openzoo 0.48.37 → 0.48.39

package/lib/launch.js

@@ -177,7 +177,7 @@ 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";'
         // 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

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;
@@ -690,6 +694,7 @@ export async function startProxy({ silent = false, requireToken = null, sessionM
             reusedBinds: spillReuses,
             spend: spillSpend,
             direct: spillDirect,
+            savedUsd: Math.max(0, spillDirect - spillSpend),
             savingX: spillSpend > 0 ? Number((spillDirect / spillSpend).toFixed(4)) : null,
           },
           spendUsd: sessionSpent,
@@ -1101,6 +1106,12 @@ export async function startProxy({ silent = false, requireToken = null, sessionM
           sessionCogs += typeof x.cogsUsd === 'number' ? x.cogsUsd : x.billedUsd / MARKUP;
           sessionDirect += typeof x.directUsd === 'number' ? x.directUsd : x.billedUsd;
           if (didSpill) {
+            // THE number that settles why a spilled call did or did not save:
+            // the gateway only prices a counterfactual when corpusTokens >
+            // promptTokens, so a tail that rivals the corpus silently falls
+            // back to markup and direct collapses onto billed.
+            const lc = x.lecore || {};
+            log(`spill priced: ${x.pricing} · corpus ${lc.corpusTokens ?? '?'} tok vs sent ${lc.tokensBefore ?? '?'} -> ${lc.tokensAfter ?? '?'} · billed ${(x.billedUsd ?? 0).toFixed(5)} direct ${(x.directUsd ?? 0).toFixed(5)}`);
             spillSpend += x.billedUsd;
             spillDirect += typeof x.directUsd === 'number' ? x.directUsd : x.billedUsd;
           }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "openzoo",
-  "version": "0.48.37",
+  "version": "0.48.39",
   "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",