mcp-context-cost 0.13.0 → 0.13.1

package/README.md

@@ -126,7 +126,7 @@ costs **more** than loading the definitions would.
 
 Three things the report will not do: it will not convert between units silently (in
 threshold mode the stack is compared as a range, because the audit counts wire bytes and the
-threshold is counted in what the client sends to the API — measured at 0.20×–1.92× across 23
+threshold is counted in what the client sends to the API — measured at 0.20×–1.92× across 24
 servers); it will not claim a posture the machine did not state readably, which is four
 refusals and not one — when two places set the same variable to different values, when a
 settings file exists and cannot be read, when the place that would decide sets the variable
@@ -204,8 +204,8 @@ INCREASE FAIL:
 Add `--claude` to annotate each server with its Anthropic-request cost from the published
 [Claude divergence](docs/METHODOLOGY.md#claude-divergence) run — an exact number when the
 published capture hash matches what you have installed, `—` (silence, not a stale guess)
-when it doesn't. The run holds 23 rows — the measured servers it covered when it last ran —
-and [results/leaderboard.md](results/leaderboard.md) prints a claude number for the 15 that
+when it doesn't. The run holds 24 rows — the measured servers it covered when it last ran —
+and [results/leaderboard.md](results/leaderboard.md) prints a claude number for the 16 that
 still match today and silence for the rest. Most installs will show a mix:
 
 ```
@@ -218,11 +218,11 @@ Add `--suggest` to place each of your tools in the measured set's tool-shape dis
 ([method](docs/METHODOLOGY.md#tool-shape)) and get advice only where the data can point at
 something. Only descriptions draw advice — schemas are functional surface; descriptions are
 prose every request carries — and only descriptions at or above the 90th percentile of the
-1,150 measured tools:
+1,430 measured tools:
 
 ```
   suggest — descriptions at or above the 90th percentile of measured tools
-  (baseline 2026-09-03: 1,150 tools across 81 measured servers):
+  (baseline 2026-09-05: 1,430 tools across 87 measured servers):
     stub · wordy — 345 tokens: description 321 (p92), schema 14
       rewriting the description toward the measured median (27) would recover ≈294 tokens on every request
     1 of 2 tools sit inside the distribution — no advice where nothing is measurably unusual.
@@ -256,7 +256,7 @@ Flags: `--json` (full report on stdout, progress on stderr), `--budget N`,
 
 The number `audit` gives you is the same measurement, run across a curated set of public
 servers — which is how you can tell it is a measurement and not this tool's opinion. It also
-shows what you are choosing between: across the 86 servers measured, cost spans **1,700×**,
+shows what you are choosing between: across the 87 servers measured, cost spans **1,700×**,
 from `postgres` at 32 tokens to `github` at 54,622. The table below is a
 sample of that range; the full range is in
 [results/leaderboard.md](results/leaderboard.md).
@@ -271,7 +271,7 @@ sample of that range; the full range is in
 | filesystem (reference) | 2,823 | 14 |
 | markitdown | 64 | 1 |
 
-*(86 of 106 popular servers measured, each row dated by its own most recent sweep — full table in
+*(87 of 106 popular servers measured, each row dated by its own most recent sweep — full table in
 [results/leaderboard.md](results/leaderboard.md); every failure is listed with its reason.
 Each measured server also has a [detail page](https://athakur3.github.io/mcp-context-cost/servers/)
 showing which tools its tokens are in.)*

package/dist/audit/deferral.d.ts

@@ -234,9 +234,15 @@ export interface WireToClientRatio {
     source: string;
 }
 /**
- * The band as published in this repository's own `results/divergence.json`
- * (claude-opus-5, 2026-08-19, 20 servers). Used when no divergence run was
- * supplied; `--claude` recomputes it from the run it fetched.
+ * The band as published in this repository's own `results/divergence.json`.
+ * Used when no divergence run was supplied; `--claude` recomputes it from the
+ * run it fetched.
+ *
+ * The model, the date and the count deliberately are not restated here. This
+ * docblock said `(claude-opus-5, 2026-08-19, 20 servers)` while the field below
+ * read 23 and the run on disk held 24 — prose beside a number, drifting from
+ * it, which is the same failure this constant's own guard exists to catch one
+ * level down. The fields are the record; `source` dates them.
  */
 export declare const PUBLISHED_WIRE_TO_CLIENT_RATIO: WireToClientRatio;
 /** Derive the band from a supplied divergence run, falling back to the published one. */

package/dist/audit/deferral.js

@@ -222,9 +222,15 @@ export function resolveToolSearchSources(sources) {
     return { ...resolved, source: resolved.readFromMachine ? (base?.source ?? null) : null };
 }
 /**
- * The band as published in this repository's own `results/divergence.json`
- * (claude-opus-5, 2026-08-19, 20 servers). Used when no divergence run was
- * supplied; `--claude` recomputes it from the run it fetched.
+ * The band as published in this repository's own `results/divergence.json`.
+ * Used when no divergence run was supplied; `--claude` recomputes it from the
+ * run it fetched.
+ *
+ * The model, the date and the count deliberately are not restated here. This
+ * docblock said `(claude-opus-5, 2026-08-19, 20 servers)` while the field below
+ * read 23 and the run on disk held 24 — prose beside a number, drifting from
+ * it, which is the same failure this constant's own guard exists to catch one
+ * level down. The fields are the record; `source` dates them.
  */
 export const PUBLISHED_WIRE_TO_CLIENT_RATIO = {
     low: 0.2,

package/dist/sweep/published-stats.d.ts

@@ -50,6 +50,18 @@ export interface PublishedStats {
         ratioMin: number;
         ratioMax: number;
     };
+    /**
+     * The published tool-shape baseline, which README quotes twice — once in
+     * prose and once inside the `--suggest` sample output. Both were hand-written
+     * and `STATIC_COUNTS` excused them as regen-maintained, which was not true of
+     * either: the file said 1,430 tools across 87 servers while the page said
+     * 1,150 across 81, and every test passed.
+     */
+    toolShape: {
+        toolCount: number;
+        serverCount: number;
+        generatedAt: string;
+    };
     deferralCostlierCount: number;
     /** Servers whose most recent cost movement went up / down (cost-regression/v1). */
     movement: {

package/dist/sweep/published-stats.js

@@ -25,7 +25,7 @@
  * never a silent skip), and only the slots are ever rewritten — spliced in
  * place, so the page's own line wrapping survives.
  */
-import { readFileSync, writeFileSync } from 'node:fs';
+import { existsSync, readFileSync, writeFileSync } from 'node:fs';
 import { join } from 'node:path';
 import { DEFAULT_CONTEXT_WINDOW } from '../audit/audit.js';
 import { fieldSelectionShare, isCurrent } from '../core/divergence.js';
@@ -128,7 +128,15 @@ export function computePublishedStats(entries, root = process.cwd()) {
     if (!githubRow?.m || typeof githubRow.m.totalTokens !== 'number') {
         throw new Error('README quotes `verify` on results/github, which has no current measurement');
     }
+    const shape = (() => {
+        const p = join(root, 'results', 'tool-shape.json');
+        if (!existsSync(p))
+            throw new Error('results/tool-shape.json is missing — README states its numbers');
+        const j = JSON.parse(readFileSync(p, 'utf8'));
+        return { toolCount: j.toolCount, serverCount: j.serverCount, generatedAt: j.generatedAt };
+    })();
     return {
+        toolShape: shape,
         candidateTotal: rows.length,
         measuredCount: measured.length,
         max,
@@ -225,6 +233,20 @@ export const PAGE_CLAIMS = [
         template: 'measured at {f}×–{f}× across {n} servers)',
         values: (s) => [s.claude.ratioMin.toFixed(2), s.claude.ratioMax.toFixed(2), fmt(s.claude.runSize)],
     },
+    {
+        file: 'README.md',
+        id: 'tool-shape:prose',
+        template: 'only descriptions at or above the 90th percentile of the {n} measured tools:',
+        values: (s) => [fmt(s.toolShape.toolCount)],
+    },
+    {
+        file: 'README.md',
+        id: 'tool-shape:sample-header',
+        // Inside the `--suggest` sample block. A reader compares their own output
+        // against it, so a stale baseline line there is read as a current one.
+        template: '(baseline {w}: {n} tools across {n} measured servers):',
+        values: (s) => [s.toolShape.generatedAt, fmt(s.toolShape.toolCount), fmt(s.toolShape.serverCount)],
+    },
     {
         file: 'README.md',
         id: 'repo-map:candidates',
@@ -317,6 +339,18 @@ export const PAGE_CLAIMS = [
             fmt(s.claude.widest.mapped),
         ],
     },
+    {
+        file: 'docs/METHODOLOGY.md',
+        id: 'divergence:band-parenthetical',
+        // The deferral section's own copy of the band. It was hand-written and held
+        // to `PUBLISHED_WIRE_TO_CLIENT_RATIO` by a test, which worked only while the
+        // constant tracked the run exactly — and the constant is a release-time
+        // snapshot that is deliberately allowed to lag. Two pages quoting two
+        // different sources for one number is the drift this file exists to end, so
+        // both pages state the run and the constant is guarded separately.
+        template: 'band ({f}×–{f}×\nacross {n} servers)',
+        values: (s) => [s.claude.ratioMin.toFixed(2), s.claude.ratioMax.toFixed(2), fmt(s.claude.runSize)],
+    },
     {
         file: 'docs/METHODOLOGY.md',
         id: 'divergence:ratio-range',

package/dist/sweep/regressions.js

@@ -103,9 +103,19 @@ export function writeCaptureIndex(entries, root = process.cwd()) {
     }
     for (const sha of ambiguous)
         delete captures[sha];
+    // Dated by the newest capture it indexes, not by the moment it was written —
+    // the same rule the dashboard was moved to. A clock stamp makes the file
+    // claim to be as fresh as the run that regenerated it, when what a reader
+    // needs to know is how far the data behind it reaches. It also means
+    // regenerating over unchanged vectors produces no diff, which is what lets a
+    // release gate ask "is anything derived here stale?" and get a real answer.
+    const newest = Object.values(captures)
+        .map((c) => c.date)
+        .sort()
+        .pop();
     const index = {
         method: CAPTURE_INDEX_METHOD,
-        generatedAt: new Date().toISOString().slice(0, 10),
+        generatedAt: newest ?? '',
         // Key order sorted so a re-run over unchanged vectors produces no diff noise.
         captures: Object.fromEntries(Object.entries(captures).sort(([a], [b]) => a.localeCompare(b))),
         current: Object.fromEntries(Object.entries(current).sort(([a], [b]) => a.localeCompare(b))),

package/dist/sweep/tool-shape.js

@@ -11,6 +11,7 @@ import { buildToolShapeBaseline } from '../core/tool-shape.js';
 import { loadRows } from './report.js';
 export function writeToolShapeBaseline(entries, root = process.cwd()) {
     const tools = [];
+    const dates = [];
     let serverCount = 0;
     for (const r of loadRows(entries, root)) {
         if (!r.m || (r.m.status !== 'measured' && r.m.status !== 'dynamic'))
@@ -19,8 +20,20 @@ export function writeToolShapeBaseline(entries, root = process.cwd()) {
             continue;
         serverCount++;
         tools.push(...r.m.tools);
+        const day = String(r.m.measuredAt ?? '').slice(0, 10);
+        if (/^\d{4}-\d{2}-\d{2}$/.test(day))
+            dates.push(day);
     }
-    const baseline = buildToolShapeBaseline(tools, { serverCount, methodologyVersion: METHODOLOGY_VERSION });
+    // Dated by the newest measurement it summarises rather than by the clock, so
+    // the file says how far its data reaches and a re-derivation over unchanged
+    // measurements produces no diff. `buildToolShapeBaseline` still defaults to
+    // today when nobody supplies one, for callers with no measurement dates.
+    const generatedAt = dates.sort().pop();
+    const baseline = buildToolShapeBaseline(tools, {
+        serverCount,
+        methodologyVersion: METHODOLOGY_VERSION,
+        ...(generatedAt ? { generatedAt } : {}),
+    });
     writeFileSync(join(root, 'results', 'tool-shape.json'), JSON.stringify(baseline, null, 2) + '\n');
     return baseline;
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "mcp-context-cost",
-  "version": "0.13.0",
+  "version": "0.13.1",
   "description": "Measure what your MCP servers cost in context tokens — audit your own config, or badge the server you publish",
   "type": "module",
   "license": "MIT",
@@ -43,7 +43,9 @@
     "session-start": "tsx src/sweep/session-start.ts",
     "sweep": "tsx src/sweep/run.ts",
     "sweep:all": "tsx src/sweep/sweep-all.ts",
-    "verify": "tsx src/cli.ts verify"
+    "verify": "tsx src/cli.ts verify",
+    "check:release": "tsx tools/release-readiness.ts",
+    "cut": "tsx tools/cut-changelog.ts"
   },
   "engines": {
     "node": ">=20"