openzoo 0.48.36 → 0.48.38

package/lib/launch.js

@@ -177,9 +177,16 @@ export async function launchClaude(argv) {
         // ALWAYS shown, 4dp, even at 1.0000. Hiding it when there is no saving
         // reads as "no data"; printing 1.0000x says "measured, and it is level"
         // — which is the difference between an omission and a fact.
-        + 'const sx=Number(j.savingX);const sv=Number(j.savedUsd)||0;'
+        + 'const sx=Number(j.savingX);const sv=Number((j.spilled||{}).savedUsd ?? j.savedUsd)||0;'
         + 'const col=(sx>1)?"\\x1b[32m":"\\x1b[90m";'
-        + 'const save=(sx==null||!isFinite(sx))?"":("  \\u00b7 "+col+sx.toFixed(4)+"x vs direct"+(sv>0?(" ($"+sv.toFixed(4)+" saved)"):"")+"\\x1b[0m");'
+        // Prefer the SPILLED-call ratio: the session-wide one averages in every
+        // small turn that had nothing to offload, so it slides toward 1.0 as a
+        // conversation grows and reads as the mechanism failing when it is not.
+        + 'const sx2=Number(sp.savingX);'
+        + 'const on=(isFinite(sx2)&&sx2>0)?sx2:sx;'
+        + 'const lbl=(isFinite(sx2)&&sx2>0)?"x on spilled":"x vs direct";'
+        + 'const col2=(on>1)?"\\x1b[32m":"\\x1b[90m";'
+        + 'const save=(on==null||!isFinite(on))?"":("  \\u00b7 "+col2+on.toFixed(4)+lbl+(sv>0?(" ($"+sv.toFixed(4)+" saved)"):"")+"\\x1b[0m");'
         + 'process.stdout.write("\\x1b[38;5;208m\\u25cf\\x1b[0m openzoo  $"+(Number(j.spendUsd)||0).toFixed(4)+"  "+(j.paidCalls||0)+" call"+((j.paidCalls||0)===1?"":"s")+save+spill+cr+"  \\u00b7 x402")}'
         + 'catch{process.stdout.write("\\x1b[38;5;208m\\u25cf\\x1b[0m openzoo  \\u00b7 x402")}})\'\n');
       fs.chmodSync(scriptPath, 0o755);

package/lib/proxy.js

@@ -355,7 +355,11 @@ async function spillTranscript(body, log) {
   // in the bound corpus and comes back through recall.
   let tailStart = cut;
   {
-    const budget = Number(process.env.OPENZOO_TAIL_MAX_CHARS || 24000);
+    // 24000 was still too fat: MEASURED on 9 live spilled calls, direct came
+    // back identical to billed to the cent, i.e. the gateway never computed a
+    // counterfactual because promptTokens >= corpusTokens. The tail has to be
+    // small enough that the BOUND corpus is the bigger number.
+    const budget = Number(process.env.OPENZOO_TAIL_MAX_CHARS || 6000);
     let used = 0;
     for (let i = msgs.length - 1; i >= cut; i--) {
       used += msgText(msgs[i]).length;
@@ -554,6 +558,13 @@ export async function startProxy({ silent = false, requireToken = null, sessionM
   let spillCalls = 0;
   let spilledChars = 0;
   let spillReuses = 0;
+  // Spend/direct for ONLY the calls that spilled. The session-wide savingX
+  // averages these with every small turn that had nothing to offload, so it
+  // slides toward 1.0 as a conversation grows — which reads as the mechanism
+  // degrading when it is just the mix changing. OBSERVED: 1.3166 -> 1.1823
+  // while spilled calls and offloaded tokens both sat completely still.
+  let spillSpend = 0;
+  let spillDirect = 0;
   // CREDIT, CACHED. Users cannot tell prepaid credit from wallet balance and
   // have to guess whether a call was even paid for ("I don't think x402 made me
   // pay this at all"). The status line runs EVERY turn, so this is refreshed at
@@ -681,6 +692,10 @@ export async function startProxy({ silent = false, requireToken = null, sessionM
             // saving is actually made of.
             tokensApprox: Math.round(spilledChars / 4),
             reusedBinds: spillReuses,
+            spend: spillSpend,
+            direct: spillDirect,
+            savedUsd: Math.max(0, spillDirect - spillSpend),
+            savingX: spillSpend > 0 ? Number((spillDirect / spillSpend).toFixed(4)) : null,
           },
           spendUsd: sessionSpent,
           creditUsd,
@@ -995,9 +1010,11 @@ export async function startProxy({ silent = false, requireToken = null, sessionM
           }
           : init.headers,
       });
+      let didSpill = false;
       let result;
       if (cached) {
         spillCalls += 1;
+        didSpill = true;
         spilledChars += cached.corpus?.length || 0;
         if (cached.reused) spillReuses += 1;
         result = await send(cached.body, cached.contextId, cached.topK);
@@ -1039,6 +1056,10 @@ export async function startProxy({ silent = false, requireToken = null, sessionM
           // attach call that is the whole bound corpus, which is why it can be
           // orders of magnitude above what was billed. directUsd is exact and
           // always present; savesVsDirect is the same number as a ratio.
+          if (didSpill) {
+            spillSpend += receipt.billedUsd || 0;
+            spillDirect += typeof receipt.directUsd === 'number' ? receipt.directUsd : (receipt.billedUsd || 0);
+          }
           sessionDirect += typeof receipt.directUsd === 'number'
             ? receipt.directUsd
             : typeof receipt.savesVsDirect === 'number'
@@ -1084,6 +1105,10 @@ export async function startProxy({ silent = false, requireToken = null, sessionM
           sessionSpent += x.billedUsd;
           sessionCogs += typeof x.cogsUsd === 'number' ? x.cogsUsd : x.billedUsd / MARKUP;
           sessionDirect += typeof x.directUsd === 'number' ? x.directUsd : x.billedUsd;
+          if (didSpill) {
+            spillSpend += x.billedUsd;
+            spillDirect += typeof x.directUsd === 'number' ? x.directUsd : x.billedUsd;
+          }
           paidCalls += 1;
           if (viaTunnel) tunnelSpent += x.billedUsd;
           say(`credit -> $${x.billedUsd.toFixed(6)}  ·  session $${sessionSpent.toFixed(6)}`);

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "openzoo",
-  "version": "0.48.36",
+  "version": "0.48.38",
   "description": "Local x402-paying proxy + MCP server for openzoo.fun — point any OpenAI-compatible harness (Cursor, Claude Code, aider, SDKs) at localhost and it pays per call from a local burner wallet. Solana and Base rails live; Robinhood experimental.",
   "license": "MIT",
   "type": "module",