@trazum/cli 1.29.0 → 1.30.0

package/dist/index.js

@@ -2,7 +2,7 @@
 import { readdir, readFile, stat, writeFile } from 'node:fs/promises';
 import { join, resolve as resolvePath } from 'node:path';
 import { gunzipSync } from 'node:zlib';
-import { applyRewrites, BASELINE_FILENAME, BASELINE_VERSION, breaches, cacheableMinimum, analyzeCachePrefix, billLevers, cacheEconomics, cacheHitRate, contextPressure, comparePrompts, compareToBaseline, computeSavings, countTokensAnthropic, DEFAULT_USAGE, detectFromSource, driversBetween, estimateTokens, evaluate, extractPrompts, findExamples, formatBaseline, formatSignedUsd, formatUsd, getMessages, getModel, hasMarker, LOCALES, MAX_BASELINE_BYTES, moneyIsComparable, mostSpecificMatch, nearestName, optimize, parseBaseline, PHRASE_LANGUAGES, plannedCalls, profilePrompt, profileToCsv, profileUsage, promptId, providerFromEnv, pruneExamples, refineWithLlm, rejectionText, reorderForCache, repriceProfile, reviewAgeDays, reviewExamples, RULES, sharedPrefixes, sharesOf, SOURCE_EXTENSIONS, suggestRewrites, toOtlpMetrics, toPromptfoo, TTL_1H_MS, UNLABELLED, withExactTokenCounts, } from '@trazum/core';
+import { applyRewrites, BASELINE_FILENAME, BASELINE_VERSION, breaches, cacheableMinimum, analyzeCachePrefix, billLevers, cacheEconomics, cacheHitRate, contextPressure, comparePrompts, compareToBaseline, computeSavings, countTokensAnthropic, DEFAULT_USAGE, detectFromSource, coverageDrift, driversBetween, estimateTokens, evaluate, extractPrompts, findExamples, formatBaseline, formatSignedUsd, formatUsd, getMessages, getModel, hasMarker, LOCALES, MAX_BASELINE_BYTES, moneyIsComparable, mostSpecificMatch, nearestName, optimize, parseBaseline, PHRASE_LANGUAGES, plannedCalls, profilePrompt, profileToCsv, profileUsage, promptId, providerFromEnv, pruneExamples, refineWithLlm, rejectionText, reorderForCache, repriceProfile, reviewAgeDays, reviewExamples, RULES, sharedPrefixes, sharesOf, SOURCE_EXTENSIONS, suggestRewrites, toOtlpMetrics, toPromptfoo, TTL_1H_MS, UNLABELLED, withExactTokenCounts, } from '@trazum/core';
 import { cacheDir, cacheStats, cachingProvider, clearCache } from './suggest-cache.js';
 import { dayOf, formatGap, median, spanDays } from './time.js';
 // Everything that reads the filesystem, on its own entry point so the web
@@ -1791,7 +1791,27 @@ async function commandProfile(args, config, pricing, t) {
         }
         if (typeof args.flags.get('max-growth-usd') === 'string' && againstDelta !== null) {
             const maxGrowth = numberFlag(args, 'max-growth-usd', 0, t);
-            if (againstDelta > maxGrowth) {
+            /**
+             * A comparison that went blind fails before it is judged.
+             *
+             * The dollars can hold flat while the current log stopped recording a
+             * field the previous one carried — and every finding that needed the
+             * field is now silent for a reason that has nothing to do with spend.
+             * A gate passing there would be certifying a comparison it could not
+             * make: "not measured" is not "did not grow", the same refusal
+             * --max-day-usd makes on a clockless log and --max-session-usd on a
+             * sessionless one. Only a collapse fails; a field that appeared means
+             * this side can see more, which is never a reason to refuse.
+             */
+            const blinded = previous !== null
+                ? coverageDrift(previous.fieldCoverage, report.fieldCoverage).filter((d) => d.delta < 0)
+                : [];
+            const worst = blinded[0];
+            if (worst !== undefined) {
+                console.error(c.red(t.profile.maxGrowthCoverageLost(blinded.map((d) => t.profile.coverageField(d.field)).join(', '), pct(worst.was), pct(worst.now))));
+                process.exitCode = 1;
+            }
+            else if (againstDelta > maxGrowth) {
                 console.error(c.red(t.profile.maxGrowthUsdFailed(formatSignedUsd(againstDelta), formatUsd(maxGrowth))));
                 process.exitCode = 1;
             }
@@ -2945,6 +2965,40 @@ async function commandProfile(args, config, pricing, t) {
                     console.log(`  ${d.delta > 0 ? c.yellow(line) : c.dim(line)}`);
                 }
             }
+            /**
+             * What the comparison stopped being able to see.
+             *
+             * Every figure above is dollars, and dollars cannot tell a finding that
+             * was fixed from a finding whose field the log stopped recording — both
+             * are silence. This is the only section that can, so it is loud: a
+             * collapse in coverage invalidates whichever findings depended on it,
+             * and reading the drop as good news is the specific mistake it exists
+             * to prevent.
+             */
+            const drifts = coverageDrift(previous.fieldCoverage, report.fieldCoverage);
+            if (drifts.length > 0) {
+                console.log();
+                for (const drift of drifts) {
+                    const line = t.profile.coverageDrift(t.profile.coverageField(drift.field), pct(drift.was), pct(drift.now));
+                    console.log(drift.delta < 0
+                        ? `  ${c.yellow('!')} ${c.bold(wrap(line, 74, '    '))}`
+                        : `  ${c.dim(wrap(line, 74, '  '))}`);
+                    /**
+                     * Which findings went with it, named. "Some findings are silent" is
+                     * not something a reader can act on; knowing that conversation
+                     * growth and the cache-TTL fit are now silence rather than absence
+                     * tells them exactly which sections of this report to distrust.
+                     */
+                    if (drift.delta < 0) {
+                        const silenced = t.profile.coverageSilenced(drift.field);
+                        if (silenced !== '')
+                            console.log(`    ${c.dim(wrap(silenced, 72, '    '))}`);
+                    }
+                }
+                if (drifts.some((d) => d.delta < 0)) {
+                    console.log(`  ${c.dim(wrap(t.profile.coverageDriftWhy(), 74, '    '))}`);
+                }
+            }
         }
     }
     for (const [heading, rows] of [