Переглянути джерело

test(agent-eval): price codegraph's fixed context cost alongside its residual (CG-7)

The first request's prompt is system + tool schemas + the question, before any
tool has answered, so differencing the arms' ctxBase prices what codegraph
occupies whether or not the agent ever calls it. Measured on gin: +775 tokens,
small because the tool is deferred -- only its name is in the initial listing.
Colby McHenry 1 місяць тому
батько
коміт
9b4df2133b

+ 8 - 1
scripts/agent-eval/parse-bench-readme.mjs

@@ -45,6 +45,7 @@ function parse(dir, label) {
     tokens: s.processed, cost: s.cost, raced: s.raced, turns: s.turns,
     segments: files.length,
     ctx: o.ctxFinal,
+    ctxBase: o.ctxBase,
     occCg: o.residual.codegraph,
     occFile: o.residualFileAccess,
     // The arm's own retrieval residual: codegraph in the with-arm, Read/Grep/Bash
@@ -107,10 +108,11 @@ console.log(`\n\nRESIDUAL CONTEXT OCCUPANCY — retrieval tokens still in the wi
 console.log(`(WITH = codegraph responses · WITHOUT = Read + Grep/Glob + Bash responses)`);
 console.log(`${anyMulti ? 'multi-turn sessions' : 'SINGLE-TURN sessions — see the caveat below'}\n`);
 console.log('repo        turns   final ctx W→WO        residual W→WO         % of ctx W→WO     % of window W→WO');
-const occ = { resid: [], shareCtx: [] };
+const occ = { resid: [], shareCtx: [], fixed: [] };
 for (const { repo, W, WO } of rows) {
   if (!W.length || !WO.length) { console.log(`${repo.padEnd(11)} (incomplete)`); continue; }
   const m = (arr, k) => median(arr.map(x => x[k]));
+  occ.fixed.push(m(W, 'ctxBase') - m(WO, 'ctxBase'));
   const wR = m(W, 'occSelf'), woR = m(WO, 'occSelf');
   const wCtx = m(W, 'ctx'), woCtx = m(WO, 'ctx');
   const wSc = m(W, 'occShareCtx'), woSc = m(WO, 'occShareCtx');
@@ -125,6 +127,11 @@ for (const { repo, W, WO } of rows) {
   );
 }
 console.log(`\nAVERAGE: retrieval residual ${avg(occ.resid)}% lower with codegraph  ·  share-of-context ${avg(occ.shareCtx)}% lower`);
+console.log(
+  `FIXED overhead: codegraph's tool schema + MCP instructions cost ${avg(occ.fixed) >= 0 ? '+' : ''}${avg(occ.fixed)} tok\n` +
+  `  of context before any tool is called (median WITH ctxBase - median WITHOUT ctxBase, averaged\n` +
+  `  over repos). It is paid whether or not the agent ever calls codegraph.`
+);
 if (!anyMulti) {
   console.log(
     `\nCAVEAT: every row above is a SINGLE-turn session, so the residual is measured at the\n` +

+ 7 - 1
scripts/agent-eval/parse-run.mjs

@@ -231,6 +231,11 @@ export function parseSession(files) {
   for (const q of queue) residual[q.family] += q.tokens;
 
   const ctxFinal = reqIdx.length ? timeline[reqIdx[reqIdx.length - 1]].ctx : 0;
+  // The FIRST request's prompt is system + tool schemas + the question, before
+  // any tool has answered. Differencing the arms' ctxBase prices codegraph's
+  // FIXED occupancy — its tool schema and MCP `initialize` instructions — which
+  // it pays whether or not the agent ever calls it.
+  const ctxBase = reqIdx.length ? timeline[reqIdx[0]].ctx : 0;
   // Multi-turn: duration/cost/tokens are per-segment, so sum them. `result.usage`
   // is cumulative WITHIN a segment (verified: its in+cache+out equals the sum of
   // that segment's per-request prompts), so summing segments is correct and does
@@ -252,7 +257,7 @@ export function parseSession(files) {
     cost: results.reduce((s, r) => s + (r.total_cost_usd || 0), 0),
     processed,
     occupancy: {
-      ctxFinal, windowTokens: WINDOW_TOKENS,
+      ctxFinal, ctxBase, windowTokens: WINDOW_TOKENS,
       charsPerToken, calibrated, compactions, evicted: Math.round(evicted),
       residual: Object.fromEntries(FAMILIES.map((f) => [f, Math.round(residual[f])])),
       contributed: Object.fromEntries(FAMILIES.map((f) => [f, Math.round(contributed[f])])),
@@ -287,6 +292,7 @@ export function formatOccupancy(s, indent = '  ') {
   row('other tools', o.residual.other, o.chars.other, o.results.other);
   const toolTotal = Object.values(o.residual).reduce((a, b) => a + b, 0);
   row('base (prompt+prose)', Math.max(0, o.ctxFinal - toolTotal));
+  out.push(`${indent}  ${'  of which fixed'.padEnd(18)}${(n(o.ctxBase) + ' tok').padStart(12)}  system + tool schemas + question, before any tool answered`);
   out.push(...rows);
   const dropped = o.contributed.codegraph + o.contributedFileAccess + o.contributed.other
     - (o.residual.codegraph + o.residualFileAccess + o.residual.other);