@decocms/parity 0.11.13 → 0.11.16

package/CHANGELOG.md

@@ -5,6 +5,42 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/).
 
+## [0.11.16](https://github.com/decocms/parity/compare/v0.11.15...v0.11.16) (2026-06-17)
+
+### Changed
+
+* **All checks now run in parallel.** `runAllChecks` previously walked the ~27 checks sequentially with `for (const check of ALL_CHECKS) await check(ctx)`, taking ~4m26s on bagaggio. Almost every check is a pure CPU aggregation over already-captured `PageCapture[]` data (string diffs, regex matches, console-entry filtering); the 3 network-bound ones (`seo-deep-audit`, `footer-links-health`, `plp-pagination`) are I/O-bound and parallelize cleanly too. Now uses `Promise.all` — the whole checks phase is dominated by the slowest single check. **Expected speedup: ~4m26s → ~30-60s** on bagaggio.
+* **Console errors now dedupe across pages.** Previously the same error message ("A chave utilizada não corresponde ao domínio…") that appeared on 4 pages produced 4 separate top-level issues, crowding the report. Now `consoleErrorsBaseline` groups by normalized error key across all page pairs and emits ONE issue per unique error with the affected-pages list inline:
+  ```
+  [high] [generic] novo erro de console em 4 páginas (/ · /s · /search · /:::desktop): A chave utilizada não corresponde…
+  ```
+  Direct response to user feedback ("console log não deveria ter um teste pra ele … fazer o dedup e descrever quais páginas tiveram aquele error"). The schema-level Issue shape is unchanged (page = first affected, details = full list).
+
+## [0.11.15](https://github.com/decocms/parity/compare/v0.11.14...v0.11.15) (2026-06-17)
+
+### Changed
+
+* **Viewports now run in parallel during collect.** Building on 0.11.14's parallel-sides change, the outer `for (const viewport of viewports)` loop is now wrapped in `runWithConcurrency` so the default `mobile,desktop` set runs fully concurrent — 4 BrowserContexts simultaneously (mobile/prod + mobile/cand + desktop/prod + desktop/cand). **Expected collect speedup: another ~40% on top of 0.11.14** (~9min → ~5–6min on bagaggio).
+* **New `--max-viewport-concurrency <n>` flag (default 2).** Lets memory-constrained machines fall back to serial-viewport behavior, and caps concurrency for runs that add a 3rd viewport (e.g. `--viewports mobile,tablet,desktop`).
+* **LLM call concurrency cap.** With 4 sides running, recovery-budget × side count could push 8–12 simultaneous LLM calls — past Anthropic tier-1's 4 RPS. Added a process-global semaphore in `src/llm/client.ts` (3 concurrent slots, queue thereafter) so calls never blow up with 429s — they queue. Default of 3 leaves headroom for the post-collect aggregate call.
+
+## [0.11.14](https://github.com/decocms/parity/compare/v0.11.13...v0.11.14) (2026-06-17)
+
+### Added
+
+* **Always-visible elapsed counter.** A 1-second ticker rewrites the spinner with `⏱ 04:32 · <current label>` from the moment "Launching browser…" appears until the report writes. No more "is it stuck?" minutes of silence — the user knows exactly how long the run has been going at any moment. Cleared in `finally` so it can't survive past `runCommand`.
+* **Per-flow timing in the bottom summary.** New `flows breakdown` block lists each flow's `max` time (the parallel-wall-clock contribution) plus per-side detail:
+  ```
+  flows breakdown (sides run in parallel within viewport)
+    purchase-journey  max 1m32s · mobile/prod 52s · mobile/cand 58s · desk/prod 89s · desk/cand 90s
+    search            max 1m48s · …
+  ```
+  Driven by the existing `FlowCapture.totalDurationMs` (already populated by `finalize` — just wasn't surfaced).
+
+### Changed
+
+* **Prod + cand now run in parallel within each viewport.** `parity run`'s collect loop was 100% sequential: `for viewport { for side { for flow {…} } }`. The two sides for a given viewport are already independent (separate BrowserContexts, separate HAR/trace paths), so the same `Promise.all([prod, cand])` pattern that `parity journey` has used for months now applies to `parity run` too. Extracted a `runOneSide(viewport, side)` helper, replaced the inner side loop with `Promise.all`, and deferred `promoteStepsFromFlow` (which mutates the shared `learned` object) until after `Promise.all` resolves so there's no race on selector promotion. **Expected speedup: ~50% on collect phase** (25m → ~17m total on bagaggio). Viewports still serialize for now — PR #2 will parallelize those too.
+
 ## [0.11.13](https://github.com/decocms/parity/compare/v0.11.12...v0.11.13) (2026-06-17)
 
 ### Fixed

package/dist/cli.js

@@ -23409,19 +23409,42 @@ function makeTimeoutSignal(ms) {
   const t = setTimeout(() => controller.abort(new Error(`LLM call timed out after ${ms}ms`)), ms);
   return { signal: controller.signal, clear: () => clearTimeout(t) };
 }
+function acquireLlmSlot() {
+  if (activeLlmCalls < MAX_CONCURRENT_LLM_CALLS) {
+    activeLlmCalls++;
+    return Promise.resolve();
+  }
+  return new Promise((resolve) => {
+    llmQueue.push(() => {
+      activeLlmCalls++;
+      resolve();
+    });
+  });
+}
+function releaseLlmSlot() {
+  activeLlmCalls--;
+  const next2 = llmQueue.shift();
+  if (next2)
+    next2();
+}
 async function callTool(params) {
   const provider = getProvider();
   if (!provider)
     return null;
   const model = resolveModel(params.feature, provider);
   const timeoutMs = params.timeoutMs ?? defaultTimeout(params);
-  if (provider === "anthropic")
-    return callAnthropicTool(params, model, timeoutMs);
-  if (provider === "openrouter")
-    return callOpenRouterTool(params, model, timeoutMs);
-  if (provider === "claude-agent-sdk")
-    return callToolSdk({ ...params, model, timeoutMs });
-  return null;
+  await acquireLlmSlot();
+  try {
+    if (provider === "anthropic")
+      return await callAnthropicTool(params, model, timeoutMs);
+    if (provider === "openrouter")
+      return await callOpenRouterTool(params, model, timeoutMs);
+    if (provider === "claude-agent-sdk")
+      return await callToolSdk({ ...params, model, timeoutMs });
+    return null;
+  } finally {
+    releaseLlmSlot();
+  }
 }
 async function callAnthropicTool(params, model, timeoutMs) {
   const client = new Anthropic({ apiKey: process.env.ANTHROPIC_API_KEY ?? "" });
@@ -23629,13 +23652,18 @@ async function callMessage(params) {
     return null;
   const model = resolveModel(params.feature, provider);
   const timeoutMs = params.timeoutMs ?? 60000;
-  if (provider === "anthropic")
-    return callAnthropicMessage(params, model, timeoutMs);
-  if (provider === "openrouter")
-    return callOpenRouterMessage(params, model, timeoutMs);
-  if (provider === "claude-agent-sdk")
-    return callMessageSdk({ ...params, model, timeoutMs });
-  return null;
+  await acquireLlmSlot();
+  try {
+    if (provider === "anthropic")
+      return await callAnthropicMessage(params, model, timeoutMs);
+    if (provider === "openrouter")
+      return await callOpenRouterMessage(params, model, timeoutMs);
+    if (provider === "claude-agent-sdk")
+      return await callMessageSdk({ ...params, model, timeoutMs });
+    return null;
+  } finally {
+    releaseLlmSlot();
+  }
 }
 async function callAnthropicMessage(params, model, timeoutMs) {
   const client = new Anthropic({ apiKey: process.env.ANTHROPIC_API_KEY ?? "" });
@@ -23695,12 +23723,13 @@ async function callOpenRouterMessage(params, model, timeoutMs) {
     clear();
   }
 }
-var LLM_MODEL_ANTHROPIC, LLM_MODEL_OPENROUTER, forcedProvider = null, llmDisabled = false, llmLanguage = "en";
+var LLM_MODEL_ANTHROPIC, LLM_MODEL_OPENROUTER, forcedProvider = null, llmDisabled = false, llmLanguage = "en", MAX_CONCURRENT_LLM_CALLS = 3, activeLlmCalls = 0, llmQueue;
 var init_client = __esm(() => {
   init_models();
   init_claude_agent_sdk();
   LLM_MODEL_ANTHROPIC = PROVIDER_MODELS.anthropic.sonnet;
   LLM_MODEL_OPENROUTER = PROVIDER_MODELS.openrouter.sonnet;
+  llmQueue = [];
 });
 
 // src/cli.ts
@@ -43862,7 +43891,7 @@ function detailFor(prod, cand, viewport) {
 function consoleErrorsBaseline(ctx) {
   const start = Date.now();
   const { pairs } = pairCaptures(ctx.prodPages, ctx.candPages);
-  const issues = [];
+  const byKey = new Map;
   let totalNew = 0;
   for (const pair of pairs) {
     const diff = diffConsole(pair.prod.console, pair.cand.console, {
@@ -43871,24 +43900,45 @@ function consoleErrorsBaseline(ctx) {
     });
     totalNew += diff.newInCand.length;
     for (const e of diff.newInCand) {
-      issues.push({
-        id: `console:${pair.key}:${hash(e.key)}`,
-        severity: e.cls === "hydration" ? "critical" : "high",
-        category: "console",
-        page: pair.key,
-        check: "console-errors-baseline",
-        summary: `[${e.cls}] novo erro de console em ${pair.key}: ${truncate4(e.entry.text, 160)}`,
-        details: e.entry.text,
-        evidence: [{ kind: "screenshot", path: pair.cand.screenshotPath, label: "cand" }]
-      });
+      const existing = byKey.get(e.key);
+      if (!existing) {
+        byKey.set(e.key, {
+          cls: e.cls,
+          sampleText: e.entry.text,
+          pages: [pair.key],
+          sampleEvidence: [
+            { kind: "screenshot", path: pair.cand.screenshotPath, label: "cand" }
+          ]
+        });
+      } else if (!existing.pages.includes(pair.key)) {
+        existing.pages.push(pair.key);
+      }
     }
   }
+  const issues = [];
+  for (const [key, agg] of byKey) {
+    const pageList = agg.pages.length <= 5 ? agg.pages.join(" · ") : `${agg.pages.slice(0, 5).join(" · ")} +${agg.pages.length - 5} more`;
+    const pageSuffix = agg.pages.length === 1 ? `em ${agg.pages[0]}` : `em ${agg.pages.length} páginas (${pageList})`;
+    const severity = agg.cls === "hydration" ? "critical" : "high";
+    issues.push({
+      id: `console:${hash(key)}`,
+      severity,
+      category: "console",
+      page: agg.pages[0],
+      check: "console-errors-baseline",
+      summary: `[${agg.cls}] novo erro de console ${pageSuffix}: ${truncate4(agg.sampleText, 160)}`,
+      details: `${agg.sampleText}
+
+Observed on: ${agg.pages.join(", ")}`,
+      evidence: agg.sampleEvidence
+    });
+  }
   return {
     name: "console-errors-baseline",
     status: issues.length > 0 ? "fail" : "pass",
     severity: "critical",
     durationMs: Date.now() - start,
-    summary: `${totalNew} novo(s) erro(s) de console em cand não presentes em prod`,
+    summary: issues.length > 0 ? `${issues.length} erro(s) único(s) novo(s) em cand (${totalNew} ocorrência(s) somando todas as páginas)` : `${totalNew} novo(s) erro(s) de console em cand não presentes em prod`,
     issues
   };
 }
@@ -47155,7 +47205,7 @@ function getCheckByName(name) {
 }
 async function runAllChecks(ctx, outResults) {
   const results = outResults ?? [];
-  for (const check of ALL_CHECKS) {
+  await Promise.all(ALL_CHECKS.map(async (check) => {
     const start = Date.now();
     try {
       const r = await check(ctx);
@@ -47170,7 +47220,7 @@ async function runAllChecks(ctx, outResults) {
         issues: []
       });
     }
-  }
+  }));
   return results;
 }
 
@@ -54522,7 +54572,7 @@ class TimingRegistry {
     };
   }
 }
-function formatTimingsSummary(timings) {
+function formatTimingsSummary(timings, flowBreakdown) {
   if (timings.phases.length === 0) {
     return `⏱  Run completed in ${formatMs(timings.totalMs)}`;
   }
@@ -54535,9 +54585,41 @@ function formatTimingsSummary(timings) {
     const pctOfTotal = timings.totalMs > 0 ? Math.round(p.durationMs / timings.totalMs * 100) : 0;
     lines.push(`    ${p.phase.padEnd(maxLabel)}  ${formatMs(p.durationMs).padStart(7)}  ${bar} ${pctOfTotal}%`);
   }
+  if (flowBreakdown && flowBreakdown.length > 0) {
+    const maxFlowLabel = Math.max(...flowBreakdown.map((f) => f.flow.length));
+    lines.push("");
+    lines.push("    flows breakdown (sides run in parallel within viewport)");
+    for (const f of flowBreakdown) {
+      const sideDetail = f.sides.map((s) => `${s.viewport}/${s.side} ${formatMs(s.durationMs)}`).join(" · ");
+      lines.push(`    ${f.flow.padEnd(maxFlowLabel)}  max ${formatMs(f.maxMs).padStart(6)} · ${sideDetail}`);
+    }
+  }
   return lines.join(`
 `);
 }
+function buildFlowBreakdown(captures) {
+  const byFlow = new Map;
+  for (const c of captures) {
+    const existing = byFlow.get(c.flow);
+    const entry = {
+      viewport: c.viewport,
+      side: c.side,
+      durationMs: c.totalDurationMs
+    };
+    if (!existing) {
+      byFlow.set(c.flow, {
+        flow: c.flow,
+        maxMs: c.totalDurationMs,
+        sides: [entry]
+      });
+    } else {
+      existing.sides.push(entry);
+      if (c.totalDurationMs > existing.maxMs)
+        existing.maxMs = c.totalDurationMs;
+    }
+  }
+  return Array.from(byFlow.values());
+}
 function formatMs(ms) {
   const totalSec = Math.floor(ms / 1000);
   if (totalSec < 60)
@@ -55069,6 +55151,21 @@ async function runCommand(rawOpts) {
   let browser = null;
   const allFlowCaptures = [];
   const allPageCaptures = [];
+  let currentSpinnerLabel = "Launching browser…";
+  const renderSpinner = () => {
+    const elapsedSec = Math.floor((Date.now() - startedAt) / 1000);
+    const min = Math.floor(elapsedSec / 60);
+    const sec = elapsedSec % 60;
+    const mmss = `${min.toString().padStart(2, "0")}:${sec.toString().padStart(2, "0")}`;
+    spinner.text = `${chalk16.dim(`⏱ ${mmss} ·`)} ${currentSpinnerLabel}`;
+  };
+  const setSpinnerLabel = (label) => {
+    currentSpinnerLabel = label;
+    renderSpinner();
+  };
+  setSpinnerLabel("Launching browser…");
+  const elapsedTicker = setInterval(renderSpinner, 1000);
+  elapsedTicker.unref?.();
   const timings = new TimingRegistry;
   let phaseStart = performance.now();
   const stampPhase = (label) => {
@@ -55113,7 +55210,7 @@ async function runCommand(rawOpts) {
       process.stderr.write(`  report parcial em ${paths.reportHtml}
 `);
       process.stderr.write(`
-${formatTimingsSummary(partialTimings)}
+${formatTimingsSummary(partialTimings, buildFlowBreakdown(allFlowCaptures))}
 `);
     } catch (err) {
       process.stderr.write(`  falha escrevendo report parcial: ${err.message}
@@ -55155,26 +55252,27 @@ ${formatTimingsSummary(partialTimings)}
         warmupSpinner.fail(`Warmup falhou — 0/${result.attempted} requests ok. Cache pode estar stale. Primeira falha: ${firstReason}`);
       }
     }
-    for (const viewport of viewports) {
-      for (const side of ["prod", "cand"]) {
-        const baseUrl = side === "prod" ? opts.prod : opts.cand;
-        spinner.text = `[${viewport}/${side}] preparing context…`;
-        const harPath = join16(paths.harDir, `${viewport}-${side}.har`);
-        const tracePath = join16(paths.tracesDir, `${viewport}-${side}.zip`);
-        const ctx = await newContext(browser, {
-          viewport,
-          harPath,
-          tracesDir: paths.tracesDir,
-          cohortCookieValue: "control",
-          noCache: opts.bypassCache
-        });
-        await installVitalsCollector(ctx);
+    const runOneSide = async (viewport, side) => {
+      const baseUrl = side === "prod" ? opts.prod : opts.cand;
+      const harPath = join16(paths.harDir, `${viewport}-${side}.har`);
+      const tracePath = join16(paths.tracesDir, `${viewport}-${side}.zip`);
+      const ctx = await newContext(browser, {
+        viewport,
+        harPath,
+        tracesDir: paths.tracesDir,
+        cohortCookieValue: "control",
+        noCache: opts.bypassCache
+      });
+      await installVitalsCollector(ctx);
+      const captures = [];
+      const pages = [];
+      try {
         for (const flow of flows) {
           const flowStart = Date.now();
           let lastStepTotal = 0;
           const sideTag = side === "prod" ? chalk16.cyan("prod") : chalk16.magenta("cand");
           const prefix = `${chalk16.dim(`[${viewport}/`)}${sideTag}${chalk16.dim("]")}`;
-          spinner.text = `${prefix} ${flow}: starting…`;
+          setSpinnerLabel(`${prefix} ${flow}: starting…`);
           const cap = await runFlow(flow, {
             baseUrl,
             side,
@@ -55190,7 +55288,7 @@ ${formatTimingsSummary(partialTimings)}
             onStep: (event) => {
               if (event.phase === "start") {
                 lastStepTotal = event.total;
-                spinner.text = `${prefix} ${flow} ${chalk16.dim(`${event.index}/${event.total}`)} ${event.name}…`;
+                setSpinnerLabel(`${prefix} ${flow} ${chalk16.dim(`${event.index}/${event.total}`)} ${event.name}…`);
               }
             }
           });
@@ -55199,7 +55297,7 @@ ${formatTimingsSummary(partialTimings)}
           const target = lastStepTotal || recorded;
           const failed = cap.steps?.find((s) => s.status === "failed");
           const lastStep = cap.steps?.[cap.steps.length - 1];
-          const elapsed = chalk16.dim(`${((Date.now() - flowStart) / 1000).toFixed(1)}s`);
+          const elapsedDim = chalk16.dim(`${((Date.now() - flowStart) / 1000).toFixed(1)}s`);
           const counts = target > 0 ? chalk16.dim(`${okSteps}/${target}`) : "";
           let glyph;
           let detail;
@@ -55213,24 +55311,42 @@ ${formatTimingsSummary(partialTimings)}
             glyph = chalk16.green("✓");
             detail = "";
           }
-          const summary = `${glyph} ${prefix} ${flow} ${counts} ${detail} ${elapsed}`.replace(/\s+/g, " ").trim();
+          const summary = `${glyph} ${prefix} ${flow} ${counts} ${detail} ${elapsedDim}`.replace(/\s+/g, " ").trim();
           spinner.stopAndPersist({ symbol: "", text: summary });
-          spinner.start(`${prefix} ${flow} done — next…`);
-          allFlowCaptures.push(cap);
+          setSpinnerLabel(`${prefix} ${flow} done — next…`);
+          spinner.start();
+          captures.push(cap);
           for (const p of cap.pages)
-            allPageCaptures.push(p);
+            pages.push(p);
+        }
+      } finally {
+        await stopTracing(ctx, tracePath).catch(() => {
+          return;
+        });
+        await ctx.close();
+      }
+      return { captures, pages };
+    };
+    const runOneViewport = async (viewport) => {
+      const [prodResult, candResult] = await Promise.all([
+        runOneSide(viewport, "prod"),
+        runOneSide(viewport, "cand")
+      ]);
+      for (const r of [prodResult, candResult]) {
+        for (const cap of r.captures) {
+          allFlowCaptures.push(cap);
           if (opts.learn !== false) {
             const result = promoteStepsFromFlow(learned, platform, prodHost, cap);
             promotedCount += result.promoted;
             deprecatedCount += result.deprecated;
           }
         }
-        await stopTracing(ctx, tracePath).catch(() => {
-          return;
-        });
-        await ctx.close();
+        for (const p of r.pages)
+          allPageCaptures.push(p);
       }
-    }
+    };
+    const maxViewportConcurrency = Math.max(1, opts.maxViewportConcurrency ?? 2);
+    await runWithConcurrency2(viewports, maxViewportConcurrency, runOneViewport);
     stampPhase("collect");
     spinner.succeed("Coleta concluída");
     if (isLlmAvailable()) {
@@ -55449,7 +55565,8 @@ ${formatTimingsSummary(partialTimings)}
       }
     }
     currentPhase = "checks";
-    spinner.start("Rodando checks…");
+    setSpinnerLabel("Rodando checks…");
+    spinner.start();
     const checksHb = attachSpinnerHeartbeat(spinner, { baseText: "Rodando checks…" });
     const cacheDir = join16(opts.output, "cache");
     const checkCtx = {
@@ -55474,7 +55591,8 @@ ${formatTimingsSummary(partialTimings)}
     }
     const allIssues = checks.flatMap((c) => c.issues);
     currentPhase = "llm-aggregate";
-    spinner.start(isLlmAvailable() ? `Agregando issues via LLM (${providerLabel()}, timeout=${llmTimeoutSec}s)…` : "Agregando issues (modo offline — sem LLM keys)…");
+    setSpinnerLabel(isLlmAvailable() ? `Agregando issues via LLM (${providerLabel()}, timeout=${llmTimeoutSec}s)…` : "Agregando issues (modo offline — sem LLM keys)…");
+    spinner.start();
     const llmHb = attachSpinnerHeartbeat(spinner, {
       baseText: isLlmAvailable() ? "Agregando issues via LLM (Sonnet 4.6)…" : "Agregando issues (modo offline)…"
     });
@@ -55556,7 +55674,7 @@ ${formatTimingsSummary(partialTimings)}
     });
     printSummary5(run, paths.reportHtml, { promotedCount, deprecatedCount, platform });
     console.log(`
-${formatTimingsSummary(runTimings)}`);
+${formatTimingsSummary(runTimings, buildFlowBreakdown(allFlowCaptures))}`);
     if (opts.ci) {
       const blocking = allIssues.filter((i) => failOn.includes(i.severity));
       if (blocking.length > 0) {
@@ -55586,6 +55704,7 @@ ${formatTimingsSummary(runTimings)}`);
     });
     return 2;
   } finally {
+    clearInterval(elapsedTicker);
     clearTimeout(globalTimeoutTimer);
     process.removeListener("SIGINT", onSignal);
     process.removeListener("SIGTERM", onSignal);
@@ -56416,7 +56535,7 @@ program.command("run").description([
   "  auto-selectors=ON (if LLM), learn=ON, cache=ON, visual-diff=ON,",
   "  warmup=OFF, bypass-cache=OFF, ci=OFF."
 ].join(`
-`)).requiredOption("--prod <url>", "Production URL (source of truth, e.g. Fresh site)").requiredOption("--cand <url>", "Candidate URL (migrated site, e.g. TanStack)").option("--preset <name>", "Bundle of defaults: smoke (fast, ~30s, no LLM) | full (deep audit) | ci (CI-tuned)").option("--flows <list>", "Comma-separated flows: homepage,plp,pdp,purchase-journey", "purchase-journey").option("--viewports <list>", "Comma-separated viewports: mobile,desktop", "mobile,desktop").option("--cep <cep>", "CEP for shipping calculation", "01310-100").option("--runs <n>", "Repeat each measurement N times (median)", "1").option("--baseline <name>", "Compare against a saved baseline").option("--output <dir>", "Output directory", "./parity-output").option("--ci", "CI mode: stricter exit codes", false).option("--fail-on <severities>", "Comma-separated severities that cause exit 1", "critical").option("--open", "Open the HTML report after the run completes", false).option("--no-auto-selectors", "Disable LLM-based selector discovery (uses defaults instead)").option("--refresh-selectors", "Bypass selector cache and re-run discovery", false).option("--no-learn", "Don't write to learned-selectors.json (read-only mode)").option("--vitals-pages <n>", "Extra pages from sitemap to crawl for Vitals coverage (default 10)", (v) => Number(v), 10).option("--visual-pages <n>", "Pages to compare visually via LLM (home + sampled PLPs/PDPs from sitemap, default 5)", (v) => Number(v), 5).option("--pages <list>", 'Comma-separated paths to compare visually (overrides sitemap discovery). E.g. "/,/account,/p/some-product". Use this when you want deterministic coverage instead of sampled.').option("--pages-file <path>", "Read paths to compare visually from a text file (one path per line). Lines starting with # are ignored. Overrides --pages when both are present.").option("--no-visual-diff", "Skip the visual diff capture pass entirely").option("--no-cache", "Disable the cross-run visual-diff cache (forces a fresh LLM judgment on every page).").option("--clear-cache", "Wipe the visual-diff verdict cache (parity-output/cache/verdicts.json) before the run.", false).option("--bypass-cache", "Bypass CDN/edge caches: append a cache-busting query param and send Cache-Control: no-cache on every request. Use right after a deploy to avoid false failures from stale CF edge content.", false).option("--warmup", "Before measurement, hit each target URL once (per viewport) with a cache-buster so the Worker serves a fresh response. Recommended after deploys.", false).option("--accept-prod-quirks", "Demote prod-side cart-empty journey failures (VTEX session quirk) from failed to skipped. The cart-reveal-mode-divergence check still emits critical if prod/cand markup intents differ, so this flag never masks a real regression. See issue #12.", false).option("--llm-timeout <seconds>", "Hard timeout for the LLM aggregation call (seconds). The run completes in offline mode if the LLM hangs past this. Default: 60.", (v) => Number(v), 60).option("--timeout <minutes>", "Hard wall-clock cap for the whole run (minutes). On expiry, parity writes a partial report and exits 130. Default: 30. Issue #56.", (v) => Number(v), 30).option("--flow <name>", "Alias for --flows when running a single flow (e.g. --flow cart). Issue #53.").option("--json <path|->", "Emit JSON-Lines (one line per check) to the given file path, or '-' for stdout. Schema versioned via leading metadata line. Issue #53.").option("--pt", "Tell the LLM to respond in Brazilian Portuguese. Only affects LLM-generated content (issues, explain, prompts) — the static HTML report stays in English. Issue #67.").option("--llm <provider>", "Force LLM provider: anthropic | openrouter | claude-code | none. Default: auto-detect (anthropic key → openrouter key → local claude CLI → none). Issue #66.").option("--llm-model <overrides>", "Per-feature model override, e.g. visual-diff=claude-opus-4-7,explain=claude-opus-4-7. Features: selector-discovery, step-recovery, search-terms, plp-matching, pdp-matching, section-understanding, visual-diff, issue-aggregation, explain. Issue #66.").option("--llm-tier-default <tier>", "Override the default tier (haiku | sonnet | opus) for every feature that doesn't have a per-feature override. Issue #66.").option("--llm-model-default <model>", "Force every LLM call to use this exact model ID, ignoring per-feature defaults. Issue #66.").option("--no-interactive", "Disable the interactive selector prompt that auto-fires in a TTY without an LLM provider. Use in scripts and CI where stdin is technically a TTY but you don't want to pause. Issue #72.").action(async (opts) => {
+`)).requiredOption("--prod <url>", "Production URL (source of truth, e.g. Fresh site)").requiredOption("--cand <url>", "Candidate URL (migrated site, e.g. TanStack)").option("--preset <name>", "Bundle of defaults: smoke (fast, ~30s, no LLM) | full (deep audit) | ci (CI-tuned)").option("--flows <list>", "Comma-separated flows: homepage,plp,pdp,purchase-journey", "purchase-journey").option("--viewports <list>", "Comma-separated viewports: mobile,desktop", "mobile,desktop").option("--cep <cep>", "CEP for shipping calculation", "01310-100").option("--runs <n>", "Repeat each measurement N times (median)", "1").option("--baseline <name>", "Compare against a saved baseline").option("--output <dir>", "Output directory", "./parity-output").option("--ci", "CI mode: stricter exit codes", false).option("--fail-on <severities>", "Comma-separated severities that cause exit 1", "critical").option("--open", "Open the HTML report after the run completes", false).option("--no-auto-selectors", "Disable LLM-based selector discovery (uses defaults instead)").option("--refresh-selectors", "Bypass selector cache and re-run discovery", false).option("--no-learn", "Don't write to learned-selectors.json (read-only mode)").option("--vitals-pages <n>", "Extra pages from sitemap to crawl for Vitals coverage (default 10)", (v) => Number(v), 10).option("--max-viewport-concurrency <n>", "How many viewports run concurrently during collect (default 2: mobile+desktop together). Set to 1 if you hit OOM with many viewports.", (v) => Number(v), 2).option("--visual-pages <n>", "Pages to compare visually via LLM (home + sampled PLPs/PDPs from sitemap, default 5)", (v) => Number(v), 5).option("--pages <list>", 'Comma-separated paths to compare visually (overrides sitemap discovery). E.g. "/,/account,/p/some-product". Use this when you want deterministic coverage instead of sampled.').option("--pages-file <path>", "Read paths to compare visually from a text file (one path per line). Lines starting with # are ignored. Overrides --pages when both are present.").option("--no-visual-diff", "Skip the visual diff capture pass entirely").option("--no-cache", "Disable the cross-run visual-diff cache (forces a fresh LLM judgment on every page).").option("--clear-cache", "Wipe the visual-diff verdict cache (parity-output/cache/verdicts.json) before the run.", false).option("--bypass-cache", "Bypass CDN/edge caches: append a cache-busting query param and send Cache-Control: no-cache on every request. Use right after a deploy to avoid false failures from stale CF edge content.", false).option("--warmup", "Before measurement, hit each target URL once (per viewport) with a cache-buster so the Worker serves a fresh response. Recommended after deploys.", false).option("--accept-prod-quirks", "Demote prod-side cart-empty journey failures (VTEX session quirk) from failed to skipped. The cart-reveal-mode-divergence check still emits critical if prod/cand markup intents differ, so this flag never masks a real regression. See issue #12.", false).option("--llm-timeout <seconds>", "Hard timeout for the LLM aggregation call (seconds). The run completes in offline mode if the LLM hangs past this. Default: 60.", (v) => Number(v), 60).option("--timeout <minutes>", "Hard wall-clock cap for the whole run (minutes). On expiry, parity writes a partial report and exits 130. Default: 30. Issue #56.", (v) => Number(v), 30).option("--flow <name>", "Alias for --flows when running a single flow (e.g. --flow cart). Issue #53.").option("--json <path|->", "Emit JSON-Lines (one line per check) to the given file path, or '-' for stdout. Schema versioned via leading metadata line. Issue #53.").option("--pt", "Tell the LLM to respond in Brazilian Portuguese. Only affects LLM-generated content (issues, explain, prompts) — the static HTML report stays in English. Issue #67.").option("--llm <provider>", "Force LLM provider: anthropic | openrouter | claude-code | none. Default: auto-detect (anthropic key → openrouter key → local claude CLI → none). Issue #66.").option("--llm-model <overrides>", "Per-feature model override, e.g. visual-diff=claude-opus-4-7,explain=claude-opus-4-7. Features: selector-discovery, step-recovery, search-terms, plp-matching, pdp-matching, section-understanding, visual-diff, issue-aggregation, explain. Issue #66.").option("--llm-tier-default <tier>", "Override the default tier (haiku | sonnet | opus) for every feature that doesn't have a per-feature override. Issue #66.").option("--llm-model-default <model>", "Force every LLM call to use this exact model ID, ignoring per-feature defaults. Issue #66.").option("--no-interactive", "Disable the interactive selector prompt that auto-fires in a TTY without an LLM provider. Use in scripts and CI where stdin is technically a TTY but you don't want to pause. Issue #72.").action(async (opts) => {
   if (opts.flow) {
     opts.flows = opts.flow;
   }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@decocms/parity",
-  "version": "0.11.13",
+  "version": "0.11.16",
   "description": "E2E parity validator for site migrations. Compares prod vs cand and reports UI, functional, SEO, visual, and Web Vitals deltas with an LLM-ranked HTML report.",
   "type": "module",
   "license": "MIT",