@trazum/cli 1.37.0 → 1.39.0

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, coverageDrift, driversBetween, explainGateFailure, assignSources, fleetRollup, labelCoverage, measuredUsage, gateMargin, GATE_MARGIN_TIGHT, 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, buildPlan, verifyPlan, cacheEconomics, cacheHitRate, contextPressure, comparePrompts, compareToBaseline, computeSavings, countTokensAnthropic, DEFAULT_USAGE, detectFromSource, coverageDrift, driversBetween, explainGateFailure, assignSources, fleetRollup, labelCoverage, measuredUsage, gateMargin, GATE_MARGIN_TIGHT, 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
@@ -26,6 +26,7 @@ const c = {
 const VALUE_FLAGS = new Set([
     'against',
     'from-log',
+    'min-usd',
     // `route` takes a path here, and the flag is deliberately not `--prompt`:
     // everywhere else in this tool `--prompt` names a marked prompt *inside* a
     // source file, and reusing it for a path would be a trap laid for the reader.
@@ -159,6 +160,12 @@ function levelFlag(args, config, t) {
  * model id. It beats the default because reading the code is better than
  * assuming, and loses to config because being told is better than reading.
  */
+/**
+ * The file names a usage log answers to, shared by every command that reads a
+ * directory of them. One list, because two commands disagreeing on what counts
+ * as a log would be the same directory billing differently by verb.
+ */
+const LOG_EXTENSIONS = ['.jsonl', '.ndjson', '.log', '.json'];
 /**
  * One usage log, gzip included, shared by every command that reads one.
  *
@@ -279,6 +286,8 @@ const COMMAND_FLAGS = {
     check: ['max-tokens', 'level', 'exact-tokens', 'markdown-out', 'baseline'],
     baseline: ['model', 'calls', 'output-tokens', 'cache-hit-rate', 'batch', 'exact-tokens', 'out', 'o'],
     profile: ['json', 'pricing', 'pricing-live', 'against', 'what-if', 'markdown-out', 'csv-out', 'csv-shape', 'max-usd', 'max-growth-usd', 'max-cache-loss-usd', 'max-day-usd', 'max-session-usd', 'label', 'since', 'until', 'dry-run', 'markdown-summary', 'by-source'],
+    plan: ['json', 'out', 'markdown-out', 'min-usd', 'pricing', 'pricing-live'],
+    verify: ['against', 'gate', 'json', 'markdown-out', 'pricing', 'pricing-live'],
     route: ['prompt-file', 'cases', 'label', 'concurrency', 'json', 'yes', 'pricing', 'pricing-live'],
     eval: ['cases', 'level', 'concurrency', 'export', 'out', 'o', 'model'],
     prune: ['cases', 'concurrency', 'json', 'yes'],
@@ -1636,6 +1645,243 @@ function isoDate() {
  * metered API calls somebody was actually billed for — the bill exists wherever
  * Trazum happens to be running, so the host has no bearing on it.
  */
+/**
+ * `trazum verify <plan.json> --against <newer.jsonl|dir>` — did it work?
+ *
+ * The plan predicted; this holds the prediction to the log that came after
+ * it. Three outcomes and never two — arrived, did not arrive, cannot be told
+ * — because "cannot be told" rendered as "arrived" is how every other tool
+ * congratulates a team for a workload that merely vanished. With `--gate`,
+ * a broken promise is a failing exit code: a different and more useful gate
+ * than "spend went up".
+ */
+async function commandVerify(args, pricing, t) {
+    const planPath = args.positional[0];
+    if (planPath === undefined)
+        throw new Error(t.verify.noTarget());
+    const againstPath = stringFlag(args, 'against');
+    if (againstPath === undefined)
+        throw new Error(t.verify.needsAgainst());
+    let plan;
+    try {
+        const parsed = JSON.parse(await readFile(planPath, 'utf8'));
+        if (parsed?.schemaVersion !== 1 || !Array.isArray(parsed.actions)) {
+            throw new Error(t.verify.badPlan(planPath));
+        }
+        plan = parsed;
+    }
+    catch (error) {
+        if (error instanceof SyntaxError)
+            throw new Error(t.verify.badPlan(planPath));
+        throw error;
+    }
+    const GZ = LOG_EXTENSIONS.map((ext) => `${ext}.gz`);
+    const READABLE = [...LOG_EXTENSIONS, ...GZ];
+    const target = await stat(againstPath).catch(() => null);
+    let files = [againstPath];
+    if (target?.isDirectory()) {
+        const entries = await readdir(againstPath, { withFileTypes: true });
+        files = entries
+            .filter((entry) => entry.isFile() && READABLE.some((ext) => entry.name.endsWith(ext)))
+            .map((entry) => join(againstPath, entry.name))
+            .sort((a, b) => a.localeCompare(b));
+        if (files.length === 0)
+            throw new Error(t.profile.noLogsInDirectory(againstPath, READABLE.join(', ')));
+    }
+    const texts = await Promise.all(files.map((file) => readUsageLog(file, t)));
+    const raw = texts.map((text) => (text.endsWith('\n') ? text : `${text}\n`)).join('');
+    const report = profileUsage(raw, { catalogue: pricing });
+    const verification = verifyPlan(plan, report, { currentPricingLastReviewed: pricing.lastReviewed });
+    const gate = boolFlag(args, 'gate');
+    const n = (value) => value.toLocaleString(t.numberLocale);
+    const lines = (md) => {
+        const out = [];
+        const actionLine = (v) => {
+            const name = v.action.label === UNLABELLED ? t.profile.unlabelled() : v.action.label;
+            const rows = [];
+            rows.push(t.verify.action(v.action.kind, name, v.action.model, v.outcome));
+            if (v.outcome === 'cannot-tell' && v.reason !== null)
+                rows.push(t.verify.reason(v.reason));
+            if (v.action.kind === 'route' || v.action.kind === 'route+batch') {
+                if (v.outcome !== 'cannot-tell') {
+                    rows.push(t.verify.routeObserved(String(v.observed.dearestModel ?? ''), formatUsd(Number(v.observed.onTargetUsd ?? 0)), formatUsd(Number(v.observed.onOldModelUsd ?? 0))));
+                }
+                if (v.action.kind === 'route+batch' && v.outcome !== 'cannot-tell')
+                    rows.push(t.verify.batchUnobservable());
+            }
+            if (v.action.kind === 'fix-truncation' && v.outcome === 'not-arrived') {
+                rows.push(t.verify.truncationObserved(formatUsd(Number(v.observed.retryBillUsd ?? 0))));
+            }
+            if (v.action.kind === 'fix-caching' && v.outcome !== 'cannot-tell') {
+                rows.push(t.verify.cacheObserved(formatUsd(Number(v.observed.deltaUsd ?? 0)), v.outcome));
+            }
+            if (v.attribution?.calls !== undefined) {
+                rows.push(t.verify.attribution(n(Math.round(v.attribution.calls.before)), n(Math.round(v.attribution.calls.after)), n(Math.round(v.attribution.outputPerCallTokens?.before ?? 0)), n(Math.round(v.attribution.outputPerCallTokens?.after ?? 0))));
+            }
+            return rows;
+        };
+        const heading = t.verify.heading(n(verification.actions.length), verification.planCreatedAt === null ? null : verification.planCreatedAt.slice(0, 10));
+        out.push(md ? `## ${heading}` : heading);
+        out.push(t.verify.counts(n(verification.arrived), n(verification.notArrived), n(verification.cannotTell)));
+        if (verification.pricesChanged) {
+            out.push(t.verify.pricesChanged(verification.planPricing, verification.currentPricing));
+        }
+        for (const v of verification.actions) {
+            out.push('');
+            const [head, ...rest] = actionLine(v);
+            out.push(md ? `### ${head}` : `→ ${head}`);
+            for (const row of rest)
+                out.push(md ? `- ${row}` : `  · ${row}`);
+        }
+        out.push('');
+        out.push(t.verify.footer());
+        return out;
+    };
+    await writeMarkdown(args, () => lines(true).join('\n'));
+    if (boolFlag(args, 'json')) {
+        console.log(JSON.stringify(verification, null, 2));
+    }
+    else {
+        const [head, ...rest] = lines(false);
+        console.log(c.bold(head));
+        for (const row of rest) {
+            console.log(row === '' ? '' : `  ${wrap(row, 74, '    ')}`);
+        }
+    }
+    if (gate) {
+        if (verification.gateFailures > 0) {
+            console.error(c.red(t.verify.gateFailed(n(verification.gateFailures), n(verification.actions.length))));
+            process.exitCode = 1;
+        }
+        else {
+            console.log(c.green(t.verify.gateOk()));
+        }
+    }
+}
+/**
+ * `trazum plan <log>` — not a list of findings, a ranked plan of what to do.
+ *
+ * The composition (route and batch on one slice never summed) happens in
+ * core's `buildPlan`; this command owns the I/O and the rendering. The plan
+ * saves as a dated JSON file on request, which is what makes verifying it
+ * against a later log possible at all — a prediction nobody wrote down is a
+ * prediction nobody can be held to.
+ */
+async function commandPlan(args, pricing, t) {
+    const path = args.positional[0];
+    if (path === undefined)
+        throw new Error(t.plan.noTarget());
+    const GZ = LOG_EXTENSIONS.map((ext) => `${ext}.gz`);
+    const READABLE = [...LOG_EXTENSIONS, ...GZ];
+    const target = await stat(path).catch(() => null);
+    let files = [path];
+    if (target?.isDirectory()) {
+        const entries = await readdir(path, { withFileTypes: true });
+        files = entries
+            .filter((entry) => entry.isFile() && READABLE.some((ext) => entry.name.endsWith(ext)))
+            .map((entry) => join(path, entry.name))
+            .sort((a, b) => a.localeCompare(b));
+        if (files.length === 0)
+            throw new Error(t.profile.noLogsInDirectory(path, READABLE.join(', ')));
+    }
+    const texts = await Promise.all(files.map((file) => readUsageLog(file, t)));
+    const raw = texts.map((text) => (text.endsWith('\n') ? text : `${text}\n`)).join('');
+    const report = profileUsage(raw, { catalogue: pricing });
+    if (report.total.calls === 0)
+        throw new Error(t.plan.nothingPriced());
+    const levers = billLevers(report, { catalogue: pricing });
+    const plan = buildPlan(report, levers, pricing.lastReviewed);
+    const minUsd = typeof args.flags.get('min-usd') === 'string' ? numberFlag(args, 'min-usd', 0, t) : 0;
+    const actions = plan.actions.filter((a) => (a.savingUsd ?? a.stakeUsd ?? 0) >= minUsd);
+    const filtered = plan.actions.length - actions.length;
+    const droppedUsd = plan.actions
+        .filter((a) => (a.savingUsd ?? a.stakeUsd ?? 0) < minUsd)
+        .reduce((sum, a) => sum + (a.savingUsd ?? a.stakeUsd ?? 0), 0);
+    const n = (value) => value.toLocaleString(t.numberLocale);
+    /**
+     * The document's totals cover the actions the document holds — a filtered
+     * plan whose totals still counted the filtered actions would be a file
+     * that contradicts itself, and 1.39's verify would hold it to money it
+     * cannot see. What --min-usd dropped is stated with its worth, never
+     * silently.
+     */
+    const stamped = {
+        ...plan,
+        actions,
+        projectedSavingUsd: actions.reduce((sum, a) => sum + (a.savingUsd ?? 0), 0),
+        measuredStakeUsd: actions.reduce((sum, a) => sum + (a.stakeUsd ?? 0), 0),
+        createdAt: new Date().toISOString(),
+    };
+    const outPath = stringFlag(args, 'out');
+    if (outPath !== undefined) {
+        await writeFile(outPath, `${JSON.stringify(stamped, null, 2)}\n`);
+    }
+    await writeMarkdown(args, () => {
+        const lines = [];
+        lines.push(`## ${t.plan.heading(n(actions.length), formatUsd(plan.totalUsd))}`);
+        lines.push('');
+        lines.push(t.plan.totals(formatUsd(stamped.projectedSavingUsd), formatUsd(stamped.measuredStakeUsd)));
+        if (plan.span === null) {
+            lines.push('');
+            lines.push(`_${t.plan.noClock()}_`);
+        }
+        for (const action of actions) {
+            const name = action.label === UNLABELLED ? t.profile.unlabelled() : action.label;
+            const money = action.savingUsd !== null
+                ? t.plan.projected(formatUsd(action.savingUsd))
+                : t.plan.staked(formatUsd(action.stakeUsd ?? 0));
+            lines.push('');
+            lines.push(`### ${t.plan.action(action.kind, name, action.model)} — ${money}`);
+            if (action.detail.routeTo !== undefined)
+                lines.push(`- ${t.plan.routeTo(action.detail.routeTo.displayName)}`);
+            for (const assumption of action.assumes)
+                lines.push(`- ${t.plan.assume(assumption)}`);
+            if (action.check !== null)
+                lines.push(`- ${t.plan.check(action.check)}`);
+        }
+        if (filtered > 0) {
+            lines.push('');
+            lines.push(`_${t.plan.filtered(n(filtered), formatUsd(minUsd), formatUsd(droppedUsd))}_`);
+        }
+        lines.push('');
+        lines.push(`_${t.plan.footer()}_`);
+        return lines.join('\n');
+    });
+    if (boolFlag(args, 'json')) {
+        console.log(JSON.stringify(stamped, null, 2));
+        return;
+    }
+    console.log(c.bold(t.plan.heading(n(actions.length), formatUsd(plan.totalUsd))));
+    console.log(`  ${wrap(t.plan.totals(formatUsd(stamped.projectedSavingUsd), formatUsd(stamped.measuredStakeUsd)), 74, '  ')}`);
+    if (plan.span === null) {
+        console.log(`  ${c.dim(wrap(t.plan.noClock(), 74, '  '))}`);
+    }
+    for (const action of actions) {
+        const name = action.label === UNLABELLED ? t.profile.unlabelled() : action.label;
+        const money = action.savingUsd !== null
+            ? t.plan.projected(formatUsd(action.savingUsd))
+            : t.plan.staked(formatUsd(action.stakeUsd ?? 0));
+        console.log();
+        console.log(`  ${c.green('→')} ${c.bold(t.plan.action(action.kind, name, action.model))}  ${money}`);
+        if (action.detail.routeTo !== undefined) {
+            console.log(`    ${c.dim(t.plan.routeTo(action.detail.routeTo.displayName))}`);
+        }
+        for (const assumption of action.assumes) {
+            console.log(`    ${c.yellow('?')} ${c.dim(wrap(t.plan.assume(assumption), 72, '      '))}`);
+        }
+        if (action.check !== null) {
+            console.log(`    ${c.dim(wrap(t.plan.check(action.check), 72, '      '))}`);
+        }
+    }
+    if (filtered > 0) {
+        console.log();
+        console.log(`  ${c.dim(wrap(t.plan.filtered(n(filtered), formatUsd(minUsd), formatUsd(droppedUsd)), 74, '  '))}`);
+    }
+    console.log();
+    console.log(`  ${c.dim(wrap(t.plan.footer(), 74, '  '))}`);
+    if (outPath !== undefined)
+        console.log(c.dim(wrap(t.plan.wrote(outPath), 74, '')));
+}
 async function commandProfile(args, config, pricing, t) {
     const path = args.positional[0];
     if (path === undefined) {
@@ -1658,7 +1904,6 @@ async function commandProfile(args, config, pricing, t) {
      * directory holding nothing readable is an error naming what it looked for,
      * not an empty report.
      */
-    const LOG_EXTENSIONS = ['.jsonl', '.ndjson', '.log', '.json'];
     /**
      * The same names, gzipped — which is what a rotated log actually looks like
      * a day after it rotates.
@@ -4801,6 +5046,12 @@ async function main() {
         case 'profile':
             await commandProfile(args, config, pricing, t);
             break;
+        case 'plan':
+            await commandPlan(args, pricing, t);
+            break;
+        case 'verify':
+            await commandVerify(args, pricing, t);
+            break;
         case 'route':
             await commandRoute(args, pricing, t);
             break;