under prefix
+ * drift alone.
+ */
+export function resolveActualFile(actualDir, goldenName, goldenPrefix, actualPrefix) {
+    const direct = path.join(actualDir, goldenName);
+    if (fs.existsSync(direct))
+        return direct;
+    const canonGolden = canonicalizePrefix(goldenName, goldenPrefix);
+    const candidate = fs.readdirSync(actualDir)
+        .filter(f => f.endsWith('.metadata.xml'))
+        .find(f => canonicalizePrefix(f, actualPrefix) === canonGolden);
+    return candidate ? path.join(actualDir, candidate) : undefined;
+}
+/**
+ * Build the `actualArtifacts` map for a multi-artifact (`--actual-dir`) run,
+ * one entry per golden artifact name.
+ *
+ * Regression: this used to key every entry by the GOLDEN's own filename
+ * (`actualArtifacts[name] = ...` inside a `for (const name of
+ * artifactNames)` loop) even when the resolved actual file had a DIFFERENT
+ * literal prefix (e.g. golden "ContosoMyContract.metadata.xml" resolved to actual
+ * file "DemoMyContract.metadata.xml" under prefix-agnostic matching —
+ * `resolveActualFile`'s whole point). `evaluateMulti`/`normalizeMultiArtifact`
+ * then canonicalises each artifact KEY with `actualPrefix` — but a key that's
+ * still the GOLDEN's literal name doesn't contain `actualPrefix` at all, so
+ * `canonicalizePrefix` is a no-op on it, and the golden side's key (correctly
+ * canonicalised from ITS OWN prefix) never matches. Every path in the
+ * artifact then shows up as wholesale `missing` (under the golden's canonical
+ * key) AND `extra` (under the actual's un-canonicalised key), even when the
+ * content is byte-identical. Keying by the RESOLVED actual file's own
+ * basename (which DOES contain `actualPrefix`) fixes the canonicalisation on
+ * both sides consistently — matching the documented multi-artifact contract
+ * (src/eval/oracle/normalize.ts's `normalizeMultiArtifact` doc comment).
+ *
+ * A golden artifact with NO resolvable actual file (genuinely missing, not a
+ * prefix-matching miss) keeps the golden's own name as the key with empty
+ * content — unchanged from before; there is no real actual basename to key it
+ * by, and the empty content correctly registers every one of that artifact's
+ * paths as `missing`.
+ */
+export function buildActualArtifactsMap(actualDir, artifactNames, goldenPrefix, actualPrefix) {
+    const actualArtifacts = {};
+    const matchedActualFiles = new Set();
+    for (const name of artifactNames) {
+        const actualFile = resolveActualFile(actualDir, name, goldenPrefix, actualPrefix);
+        if (actualFile) {
+            const actualBasename = path.basename(actualFile);
+            actualArtifacts[actualBasename] = fs.readFileSync(actualFile, 'utf8');
+            matchedActualFiles.add(actualBasename);
+        }
+        else {
+            actualArtifacts[name] = '';
+        }
+    }
+    return { actualArtifacts, matchedActualFiles };
+}
+//# sourceMappingURL=actualArtifactResolution.js.map

package/dist/eval/oracle/actualArtifactResolution.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"actualArtifactResolution.js","sourceRoot":"","sources":["../../../src/eval/oracle/actualArtifactResolution.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,OAAO,KAAK,EAAE,MAAM,IAAI,CAAC;AACzB,OAAO,KAAK,IAAI,MAAM,MAAM,CAAC;AAC7B,OAAO,EAAE,kBAAkB,EAAE,MAAM,gBAAgB,CAAC;AAEpD;;;;;;;;;;GAUG;AACH,MAAM,UAAU,iBAAiB,CAC/B,SAAiB,EACjB,UAAkB,EAClB,YAAoB,EACpB,YAAoB;IAEpB,MAAM,MAAM,GAAG,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,UAAU,CAAC,CAAC;IAChD,IAAI,EAAE,CAAC,UAAU,CAAC,MAAM,CAAC;QAAE,OAAO,MAAM,CAAC;IACzC,MAAM,WAAW,GAAG,kBAAkB,CAAC,UAAU,EAAE,YAAY,CAAC,CAAC;IACjE,MAAM,SAAS,GAAG,EAAE,CAAC,WAAW,CAAC,SAAS,CAAC;SACxC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,eAAe,CAAC,CAAC;SACxC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,kBAAkB,CAAC,CAAC,EAAE,YAAY,CAAC,KAAK,WAAW,CAAC,CAAC;IAClE,OAAO,SAAS,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,SAAS,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;AACjE,CAAC;AAED;;;;;;;;;;;;;;;;;;;;;;;;;;GA0BG;AACH,MAAM,UAAU,uBAAuB,CACrC,SAAiB,EACjB,aAAuB,EACvB,YAAoB,EACpB,YAAoB;IAEpB,MAAM,eAAe,GAA2B,EAAE,CAAC;IACnD,MAAM,kBAAkB,GAAG,IAAI,GAAG,EAAU,CAAC;IAC7C,KAAK,MAAM,IAAI,IAAI,aAAa,EAAE,CAAC;QACjC,MAAM,UAAU,GAAG,iBAAiB,CAAC,SAAS,EAAE,IAAI,EAAE,YAAY,EAAE,YAAY,CAAC,CAAC;QAClF,IAAI,UAAU,EAAE,CAAC;YACf,MAAM,cAAc,GAAG,IAAI,CAAC,QAAQ,CAAC,UAAU,CAAC,CAAC;YACjD,eAAe,CAAC,cAAc,CAAC,GAAG,EAAE,CAAC,YAAY,CAAC,UAAU,EAAE,MAAM,CAAC,CAAC;YACtE,kBAAkB,CAAC,GAAG,CAAC,cAAc,CAAC,CAAC;QACzC,CAAC;aAAM,CAAC;YACN,eAAe,CAAC,IAAI,CAAC,GAAG,EAAE,CAAC;QAC7B,CAAC;IACH,CAAC;IACD,OAAO,EAAE,eAAe,EAAE,kBAAkB,EAAE,CAAC;AACjD,CAAC"}

package/dist/eval/oracle/cli.d.ts

@@ -0,0 +1,30 @@
+/**
+ * Eval oracle CLI — score a produced artifact against its golden and (optionally)
+ * write a corpus record. VM-free: run after capturing the build result.
+ *
+ *   tsx src/eval/oracle/cli.ts <caseId> <actualXml> [options]
+ *     --golden <path>     explicit golden file (default: first *.metadata.xml in eval/goldens/<caseId>/)
+ *     --actual-dir <dir>  MULTI-ARTIFACT mode: score every *.metadata.xml golden in
+ *                         eval/goldens/<caseId>/ against a same-named file in <dir>
+ *                         (L3/L4 cases that produce several objects). Mutually
+ *                         exclusive with the single <actualXml> positional/--golden.
+ *     --build-failed      mark build as failed (default: succeeded)
+ *     --bp-warnings <n>   number of BP warnings (default: 0)
+ *     --systest <file>    text file with the `run_systest_class` output (runtime oracle)
+ *     --classification <C> rubric class for the record (default: derived)
+ *     --golden-prefix <p> EXTENSION_PREFIX the golden was captured under (default: GOLDEN_CAPTURE_PREFIX, "Contoso")
+ *     --actual-prefix <p> EXTENSION_PREFIX the actual was produced under (default: read from THIS
+ *                         process's EXTENSION_PREFIX env var — the session that ran the case)
+ *     --write             append a corpus record to eval/corpus/runs/
+ *
+ * `<actualXml>` may itself be a golden path to self-check the oracle (expect match).
+ *
+ * Root-object-name (and other prefixed-identifier) comparisons are
+ * prefix-agnostic by default: an actual object built under a DIFFERENT
+ * EXTENSION_PREFIX session than the one the golden was captured under still
+ * scores golden_match=1 as long as the object is otherwise identical (see
+ * docs/AGENT_EVAL_LOOP.md §6.2 and the corpus record that surfaced this —
+ * eval/corpus/runs/2026-07-06T10__L0-edt-basic__4fafcd8.json).
+ */
+export {};
+//# sourceMappingURL=cli.d.ts.map

package/dist/eval/oracle/cli.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"cli.d.ts","sourceRoot":"","sources":["../../../src/eval/oracle/cli.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;GA2BG"}

package/dist/eval/oracle/cli.js

@@ -0,0 +1,179 @@
+/**
+ * Eval oracle CLI — score a produced artifact against its golden and (optionally)
+ * write a corpus record. VM-free: run after capturing the build result.
+ *
+ *   tsx src/eval/oracle/cli.ts <caseId> <actualXml> [options]
+ *     --golden <path>     explicit golden file (default: first *.metadata.xml in eval/goldens/<caseId>/)
+ *     --actual-dir <dir>  MULTI-ARTIFACT mode: score every *.metadata.xml golden in
+ *                         eval/goldens/<caseId>/ against a same-named file in <dir>
+ *                         (L3/L4 cases that produce several objects). Mutually
+ *                         exclusive with the single <actualXml> positional/--golden.
+ *     --build-failed      mark build as failed (default: succeeded)
+ *     --bp-warnings <n>   number of BP warnings (default: 0)
+ *     --systest <file>    text file with the `run_systest_class` output (runtime oracle)
+ *     --classification <C> rubric class for the record (default: derived)
+ *     --golden-prefix <p> EXTENSION_PREFIX the golden was captured under (default: GOLDEN_CAPTURE_PREFIX, "Contoso")
+ *     --actual-prefix <p> EXTENSION_PREFIX the actual was produced under (default: read from THIS
+ *                         process's EXTENSION_PREFIX env var — the session that ran the case)
+ *     --write             append a corpus record to eval/corpus/runs/
+ *
+ * `<actualXml>` may itself be a golden path to self-check the oracle (expect match).
+ *
+ * Root-object-name (and other prefixed-identifier) comparisons are
+ * prefix-agnostic by default: an actual object built under a DIFFERENT
+ * EXTENSION_PREFIX session than the one the golden was captured under still
+ * scores golden_match=1 as long as the object is otherwise identical (see
+ * docs/AGENT_EVAL_LOOP.md §6.2 and the corpus record that surfaced this —
+ * eval/corpus/runs/2026-07-06T10__L0-edt-basic__4fafcd8.json).
+ */
+import * as fs from 'fs';
+import * as path from 'path';
+import { fileURLToPath } from 'url';
+import { execSync } from 'child_process';
+import { evaluate, evaluateMulti, renderDiff, renderNormalized, normalizeAotXml, parseSysTestResult, GOLDEN_CAPTURE_PREFIX, } from './index.js';
+import { resolveRegularObjectPrefixToken } from '../../utils/modelClassifier.js';
+import { buildActualArtifactsMap } from './actualArtifactResolution.js';
+const HERE = path.dirname(fileURLToPath(import.meta.url));
+const REPO_ROOT = path.resolve(HERE, '..', '..', '..');
+function arg(flag) {
+    const i = process.argv.indexOf(flag);
+    return i >= 0 ? process.argv[i + 1] : undefined;
+}
+function flagSet(flag) {
+    return process.argv.includes(flag);
+}
+function goldenDir(caseId) {
+    return path.join(REPO_ROOT, 'eval', 'goldens', caseId);
+}
+function findGolden(caseId) {
+    const dir = goldenDir(caseId);
+    const file = fs.readdirSync(dir).find(f => f.endsWith('.metadata.xml'));
+    if (!file)
+        throw new Error(`No *.metadata.xml golden in ${dir}`);
+    return path.join(dir, file);
+}
+/** All *.metadata.xml golden filenames for a case, e.g. for a multi-artifact L3/L4 case. */
+function listGoldenArtifacts(caseId) {
+    const dir = goldenDir(caseId);
+    return fs.readdirSync(dir).filter(f => f.endsWith('.metadata.xml')).sort();
+}
+function shortSha() {
+    try {
+        return execSync('git rev-parse --short HEAD', { cwd: REPO_ROOT }).toString().trim();
+    }
+    catch {
+        return 'unknown';
+    }
+}
+/** Flags that consume the following argv element as their value. */
+const VALUE_FLAGS = [
+    '--golden', '--actual-dir', '--bp-warnings', '--systest', '--classification',
+    '--golden-prefix', '--actual-prefix',
+];
+function positionalArgs(argv) {
+    const out = [];
+    for (let i = 0; i < argv.length; i++) {
+        if (argv[i].startsWith('--')) {
+            if (VALUE_FLAGS.includes(argv[i]))
+                i++; // skip its value
+            continue;
+        }
+        out.push(argv[i]);
+    }
+    return out;
+}
+async function main() {
+    const positionals = positionalArgs(process.argv.slice(2));
+    const caseId = positionals[0];
+    const actualDir = arg('--actual-dir');
+    const actualPath = actualDir ? undefined : positionals[1];
+    if (!caseId || (!actualDir && !actualPath)) {
+        console.error('usage: tsx src/eval/oracle/cli.ts <caseId> <actualXml> [--golden p] [--build-failed] [--bp-warnings n] [--write]');
+        console.error('   or: tsx src/eval/oracle/cli.ts <caseId> --actual-dir <dir> [options]   (multi-artifact)');
+        process.exit(2);
+    }
+    const caseSpec = JSON.parse(fs.readFileSync(path.join(REPO_ROOT, 'eval', 'cases', `${caseId}.json`), 'utf8'));
+    const buildSucceeded = !flagSet('--build-failed');
+    const bpCount = Number(arg('--bp-warnings') ?? '0');
+    const build = { succeeded: buildSucceeded, bpWarnings: Array.from({ length: bpCount }, () => ({})) };
+    const systestFile = arg('--systest');
+    const systest = systestFile
+        ? parseSysTestResult(fs.readFileSync(path.resolve(systestFile), 'utf8'))
+        : undefined;
+    // The golden corpus was captured under a fixed prefix (docs/AGENT_EVAL_LOOP.md §6.4); the
+    // actual defaults to THIS process's own EXTENSION_PREFIX — i.e. whatever session ran the
+    // case (the eval-implementer's VM env), not a guess. When EXTENSION_PREFIX isn't set at all
+    // (e.g. a local self-check comparing a golden against itself/another golden-shaped fixture,
+    // no VM session involved) fall back to the golden's own prefix so that unprefixed-env usage
+    // keeps matching exactly as before. Either is overridable for edge cases.
+    const goldenPrefix = arg('--golden-prefix') ?? GOLDEN_CAPTURE_PREFIX;
+    const actualPrefix = arg('--actual-prefix') ?? (resolveRegularObjectPrefixToken() || GOLDEN_CAPTURE_PREFIX);
+    let goldenDiff, score, systestOut, generatedArtifacts, debugLabel;
+    if (actualDir) {
+        const resolvedActualDir = path.resolve(actualDir);
+        const artifactNames = listGoldenArtifacts(caseId);
+        if (artifactNames.length === 0)
+            throw new Error(`No *.metadata.xml goldens in ${goldenDir(caseId)}`);
+        const goldenArtifacts = {};
+        for (const name of artifactNames) {
+            goldenArtifacts[name] = fs.readFileSync(path.join(goldenDir(caseId), name), 'utf8');
+        }
+        const { actualArtifacts, matchedActualFiles } = buildActualArtifactsMap(resolvedActualDir, artifactNames, goldenPrefix, actualPrefix);
+        // Surface extra actual files (produced but not golden-expected, and not already
+        // matched to a golden artifact above under prefix-canonicalised filename matching) too.
+        for (const f of fs.readdirSync(resolvedActualDir).filter(f => f.endsWith('.metadata.xml'))) {
+            if (!matchedActualFiles.has(f) && !(f in actualArtifacts)) {
+                actualArtifacts[f] = fs.readFileSync(path.join(resolvedActualDir, f), 'utf8');
+            }
+        }
+        ({ goldenDiff, score, systest: systestOut } = await evaluateMulti({
+            caseSpec, actualArtifacts, goldenArtifacts, build, systest, goldenPrefix, actualPrefix,
+        }));
+        generatedArtifacts = Object.keys(actualArtifacts);
+        debugLabel = `${artifactNames.length} artifact(s) in ${actualDir}`;
+    }
+    else {
+        const goldenPath = arg('--golden') ?? findGolden(caseId);
+        const goldenXml = fs.readFileSync(goldenPath, 'utf8');
+        const actualXml = fs.readFileSync(path.resolve(actualPath), 'utf8');
+        ({ goldenDiff, score, systest: systestOut } = await evaluate({
+            caseSpec, actualXml, goldenXml, build, systest, goldenPrefix, actualPrefix,
+        }));
+        generatedArtifacts = [path.basename(actualPath)];
+        debugLabel = path.basename(goldenPath);
+        if (flagSet('--debug')) {
+            console.error('\n--- normalized actual ---\n' + renderNormalized(await normalizeAotXml(actualXml, caseSpec.ignore ?? [], actualPrefix)));
+        }
+    }
+    console.error(`# Oracle: ${caseId}  (golden: ${debugLabel})`);
+    console.error(renderDiff(goldenDiff));
+    if (systestFile)
+        console.error(`SysTest: ran=${systestOut.ran} passed=${systestOut.passed} failures=${systestOut.failures.length}`);
+    console.error(`\nScore: ${JSON.stringify(score)}`);
+    if (flagSet('--write')) {
+        const classification = arg('--classification')
+            ?? (score.build === 1 && score.golden_match === 1 ? (score.bp_clean === 1 ? 'PASS' : 'KNOWLEDGE_GAP') : 'TOOL_DEFECT');
+        const ts = new Date().toISOString().replace(/[:.]/g, '').slice(0, 13);
+        const sha = shortSha();
+        const record = {
+            run_id: `${ts}__${caseId}__${sha}`,
+            case_id: caseId,
+            tier: caseSpec.tier,
+            timestamp: new Date().toISOString(),
+            server_git_sha: sha,
+            generated_artifacts: generatedArtifacts,
+            build: { succeeded: build.succeeded, errors: [], bpWarnings: build.bpWarnings },
+            golden_diff: goldenDiff,
+            systest: systestOut,
+            score,
+            classification,
+        };
+        const outDir = path.join(REPO_ROOT, 'eval', 'corpus', 'runs');
+        const outFile = path.join(outDir, `${record.run_id}.json`);
+        fs.writeFileSync(outFile, JSON.stringify(record, null, 2) + '\n');
+        console.error(`\nWrote corpus record: ${path.relative(REPO_ROOT, outFile)}`);
+    }
+    process.exit(goldenDiff.matched && score.build === 1 ? 0 : 1);
+}
+main().catch(e => { console.error(e); process.exit(2); });
+//# sourceMappingURL=cli.js.map

package/dist/eval/oracle/cli.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"cli.js","sourceRoot":"","sources":["../../../src/eval/oracle/cli.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;GA2BG;AAEH,OAAO,KAAK,EAAE,MAAM,IAAI,CAAC;AACzB,OAAO,KAAK,IAAI,MAAM,MAAM,CAAC;AAC7B,OAAO,EAAE,aAAa,EAAE,MAAM,KAAK,CAAC;AACpC,OAAO,EAAE,QAAQ,EAAE,MAAM,eAAe,CAAC;AACzC,OAAO,EACL,QAAQ,EAAE,aAAa,EAAE,UAAU,EAAE,gBAAgB,EAAE,eAAe,EAAE,kBAAkB,EAC1F,qBAAqB,GACtB,MAAM,YAAY,CAAC;AACpB,OAAO,EAAE,+BAA+B,EAAE,MAAM,gCAAgC,CAAC;AACjF,OAAO,EAAE,uBAAuB,EAAE,MAAM,+BAA+B,CAAC;AAExE,MAAM,IAAI,GAAG,IAAI,CAAC,OAAO,CAAC,aAAa,CAAC,OAAO,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC;AAC1D,MAAM,SAAS,GAAG,IAAI,CAAC,OAAO,CAAC,IAAI,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI,CAAC,CAAC;AAEvD,SAAS,GAAG,CAAC,IAAY;IACvB,MAAM,CAAC,GAAG,OAAO,CAAC,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC;IACrC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;AAClD,CAAC;AACD,SAAS,OAAO,CAAC,IAAY;IAC3B,OAAO,OAAO,CAAC,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC;AACrC,CAAC;AAED,SAAS,SAAS,CAAC,MAAc;IAC/B,OAAO,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,MAAM,EAAE,SAAS,EAAE,MAAM,CAAC,CAAC;AACzD,CAAC;AAED,SAAS,UAAU,CAAC,MAAc;IAChC,MAAM,GAAG,GAAG,SAAS,CAAC,MAAM,CAAC,CAAC;IAC9B,MAAM,IAAI,GAAG,EAAE,CAAC,WAAW,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,eAAe,CAAC,CAAC,CAAC;IACxE,IAAI,CAAC,IAAI;QAAE,MAAM,IAAI,KAAK,CAAC,+BAA+B,GAAG,EAAE,CAAC,CAAC;IACjE,OAAO,IAAI,CAAC,IAAI,CAAC,GAAG,EAAE,IAAI,CAAC,CAAC;AAC9B,CAAC;AAED,4FAA4F;AAC5F,SAAS,mBAAmB,CAAC,MAAc;IACzC,MAAM,GAAG,GAAG,SAAS,CAAC,MAAM,CAAC,CAAC;IAC9B,OAAO,EAAE,CAAC,WAAW,CAAC,GAAG,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,eAAe,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;AAC7E,CAAC;AAED,SAAS,QAAQ;IACf,IAAI,CAAC;QAAC,OAAO,QAAQ,CAAC,4BAA4B,EAAE,EAAE,GAAG,EAAE,SAAS,EAAE,CAAC,CAAC,QAAQ,EAAE,CAAC,IAAI,EAAE,CAAC;IAAC,CAAC;IAC5F,MAAM,CAAC;QAAC,OAAO,SAAS,CAAC;IAAC,CAAC;AAC7B,CAAC;AAED,oEAAoE;AACpE,MAAM,WAAW,GAAG;IAClB,UAAU,EAAE,cAAc,EAAE,eAAe,EAAE,WAAW,EAAE,kBAAkB;IAC5E,iBAAiB,EAAE,iBAAiB;CACrC,CAAC;AAEF,SAAS,cAAc,CAAC,IAAc;IACpC,MAAM,GAAG,GAAa,EAAE,CAAC;IACzB,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,IAAI,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QACrC,IAAI,IAAI,CAAC,CAAC,CAAC,CAAC,UAAU,CAAC,IAAI,CAAC,EAAE,CAAC;YAC7B,IAAI,WAAW,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;gBAAE,CAAC,EAAE,CAAC,CAAC,iBAAiB;YACzD,SAAS;QACX,CAAC;QACD,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC;IACpB,CAAC;IACD,OAAO,GAAG,CAAC;AACb,CAAC;AAED,KAAK,UAAU,IAAI;IACjB,MAAM,WAAW,GAAG,cAAc,CAAC,OAAO,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC;IAC1D,MAAM,MAAM,GAAG,WAAW,CAAC,CAAC,CAAC,CAAC;IAC9B,MAAM,SAAS,GAAG,GAAG,CAAC,cAAc,CAAC,CAAC;IACtC,MAAM,UAAU,GAAG,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,CAAC;IAE1D,IAAI,CAAC,MAAM,IAAI,CAAC,CAAC,SAAS,IAAI,CAAC,UAAU,CAAC,EAAE,CAAC;QAC3C,OAAO,CAAC,KAAK,CAAC,kHAAkH,CAAC,CAAC;QAClI,OAAO,CAAC,KAAK,CAAC,4FAA4F,CAAC,CAAC;QAC5G,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IAClB,CAAC;IAED,MAAM,QAAQ,GAAG,IAAI,CAAC,KAAK,CACzB,EAAE,CAAC,YAAY,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,MAAM,EAAE,OAAO,EAAE,GAAG,MAAM,OAAO,CAAC,EAAE,MAAM,CAAC,CAC7C,CAAC;IAEtC,MAAM,cAAc,GAAG,CAAC,OAAO,CAAC,gBAAgB,CAAC,CAAC;IAClD,MAAM,OAAO,GAAG,MAAM,CAAC,GAAG,CAAC,eAAe,CAAC,IAAI,GAAG,CAAC,CAAC;IACpD,MAAM,KAAK,GAAG,EAAE,SAAS,EAAE,cAAc,EAAE,UAAU,EAAE,KAAK,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,OAAO,EAAE,EAAE,GAAG,EAAE,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE,CAAC;IAErG,MAAM,WAAW,GAAG,GAAG,CAAC,WAAW,CAAC,CAAC;IACrC,MAAM,OAAO,GAAG,WAAW;QACzB,CAAC,CAAC,kBAAkB,CAAC,EAAE,CAAC,YAAY,CAAC,IAAI,CAAC,OAAO,CAAC,WAAW,CAAC,EAAE,MAAM,CAAC,CAAC;QACxE,CAAC,CAAC,SAAS,CAAC;IAEd,0FAA0F;IAC1F,yFAAyF;IACzF,4FAA4F;IAC5F,4FAA4F;IAC5F,4FAA4F;IAC5F,0EAA0E;IAC1E,MAAM,YAAY,GAAG,GAAG,CAAC,iBAAiB,CAAC,IAAI,qBAAqB,CAAC;IACrE,MAAM,YAAY,GAAG,GAAG,CAAC,iBAAiB,CAAC,IAAI,CAAC,+BAA+B,EAAE,IAAI,qBAAqB,CAAC,CAAC;IAE5G,IAAI,UAAU,EAAE,KAAK,EAAE,UAAU,EAAE,kBAA4B,EAAE,UAAkB,CAAC;IAEpF,IAAI,SAAS,EAAE,CAAC;QACd,MAAM,iBAAiB,GAAG,IAAI,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC;QAClD,MAAM,aAAa,GAAG,mBAAmB,CAAC,MAAM,CAAC,CAAC;QAClD,IAAI,aAAa,CAAC,MAAM,KAAK,CAAC;YAAE,MAAM,IAAI,KAAK,CAAC,gCAAgC,SAAS,CAAC,MAAM,CAAC,EAAE,CAAC,CAAC;QACrG,MAAM,eAAe,GAA2B,EAAE,CAAC;QACnD,KAAK,MAAM,IAAI,IAAI,aAAa,EAAE,CAAC;YACjC,eAAe,CAAC,IAAI,CAAC,GAAG,EAAE,CAAC,YAAY,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,EAAE,IAAI,CAAC,EAAE,MAAM,CAAC,CAAC;QACtF,CAAC;QACD,MAAM,EAAE,eAAe,EAAE,kBAAkB,EAAE,GAC3C,uBAAuB,CAAC,iBAAiB,EAAE,aAAa,EAAE,YAAY,EAAE,YAAY,CAAC,CAAC;QACxF,gFAAgF;QAChF,wFAAwF;QACxF,KAAK,MAAM,CAAC,IAAI,EAAE,CAAC,WAAW,CAAC,iBAAiB,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,eAAe,CAAC,CAAC,EAAE,CAAC;YAC3F,IAAI,CAAC,kBAAkB,CAAC,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,eAAe,CAAC,EAAE,CAAC;gBAC1D,eAAe,CAAC,CAAC,CAAC,GAAG,EAAE,CAAC,YAAY,CAAC,IAAI,CAAC,IAAI,CAAC,iBAAiB,EAAE,CAAC,CAAC,EAAE,MAAM,CAAC,CAAC;YAChF,CAAC;QACH,CAAC;QACD,CAAC,EAAE,UAAU,EAAE,KAAK,EAAE,OAAO,EAAE,UAAU,EAAE,GAAG,MAAM,aAAa,CAAC;YAChE,QAAQ,EAAE,eAAe,EAAE,eAAe,EAAE,KAAK,EAAE,OAAO,EAAE,YAAY,EAAE,YAAY;SACvF,CAAC,CAAC,CAAC;QACJ,kBAAkB,GAAG,MAAM,CAAC,IAAI,CAAC,eAAe,CAAC,CAAC;QAClD,UAAU,GAAG,GAAG,aAAa,CAAC,MAAM,mBAAmB,SAAS,EAAE,CAAC;IACrE,CAAC;SAAM,CAAC;QACN,MAAM,UAAU,GAAG,GAAG,CAAC,UAAU,CAAC,IAAI,UAAU,CAAC,MAAM,CAAC,CAAC;QACzD,MAAM,SAAS,GAAG,EAAE,CAAC,YAAY,CAAC,UAAU,EAAE,MAAM,CAAC,CAAC;QACtD,MAAM,SAAS,GAAG,EAAE,CAAC,YAAY,CAAC,IAAI,CAAC,OAAO,CAAC,UAAW,CAAC,EAAE,MAAM,CAAC,CAAC;QACrE,CAAC,EAAE,UAAU,EAAE,KAAK,EAAE,OAAO,EAAE,UAAU,EAAE,GAAG,MAAM,QAAQ,CAAC;YAC3D,QAAQ,EAAE,SAAS,EAAE,SAAS,EAAE,KAAK,EAAE,OAAO,EAAE,YAAY,EAAE,YAAY;SAC3E,CAAC,CAAC,CAAC;QACJ,kBAAkB,GAAG,CAAC,IAAI,CAAC,QAAQ,CAAC,UAAW,CAAC,CAAC,CAAC;QAClD,UAAU,GAAG,IAAI,CAAC,QAAQ,CAAC,UAAU,CAAC,CAAC;QACvC,IAAI,OAAO,CAAC,SAAS,CAAC,EAAE,CAAC;YACvB,OAAO,CAAC,KAAK,CAAC,+BAA+B,GAAG,gBAAgB,CAAC,MAAM,eAAe,CAAC,SAAS,EAAE,QAAQ,CAAC,MAAM,IAAI,EAAE,EAAE,YAAY,CAAC,CAAC,CAAC,CAAC;QAC3I,CAAC;IACH,CAAC;IAED,OAAO,CAAC,KAAK,CAAC,aAAa,MAAM,cAAc,UAAU,GAAG,CAAC,CAAC;IAC9D,OAAO,CAAC,KAAK,CAAC,UAAU,CAAC,UAAU,CAAC,CAAC,CAAC;IACtC,IAAI,WAAW;QAAE,OAAO,CAAC,KAAK,CAAC,gBAAgB,UAAU,CAAC,GAAG,WAAW,UAAU,CAAC,MAAM,aAAa,UAAU,CAAC,QAAQ,CAAC,MAAM,EAAE,CAAC,CAAC;IACpI,OAAO,CAAC,KAAK,CAAC,YAAY,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;IAEnD,IAAI,OAAO,CAAC,SAAS,CAAC,EAAE,CAAC;QACvB,MAAM,cAAc,GAAG,GAAG,CAAC,kBAAkB,CAAC;eACzC,CAAC,KAAK,CAAC,KAAK,KAAK,CAAC,IAAI,KAAK,CAAC,YAAY,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,QAAQ,KAAK,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,eAAe,CAAC,CAAC,CAAC,CAAC,aAAa,CAAC,CAAC;QACzH,MAAM,EAAE,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC,OAAO,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;QACtE,MAAM,GAAG,GAAG,QAAQ,EAAE,CAAC;QACvB,MAAM,MAAM,GAAG;YACb,MAAM,EAAE,GAAG,EAAE,KAAK,MAAM,KAAK,GAAG,EAAE;YAClC,OAAO,EAAE,MAAM;YACf,IAAI,EAAE,QAAQ,CAAC,IAAI;YACnB,SAAS,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;YACnC,cAAc,EAAE,GAAG;YACnB,mBAAmB,EAAE,kBAAkB;YACvC,KAAK,EAAE,EAAE,SAAS,EAAE,KAAK,CAAC,SAAS,EAAE,MAAM,EAAE,EAAE,EAAE,UAAU,EAAE,KAAK,CAAC,UAAU,EAAE;YAC/E,WAAW,EAAE,UAAU;YACvB,OAAO,EAAE,UAAU;YACnB,KAAK;YACL,cAAc;SACf,CAAC;QACF,MAAM,MAAM,GAAG,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,MAAM,EAAE,QAAQ,EAAE,MAAM,CAAC,CAAC;QAC9D,MAAM,OAAO,GAAG,IAAI,CAAC,IAAI,CAAC,MAAM,EAAE,GAAG,MAAM,CAAC,MAAM,OAAO,CAAC,CAAC;QAC3D,EAAE,CAAC,aAAa,CAAC,OAAO,EAAE,IAAI,CAAC,SAAS,CAAC,MAAM,EAAE,IAAI,EAAE,CAAC,CAAC,GAAG,IAAI,CAAC,CAAC;QAClE,OAAO,CAAC,KAAK,CAAC,0BAA0B,IAAI,CAAC,QAAQ,CAAC,SAAS,EAAE,OAAO,CAAC,EAAE,CAAC,CAAC;IAC/E,CAAC;IAED,OAAO,CAAC,IAAI,CAAC,UAAU,CAAC,OAAO,IAAI,KAAK,CAAC,KAAK,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;AAChE,CAAC;AAED,IAAI,EAAE,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,GAAG,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC"}

package/dist/eval/oracle/diff.d.ts

@@ -0,0 +1,26 @@
+/**
+ * Structural diff for the eval golden oracle. Compares two normalized
+ * `path → value` maps (see normalize.ts) and classifies each delta as
+ * missing / extra / changed, mirroring eval/corpus/schema.json `golden_diff`.
+ */
+export interface GoldenDiff {
+    matched: boolean;
+    /** Paths present in the golden but absent from the actual. */
+    missing: string[];
+    /** Paths present in the actual but absent from the golden. */
+    extra: string[];
+    /** Paths present in both with differing values. */
+    changed: Array<{
+        path: string;
+        expected: string;
+        actual: string;
+    }>;
+}
+/**
+ * Diff a normalized actual against a normalized golden (expected).
+ * `expected` = the golden; `actual` = what the run produced.
+ */
+export declare function diffNormalized(expected: Map<string, string>, actual: Map<string, string>): GoldenDiff;
+/** Render a GoldenDiff as a short human-readable report. */
+export declare function renderDiff(d: GoldenDiff): string;
+//# sourceMappingURL=diff.d.ts.map

package/dist/eval/oracle/diff.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"diff.d.ts","sourceRoot":"","sources":["../../../src/eval/oracle/diff.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,MAAM,WAAW,UAAU;IACzB,OAAO,EAAE,OAAO,CAAC;IACjB,8DAA8D;IAC9D,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,8DAA8D;IAC9D,KAAK,EAAE,MAAM,EAAE,CAAC;IAChB,mDAAmD;IACnD,OAAO,EAAE,KAAK,CAAC;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,QAAQ,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;CACpE;AAED;;;GAGG;AACH,wBAAgB,cAAc,CAC5B,QAAQ,EAAE,GAAG,CAAC,MAAM,EAAE,MAAM,CAAC,EAC7B,MAAM,EAAE,GAAG,CAAC,MAAM,EAAE,MAAM,CAAC,GAC1B,UAAU,CA2BZ;AAED,4DAA4D;AAC5D,wBAAgB,UAAU,CAAC,CAAC,EAAE,UAAU,GAAG,MAAM,CAOhD"}

package/dist/eval/oracle/diff.js

@@ -0,0 +1,51 @@
+/**
+ * Structural diff for the eval golden oracle. Compares two normalized
+ * `path → value` maps (see normalize.ts) and classifies each delta as
+ * missing / extra / changed, mirroring eval/corpus/schema.json `golden_diff`.
+ */
+/**
+ * Diff a normalized actual against a normalized golden (expected).
+ * `expected` = the golden; `actual` = what the run produced.
+ */
+export function diffNormalized(expected, actual) {
+    const missing = [];
+    const changed = [];
+    const extra = [];
+    for (const [path, expVal] of expected) {
+        if (!actual.has(path)) {
+            missing.push(path);
+        }
+        else {
+            const actVal = actual.get(path);
+            if (actVal !== expVal)
+                changed.push({ path, expected: expVal, actual: actVal });
+        }
+    }
+    for (const path of actual.keys()) {
+        if (!expected.has(path))
+            extra.push(path);
+    }
+    missing.sort();
+    extra.sort();
+    changed.sort((a, b) => a.path.localeCompare(b.path));
+    return {
+        matched: missing.length === 0 && extra.length === 0 && changed.length === 0,
+        missing,
+        extra,
+        changed,
+    };
+}
+/** Render a GoldenDiff as a short human-readable report. */
+export function renderDiff(d) {
+    if (d.matched)
+        return '✅ golden match — no structural deltas.';
+    const lines = [`❌ golden mismatch — ${d.missing.length} missing, ${d.extra.length} extra, ${d.changed.length} changed.`];
+    for (const p of d.missing)
+        lines.push(`  − missing: ${p}`);
+    for (const p of d.extra)
+        lines.push(`  + extra:   ${p}`);
+    for (const c of d.changed)
+        lines.push(`  ~ changed: ${c.path}  (golden="${c.expected}" actual="${c.actual}")`);
+    return lines.join('\n');
+}
+//# sourceMappingURL=diff.js.map

package/dist/eval/oracle/diff.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"diff.js","sourceRoot":"","sources":["../../../src/eval/oracle/diff.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAYH;;;GAGG;AACH,MAAM,UAAU,cAAc,CAC5B,QAA6B,EAC7B,MAA2B;IAE3B,MAAM,OAAO,GAAa,EAAE,CAAC;IAC7B,MAAM,OAAO,GAA0B,EAAE,CAAC;IAC1C,MAAM,KAAK,GAAa,EAAE,CAAC;IAE3B,KAAK,MAAM,CAAC,IAAI,EAAE,MAAM,CAAC,IAAI,QAAQ,EAAE,CAAC;QACtC,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC;YACtB,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QACrB,CAAC;aAAM,CAAC;YACN,MAAM,MAAM,GAAG,MAAM,CAAC,GAAG,CAAC,IAAI,CAAE,CAAC;YACjC,IAAI,MAAM,KAAK,MAAM;gBAAE,OAAO,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,QAAQ,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,EAAE,CAAC,CAAC;QAClF,CAAC;IACH,CAAC;IACD,KAAK,MAAM,IAAI,IAAI,MAAM,CAAC,IAAI,EAAE,EAAE,CAAC;QACjC,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,IAAI,CAAC;YAAE,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAC5C,CAAC;IAED,OAAO,CAAC,IAAI,EAAE,CAAC;IACf,KAAK,CAAC,IAAI,EAAE,CAAC;IACb,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC;IAErD,OAAO;QACL,OAAO,EAAE,OAAO,CAAC,MAAM,KAAK,CAAC,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC;QAC3E,OAAO;QACP,KAAK;QACL,OAAO;KACR,CAAC;AACJ,CAAC;AAED,4DAA4D;AAC5D,MAAM,UAAU,UAAU,CAAC,CAAa;IACtC,IAAI,CAAC,CAAC,OAAO;QAAE,OAAO,wCAAwC,CAAC;IAC/D,MAAM,KAAK,GAAa,CAAC,uBAAuB,CAAC,CAAC,OAAO,CAAC,MAAM,aAAa,CAAC,CAAC,KAAK,CAAC,MAAM,WAAW,CAAC,CAAC,OAAO,CAAC,MAAM,WAAW,CAAC,CAAC;IACnI,KAAK,MAAM,CAAC,IAAI,CAAC,CAAC,OAAO;QAAE,KAAK,CAAC,IAAI,CAAC,gBAAgB,CAAC,EAAE,CAAC,CAAC;IAC3D,KAAK,MAAM,CAAC,IAAI,CAAC,CAAC,KAAK;QAAE,KAAK,CAAC,IAAI,CAAC,gBAAgB,CAAC,EAAE,CAAC,CAAC;IACzD,KAAK,MAAM,CAAC,IAAI,CAAC,CAAC,OAAO;QAAE,KAAK,CAAC,IAAI,CAAC,gBAAgB,CAAC,CAAC,IAAI,cAAc,CAAC,CAAC,QAAQ,aAAa,CAAC,CAAC,MAAM,IAAI,CAAC,CAAC;IAC/G,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAC1B,CAAC"}

package/dist/eval/oracle/index.d.ts

@@ -0,0 +1,81 @@
+/**
+ * Eval golden oracle — orchestrator.
+ *
+ * Given a case spec, the actual produced XML, the golden XML, and the build
+ * result, compute the structural golden diff and the scorecard. VM-free: this is
+ * the piece the improver/CI runs without the D365FO platform.
+ */
+import { type GoldenDiff } from './diff.js';
+import { type BuildResult, type Score } from './score.js';
+import { type SysTestResult } from './systest.js';
+export { normalizeAotXml, normalizeMultiArtifact, renderNormalized, globToRegExp, GOLDEN_CAPTURE_PREFIX, canonicalizePrefix, } from './normalize.js';
+export { diffNormalized, renderDiff, type GoldenDiff } from './diff.js';
+export { scoreRun, type Score, type BuildResult, type ScoreInput } from './score.js';
+export { parseSysTestResult, type SysTestResult, type SysTestFailure } from './systest.js';
+export interface CaseSpec {
+    id: string;
+    tier: number;
+    ignore?: string[];
+}
+export interface EvaluateInput {
+    caseSpec: CaseSpec;
+    actualXml: string;
+    goldenXml: string;
+    build: BuildResult;
+    /** Runtime-oracle result (e.g. from parseSysTestResult); omit when the case has no SysTest. */
+    systest?: SysTestResult | {
+        passed: boolean | null;
+    } | null;
+    /**
+     * EXTENSION_PREFIX the golden was captured under, to canonicalise away
+     * before comparing (docs/AGENT_EVAL_LOOP.md §6.2). Pass
+     * `GOLDEN_CAPTURE_PREFIX` for the committed `eval/goldens/` corpus. Defaults
+     * to '' (no canonicalisation — legacy literal-string comparison), so
+     * existing callers that don't pass one are unaffected.
+     */
+    goldenPrefix?: string;
+    /**
+     * EXTENSION_PREFIX the actual artifact was produced under (the CURRENT
+     * session's configured prefix — see
+     * `resolveRegularObjectPrefixToken()` in src/utils/modelClassifier.ts).
+     * Defaults to '' (no canonicalisation), so callers that don't know/care
+     * about prefix drift keep the legacy literal-string comparison.
+     */
+    actualPrefix?: string;
+}
+export interface EvaluateResult {
+    goldenDiff: GoldenDiff;
+    score: Score;
+    /** Echoed runtime result for the corpus record, when a SysTest was supplied. */
+    systest: SysTestResult | {
+        ran: false;
+        passed: null;
+        failures: [];
+    };
+}
+export declare function evaluate(input: EvaluateInput): Promise<EvaluateResult>;
+export interface EvaluateMultiInput {
+    caseSpec: CaseSpec;
+    /** filename → produced XML, e.g. { "MyContract.metadata.xml": "<AxClass>..." } */
+    actualArtifacts: Record<string, string>;
+    /** filename → golden XML, same keys expected (a missing/extra key shows up in the diff). */
+    goldenArtifacts: Record<string, string>;
+    build: BuildResult;
+    systest?: SysTestResult | {
+        passed: boolean | null;
+    } | null;
+    /** See EvaluateInput.goldenPrefix. */
+    goldenPrefix?: string;
+    /** See EvaluateInput.actualPrefix. */
+    actualPrefix?: string;
+}
+/**
+ * Multi-artifact variant of `evaluate` for L3/L4 cases that produce several
+ * objects (e.g. a SysOperation's Contract + DP + Controller, or a data entity +
+ * its security chain). Each artifact's normalized paths are prefixed with
+ * `<filename>::` and merged into one combined map, then diffed/scored with the
+ * same single-document machinery — a wholly missing or extra artifact shows up
+ * as every one of its paths being missing/extra under that prefix.
+ */
+export declare function evaluateMulti(input: EvaluateMultiInput): Promise<EvaluateResult>;
+//# sourceMappingURL=index.d.ts.map

package/dist/eval/oracle/index.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/eval/oracle/index.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AAGH,OAAO,EAAkB,KAAK,UAAU,EAAE,MAAM,WAAW,CAAC;AAC5D,OAAO,EAAY,KAAK,WAAW,EAAE,KAAK,KAAK,EAAE,MAAM,YAAY,CAAC;AACpE,OAAO,EAAE,KAAK,aAAa,EAAE,MAAM,cAAc,CAAC;AAElD,OAAO,EACL,eAAe,EAAE,sBAAsB,EAAE,gBAAgB,EAAE,YAAY,EACvE,qBAAqB,EAAE,kBAAkB,GAC1C,MAAM,gBAAgB,CAAC;AACxB,OAAO,EAAE,cAAc,EAAE,UAAU,EAAE,KAAK,UAAU,EAAE,MAAM,WAAW,CAAC;AACxE,OAAO,EAAE,QAAQ,EAAE,KAAK,KAAK,EAAE,KAAK,WAAW,EAAE,KAAK,UAAU,EAAE,MAAM,YAAY,CAAC;AACrF,OAAO,EAAE,kBAAkB,EAAE,KAAK,aAAa,EAAE,KAAK,cAAc,EAAE,MAAM,cAAc,CAAC;AAE3F,MAAM,WAAW,QAAQ;IACvB,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,MAAM,CAAC;IACb,MAAM,CAAC,EAAE,MAAM,EAAE,CAAC;CACnB;AAED,MAAM,WAAW,aAAa;IAC5B,QAAQ,EAAE,QAAQ,CAAC;IACnB,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;IAClB,KAAK,EAAE,WAAW,CAAC;IACnB,+FAA+F;IAC/F,OAAO,CAAC,EAAE,aAAa,GAAG;QAAE,MAAM,EAAE,OAAO,GAAG,IAAI,CAAA;KAAE,GAAG,IAAI,CAAC;IAC5D;;;;;;OAMG;IACH,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB;;;;;;OAMG;IACH,YAAY,CAAC,EAAE,MAAM,CAAC;CACvB;AAED,MAAM,WAAW,cAAc;IAC7B,UAAU,EAAE,UAAU,CAAC;IACvB,KAAK,EAAE,KAAK,CAAC;IACb,gFAAgF;IAChF,OAAO,EAAE,aAAa,GAAG;QAAE,GAAG,EAAE,KAAK,CAAC;QAAC,MAAM,EAAE,IAAI,CAAC;QAAC,QAAQ,EAAE,EAAE,CAAA;KAAE,CAAC;CACrE;AAED,wBAAsB,QAAQ,CAAC,KAAK,EAAE,aAAa,GAAG,OAAO,CAAC,cAAc,CAAC,CAe5E;AAED,MAAM,WAAW,kBAAkB;IACjC,QAAQ,EAAE,QAAQ,CAAC;IACnB,kFAAkF;IAClF,eAAe,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACxC,4FAA4F;IAC5F,eAAe,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACxC,KAAK,EAAE,WAAW,CAAC;IACnB,OAAO,CAAC,EAAE,aAAa,GAAG;QAAE,MAAM,EAAE,OAAO,GAAG,IAAI,CAAA;KAAE,GAAG,IAAI,CAAC;IAC5D,sCAAsC;IACtC,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,sCAAsC;IACtC,YAAY,CAAC,EAAE,MAAM,CAAC;CACvB;AAED;;;;;;;GAOG;AACH,wBAAsB,aAAa,CAAC,KAAK,EAAE,kBAAkB,GAAG,OAAO,CAAC,cAAc,CAAC,CAetF"}

package/dist/eval/oracle/index.js

@@ -0,0 +1,55 @@
+/**
+ * Eval golden oracle — orchestrator.
+ *
+ * Given a case spec, the actual produced XML, the golden XML, and the build
+ * result, compute the structural golden diff and the scorecard. VM-free: this is
+ * the piece the improver/CI runs without the D365FO platform.
+ */
+import { normalizeAotXml, normalizeMultiArtifact } from './normalize.js';
+import { diffNormalized } from './diff.js';
+import { scoreRun } from './score.js';
+export { normalizeAotXml, normalizeMultiArtifact, renderNormalized, globToRegExp, GOLDEN_CAPTURE_PREFIX, canonicalizePrefix, } from './normalize.js';
+export { diffNormalized, renderDiff } from './diff.js';
+export { scoreRun } from './score.js';
+export { parseSysTestResult } from './systest.js';
+export async function evaluate(input) {
+    const { caseSpec, actualXml, goldenXml, build, systest } = input;
+    const goldenPrefix = input.goldenPrefix ?? '';
+    const actualPrefix = input.actualPrefix ?? '';
+    const ignore = caseSpec.ignore ?? [];
+    const [expected, actual] = await Promise.all([
+        normalizeAotXml(goldenXml, ignore, goldenPrefix),
+        normalizeAotXml(actualXml, ignore, actualPrefix),
+    ]);
+    const goldenDiff = diffNormalized(expected, actual);
+    const score = scoreRun({ build, goldenDiff, tier: caseSpec.tier, systest });
+    const systestOut = systest && 'ran' in systest
+        ? systest
+        : { ran: false, passed: null, failures: [] };
+    return { goldenDiff, score, systest: systestOut };
+}
+/**
+ * Multi-artifact variant of `evaluate` for L3/L4 cases that produce several
+ * objects (e.g. a SysOperation's Contract + DP + Controller, or a data entity +
+ * its security chain). Each artifact's normalized paths are prefixed with
+ * `<filename>::` and merged into one combined map, then diffed/scored with the
+ * same single-document machinery — a wholly missing or extra artifact shows up
+ * as every one of its paths being missing/extra under that prefix.
+ */
+export async function evaluateMulti(input) {
+    const { caseSpec, actualArtifacts, goldenArtifacts, build, systest } = input;
+    const goldenPrefix = input.goldenPrefix ?? '';
+    const actualPrefix = input.actualPrefix ?? '';
+    const ignore = caseSpec.ignore ?? [];
+    const [expected, actual] = await Promise.all([
+        normalizeMultiArtifact(goldenArtifacts, ignore, goldenPrefix),
+        normalizeMultiArtifact(actualArtifacts, ignore, actualPrefix),
+    ]);
+    const goldenDiff = diffNormalized(expected, actual);
+    const score = scoreRun({ build, goldenDiff, tier: caseSpec.tier, systest });
+    const systestOut = systest && 'ran' in systest
+        ? systest
+        : { ran: false, passed: null, failures: [] };
+    return { goldenDiff, score, systest: systestOut };
+}
+//# sourceMappingURL=index.js.map

package/dist/eval/oracle/index.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../../../src/eval/oracle/index.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AAEH,OAAO,EAAE,eAAe,EAAE,sBAAsB,EAAE,MAAM,gBAAgB,CAAC;AACzE,OAAO,EAAE,cAAc,EAAmB,MAAM,WAAW,CAAC;AAC5D,OAAO,EAAE,QAAQ,EAAgC,MAAM,YAAY,CAAC;AAGpE,OAAO,EACL,eAAe,EAAE,sBAAsB,EAAE,gBAAgB,EAAE,YAAY,EACvE,qBAAqB,EAAE,kBAAkB,GAC1C,MAAM,gBAAgB,CAAC;AACxB,OAAO,EAAE,cAAc,EAAE,UAAU,EAAmB,MAAM,WAAW,CAAC;AACxE,OAAO,EAAE,QAAQ,EAAiD,MAAM,YAAY,CAAC;AACrF,OAAO,EAAE,kBAAkB,EAA2C,MAAM,cAAc,CAAC;AAwC3F,MAAM,CAAC,KAAK,UAAU,QAAQ,CAAC,KAAoB;IACjD,MAAM,EAAE,QAAQ,EAAE,SAAS,EAAE,SAAS,EAAE,KAAK,EAAE,OAAO,EAAE,GAAG,KAAK,CAAC;IACjE,MAAM,YAAY,GAAG,KAAK,CAAC,YAAY,IAAI,EAAE,CAAC;IAC9C,MAAM,YAAY,GAAG,KAAK,CAAC,YAAY,IAAI,EAAE,CAAC;IAC9C,MAAM,MAAM,GAAG,QAAQ,CAAC,MAAM,IAAI,EAAE,CAAC;IACrC,MAAM,CAAC,QAAQ,EAAE,MAAM,CAAC,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC;QAC3C,eAAe,CAAC,SAAS,EAAE,MAAM,EAAE,YAAY,CAAC;QAChD,eAAe,CAAC,SAAS,EAAE,MAAM,EAAE,YAAY,CAAC;KACjD,CAAC,CAAC;IACH,MAAM,UAAU,GAAG,cAAc,CAAC,QAAQ,EAAE,MAAM,CAAC,CAAC;IACpD,MAAM,KAAK,GAAG,QAAQ,CAAC,EAAE,KAAK,EAAE,UAAU,EAAE,IAAI,EAAE,QAAQ,CAAC,IAAI,EAAE,OAAO,EAAE,CAAC,CAAC;IAC5E,MAAM,UAAU,GAAG,OAAO,IAAI,KAAK,IAAI,OAAO;QAC5C,CAAC,CAAC,OAAO;QACT,CAAC,CAAC,EAAE,GAAG,EAAE,KAAc,EAAE,MAAM,EAAE,IAAI,EAAE,QAAQ,EAAE,EAAQ,EAAE,CAAC;IAC9D,OAAO,EAAE,UAAU,EAAE,KAAK,EAAE,OAAO,EAAE,UAAU,EAAE,CAAC;AACpD,CAAC;AAgBD;;;;;;;GAOG;AACH,MAAM,CAAC,KAAK,UAAU,aAAa,CAAC,KAAyB;IAC3D,MAAM,EAAE,QAAQ,EAAE,eAAe,EAAE,eAAe,EAAE,KAAK,EAAE,OAAO,EAAE,GAAG,KAAK,CAAC;IAC7E,MAAM,YAAY,GAAG,KAAK,CAAC,YAAY,IAAI,EAAE,CAAC;IAC9C,MAAM,YAAY,GAAG,KAAK,CAAC,YAAY,IAAI,EAAE,CAAC;IAC9C,MAAM,MAAM,GAAG,QAAQ,CAAC,MAAM,IAAI,EAAE,CAAC;IACrC,MAAM,CAAC,QAAQ,EAAE,MAAM,CAAC,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC;QAC3C,sBAAsB,CAAC,eAAe,EAAE,MAAM,EAAE,YAAY,CAAC;QAC7D,sBAAsB,CAAC,eAAe,EAAE,MAAM,EAAE,YAAY,CAAC;KAC9D,CAAC,CAAC;IACH,MAAM,UAAU,GAAG,cAAc,CAAC,QAAQ,EAAE,MAAM,CAAC,CAAC;IACpD,MAAM,KAAK,GAAG,QAAQ,CAAC,EAAE,KAAK,EAAE,UAAU,EAAE,IAAI,EAAE,QAAQ,CAAC,IAAI,EAAE,OAAO,EAAE,CAAC,CAAC;IAC5E,MAAM,UAAU,GAAG,OAAO,IAAI,KAAK,IAAI,OAAO;QAC5C,CAAC,CAAC,OAAO;QACT,CAAC,CAAC,EAAE,GAAG,EAAE,KAAc,EAAE,MAAM,EAAE,IAAI,EAAE,QAAQ,EAAE,EAAQ,EAAE,CAAC;IAC9D,OAAO,EAAE,UAAU,EAAE,KAAK,EAAE,OAAO,EAAE,UAAU,EAAE,CAAC;AACpD,CAAC"}

package/dist/eval/oracle/normalize.d.ts

@@ -0,0 +1,85 @@
+/**
+ * AOT XML normalizer for the eval golden oracle.
+ *
+ * Flattens an AOT metadata XML document into a stable `path → value` map so two
+ * documents can be diffed structurally (not textually). Key properties:
+ *
+ *  - Collection members (AxTableField, AxEnumValue, AxTableFieldGroup, …) are
+ *    keyed by their `<Name>`/`<DataField>` rather than position, so reordering an
+ *    unordered collection does not register as a diff.
+ *  - Volatile nodes (`ModelSaveInfo`, `@Id`) and any per-case `ignore` globs are
+ *    stripped before the map is built.
+ *  - Attributes are included (notably `i:type`, which carries a field's base type
+ *    e.g. AxTableFieldInt vs AxTableFieldString) under `@type` etc.
+ *
+ * See docs/AGENT_EVAL_LOOP.md §6.2.
+ */
+/**
+ * The EXTENSION_PREFIX in effect when the committed `eval/goldens/` corpus was
+ * authored (docs/AGENT_EVAL_LOOP.md §6.4). A golden's root object Name (and any
+ * other prefixed identifier baked into it, e.g. an extension's added field
+ * DataField, or a dot-notation extension suffix) is a literal string captured
+ * under THIS prefix — it is not re-derived per run.
+ *
+ * A later eval run is free to configure ANY EXTENSION_PREFIX for its sandbox
+ * session (e.g. "Demo") — `d365fo_file`/`generate_object` correctly apply the
+ * CURRENT session's prefix to every new object per their documented contract
+ * (src/utils/modelClassifier.ts). An object named "DemoXyzNoteSubject" and a
+ * golden named "ContosoXyzNoteSubject" describe the SAME object under two
+ * different prefix sessions — that is not a semantic difference, and must not
+ * fail `golden_match` (see the corpus record that surfaced this:
+ * eval/corpus/runs/2026-07-06T10__L0-edt-basic__4fafcd8.json).
+ */
+export declare const GOLDEN_CAPTURE_PREFIX = "Contoso";
+/**
+ * Canonicalise occurrences of `prefix` — the model-naming EXTENSION_PREFIX in
+ * effect for THIS document (the golden's fixed capture-time prefix, or the
+ * actual's current session-configured prefix) — into a stable placeholder, so
+ * a value/key built under one prefix session compares equal to the same
+ * value/key built under a different one.
+ *
+ * Matches are anchored at an identifier-start boundary (string start, or
+ * immediately after a non-alphanumeric character — `.`, `(`, `,`, `_`,
+ * whitespace, …) AND require the prefix to be immediately followed by an
+ * uppercase letter (the PascalCase continuation of the object's own name,
+ * e.g. `ContosoXyzNoteSubject`, `CustGroup.ContosoExtension`, `classStr(ContosoXyzNoteSubject)`).
+ * This keeps the substitution narrow: an incidental occurrence of the prefix
+ * text inside unrelated free-form content (e.g. a label) is left alone.
+ */
+export declare function canonicalizePrefix(value: string, prefix: string): string;
+/** Compile a path glob (`**`, `*`, literal) to an anchored RegExp. */
+export declare function globToRegExp(glob: string): RegExp;
+/**
+ * Normalise an AOT XML document into a sorted `path → value` map. `ignore` is the
+ * per-case glob list from the case spec; built-in ignores are always applied.
+ *
+ * `prefix` is the EXTENSION_PREFIX in effect for THIS document — pass
+ * `GOLDEN_CAPTURE_PREFIX` when normalising a committed golden, and the
+ * CURRENT session's configured prefix (e.g. via
+ * `resolveRegularObjectPrefixToken()` from src/utils/modelClassifier.ts) when
+ * normalising an actual produced artifact — so a prefixed identifier
+ * canonicalises to the same placeholder on both sides regardless of which
+ * EXTENSION_PREFIX session produced it (see `canonicalizePrefix` above).
+ * Defaults to '' (no canonicalisation — legacy literal-string comparison)
+ * for callers that don't pass one.
+ */
+export declare function normalizeAotXml(xml: string, ignore?: string[], prefix?: string): Promise<Map<string, string>>;
+/** Render a normalized map as a stable, human-readable string (for snapshots/logs). */
+export declare function renderNormalized(map: Map<string, string>): string;
+/**
+ * Normalise a SET of artifacts (L3/L4 cases that produce several objects, e.g.
+ * a SysOperation's Contract + DP + Controller) into ONE combined `path → value`
+ * map, each artifact's paths prefixed with `<filename>::`. This lets the
+ * existing single-document diff/score machinery (diffNormalized, scoreRun)
+ * handle multi-artifact cases unchanged: an entirely missing/extra artifact
+ * just shows up as every one of its paths being missing/extra, prefixed.
+ *
+ * `artifacts` keys are filenames (e.g. "MyContract.metadata.xml") — stable,
+ * case-author-chosen identifiers, not full paths. The filename itself is
+ * typically the prefixed object name (e.g. "ContosoMyContract.metadata.xml"), so
+ * it is canonicalised with `prefix` too (see `normalizeAotXml`) — otherwise a
+ * golden captured under one prefix and an actual produced under another would
+ * combine under different `<filename>::` keys and false-mismatch wholesale.
+ */
+export declare function normalizeMultiArtifact(artifacts: Record<string, string>, ignore?: string[], prefix?: string): Promise<Map<string, string>>;
+//# sourceMappingURL=normalize.d.ts.map

package/dist/eval/oracle/normalize.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"normalize.d.ts","sourceRoot":"","sources":["../../../src/eval/oracle/normalize.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;GAeG;AAYH;;;;;;;;;;;;;;;GAeG;AACH,eAAO,MAAM,qBAAqB,YAAY,CAAC;AAU/C;;;;;;;;;;;;;;GAcG;AACH,wBAAgB,kBAAkB,CAAC,KAAK,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,GAAG,MAAM,CAIxE;AAED,sEAAsE;AACtE,wBAAgB,YAAY,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,CAmBjD;AA+KD;;;;;;;;;;;;;GAaG;AACH,wBAAsB,eAAe,CACnC,GAAG,EAAE,MAAM,EACX,MAAM,GAAE,MAAM,EAAO,EACrB,MAAM,SAAK,GACV,OAAO,CAAC,GAAG,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC,CAkB9B;AAED,uFAAuF;AACvF,wBAAgB,gBAAgB,CAAC,GAAG,EAAE,GAAG,CAAC,MAAM,EAAE,MAAM,CAAC,GAAG,MAAM,CAEjE;AAED;;;;;;;;;;;;;;GAcG;AACH,wBAAsB,sBAAsB,CAC1C,SAAS,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,EACjC,MAAM,GAAE,MAAM,EAAO,EACrB,MAAM,SAAK,GACV,OAAO,CAAC,GAAG,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC,CAQ9B"}