@dailephd/my-dev-kit-lab 0.7.0 → 0.7.2

package/dist/src/report/experiments/renderPluginExperimentReportHtml.js

@@ -3,6 +3,7 @@ import { renderWarmIndexReuseHtml } from "./renderWarmIndexReuseHtml.js";
 import { renderIncrementalChangeStalenessHtml } from "./renderIncrementalChangeStalenessHtml.js";
 import { renderContextWindowScalingHtml } from "./renderContextWindowScalingHtml.js";
 export function renderPluginExperimentReportHtml(report) {
+    const redactedTarget = report.target.privacyProjection === "external-local-redacted";
     return `<!DOCTYPE html>
 <html lang="en">
 <head>
@@ -49,10 +50,10 @@ export function renderPluginExperimentReportHtml(report) {
         ["Mode", report.target.mode],
         ["Tool root", report.target.toolRoot],
         ["Target root", report.target.targetRoot],
-        ["Target package", report.target.packageName ? `${report.target.packageName}@${report.target.packageVersion ?? "unknown"}` : "unavailable"],
+        ["Target package", redactedTarget ? "redacted" : report.target.packageName ? `${report.target.packageName}@${report.target.packageVersion ?? "unknown"}` : "unavailable"],
         ["Target branch/commit", report.target.hasGit ? `${report.target.branch ?? "unknown"} / ${report.target.commit ?? "unknown"}` : "unavailable"],
-        ["Has package.json", String(report.target.hasPackageJson)],
-        ["Has lockfile", String(report.target.hasLockfile)],
+        ["Has package.json", redactedTarget ? "redacted" : String(report.target.hasPackageJson)],
+        ["Has lockfile", redactedTarget ? "redacted" : String(report.target.hasLockfile)],
         ["Output root", report.metadata.outputRoot ?? "unavailable"],
     ])}
   </section>

package/dist/src/report/experiments/writePluginExperimentReports.js

@@ -17,13 +17,19 @@ export async function writePluginExperimentReports(args) {
         textPath: resolveWithinRoot(outDir, "report.txt"),
     };
     await mkdir(outputPaths.outDir, { recursive: true });
-    const report = buildPluginExperimentReport({
+    const builtReport = buildPluginExperimentReport({
         run: args.run,
         plugin: args.plugin,
         outputRoot: outputPaths.outDir,
         generatedAt: args.generatedAt,
     });
-    await writeFile(outputPaths.jsonPath, `${JSON.stringify({ report, outputPaths }, null, 2)}\n`, "utf8");
+    const report = args.redactOutputRoot
+        ? { ...builtReport, metadata: { ...builtReport.metadata, outputRoot: "[redacted]" } }
+        : builtReport;
+    const serializedPaths = args.redactOutputRoot
+        ? { outDir: "[redacted]", jsonPath: "report.json", htmlPath: "report.html", textPath: "report.txt" }
+        : outputPaths;
+    await writeFile(outputPaths.jsonPath, `${JSON.stringify({ report, outputPaths: serializedPaths }, null, 2)}\n`, "utf8");
     await writeFile(outputPaths.htmlPath, renderPluginExperimentReportHtml(report), "utf8");
     await writeFile(outputPaths.textPath, renderPluginExperimentReportText(report), "utf8");
     return { report, outputPaths };

package/docs/METRICS.md

@@ -598,7 +598,7 @@ All six metrics are warm-side only (they are not on the raw-full-file side, beca
 
 ## Context-window scaling evidence (v0.7.0)
 
-Status: **released in v0.7.0**. The `context-window-scaling` plugin compares `raw-full-file` and `my-dev-kit-guided` treatment contexts against selected estimated-token budgets. It runs the bundled case catalog against the fixed self target.
+Status: **released in v0.7.0**. The `context-window-scaling` plugin compares `raw-full-file` and `my-dev-kit-guided` treatment contexts against selected estimated-token budgets. By default it runs the bundled case catalog against the fixed self target; the v0.7.1 `--synthetic-config` input runs the same metrics over deterministic generated repositories without changing any formula below.
 
 ### Measurement and budget-cell semantics
 
@@ -620,6 +620,10 @@ The generic `ExperimentMetric` projection emits run-level case and budget counts
 
 The report artifact schema is `my-dev-kit-lab-context-window-scaling-report-v1`. It includes run summaries, budget-treatment summaries, per-treatment context measurements, budget cells, relevant-file summary, and fixed interpretation limits. The execution artifact is `context-window-scaling-execution.json` (`my-dev-kit-lab-context-window-scaling-execution-v1`) and does not embed context text.
 
+**Synthetic-repository runs (v0.7.1).** The synthetic input introduces no new scientific metric: fit, utilization, correctness, success, and omitted-relevant-file semantics, the execution artifact schema, the report schema, and the plot set are identical to the bundled-corpus run. Each generated repository's `synthetic-repository-manifest.json` records generation evidence (per-file byte length, character count, SHA-256, generated-content estimated tokens, aggregate totals, and `repositoryContentIdentity`); these are reproducibility measurements, not context-window outcome metrics. The raw-full-file treatment context is measured by the existing baseline, which adds file headers and separators, so its estimated tokens differ from the manifest's generated-content estimate. Retrieval precision/recall remains v0.8.0 scope.
+
+**Local-repository runs (v0.7.2).** Running over an explicitly selected local Git repository introduces no new scientific metric: fit, utilization, correctness, success, and omitted-relevant-file semantics, the execution artifact and report schemas, and the plot set are the same. In this mode file identity lists are replaced by numbered placeholders (`<redacted file N>`) with the same length as the real list, and each treatment carries `fileIdentityRedaction: "redacted"`; a placeholder is a withheld identity, not an empty result, and an empty list or `null` keeps its usual meaning. Operational compatibility with a repository, the 1 MiB file-size limit, and the Git-ignore exclusions are safety policy, not quality measures: they do not imply retrieval precision, recall, ranking quality, or a retrieval winner.
+
 There is no composite score, winner, or ranking. Budgets are experiment thresholds, not assertions about actual provider context-window capacities. See [COMMANDS.md](COMMANDS.md), [WORKFLOWS.md](WORKFLOWS.md), and [ARCHITECTURE.md](ARCHITECTURE.md).
 
 ## Incremental-change and staleness evidence (v0.6.2)

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@dailephd/my-dev-kit-lab",
-  "version": "0.7.0",
+  "version": "0.7.2",
   "type": "module",
   "description": "Evidence, benchmark, and evaluation companion for my-dev-kit.",
   "bin": {