fix: no token ceiling unless the budget forces one

Removing the caps rather than tuning them. Every number I picked was a number I
invented, and 6000 was already tight enough to truncate a real replay article,
which is the failure that was called out when the cap first went in.

The coordinator now sends no max_tokens at all. The 402 handler supplies one only
when openrouter says the budget cannot cover an open ended request, so the
ceiling exists exactly when it has to and never otherwise. Verified unbounded is
accepted against the live key before making this the default.

The caps on the signal, augor, consolidation and graph workers are gone too. They
were added to work around an empty account, not because any of them ever produced
too much, and unlike the coordinator none of them detect truncation, so an
invented ceiling there risked silently corrupting company facts. A budget failure
in those is at least loud.

The 220 token cap in crawlerClassifier is left alone, it predates this and bounds
a genuinely tiny classification.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WnNxwxfXSbeNtjvtz5gayb

src/autonomy/llm.js

@@ -24,6 +24,9 @@ async function callCoordinator(config, prompt, options = {}) {
   const apiKey = String(config?.openRouter?.apiKey || '').trim();
   if (!apiKey) throw new Error('OpenRouter API key is not configured');
   const timeoutMs = Math.max(1000, Number(config?.openRouter?.timeoutMs || process.env.OPEN_ROUTER_TIMEOUT_MS) || 60000);
+  // only set when a budget retry forced one, or an operator asked for one
+  const maxTokens = options.maxTokens
+    || Number(config?.openRouter?.maxTokens || process.env.OPEN_ROUTER_MAX_TOKENS) || null;
   const controller = new AbortController();
   const timeout = setTimeout(() => controller.abort(), timeoutMs);
   let response;
@@ -35,15 +38,14 @@ async function callCoordinator(config, prompt, options = {}) {
       body: JSON.stringify({
         model: config.openRouter.llmModel,
         temperature: 0,
-        // Deliberately far above any output we have seen, because the cap exists to
-        // satisfy openrouter's affordability check, not to ration tokens: you are
-        // billed for what is used, not what is reserved. 6000 was tight enough to
-        // truncate a real replay article. The 402 handler below walks this down
-        // automatically when the budget cannot cover it, so a high ceiling costs
-        // nothing while funded and degrades on its own when not.
-        max_tokens: options.maxTokens
-          || Math.max(512, Number(config?.openRouter?.maxTokens || process.env.OPEN_ROUTER_MAX_TOKENS) || 32000),
         response_format: { type: 'json_object' },
+        // No ceiling by default. Any number we pick is a number we invented, and
+        // 6000 was already tight enough to truncate a real replay article. A cap
+        // only exists to satisfy openrouter's affordability reservation, so it is
+        // supplied by the 402 handler below when the budget genuinely cannot cover
+        // an open ended request, and never otherwise. Set OPEN_ROUTER_MAX_TOKENS
+        // if you ever want one imposed deliberately.
+        ...(maxTokens ? { max_tokens: maxTokens } : {}),
         messages: [
           { role: 'system', content: 'You are a coordinator. Extract only evidence-backed categorical hypotheses. Never output probabilities, expected returns, confidence scores, position sizes, or trade actions.' },
           { role: 'user', content: prompt },

workers/augorWorker.js

@@ -306,9 +306,6 @@ async function callLlm(llmConfig, prompt) {
     model: llmConfig.llmModel || llmConfig.model,
     messages: [{ role: "user", content: prompt }],
     temperature: 0.1,
-    // Unbounded requests get a 402 for reserving the model's whole output
-    // window against the remaining key budget, before running anything.
-    max_tokens: Math.max(512, Number(process.env.OPEN_ROUTER_MAX_TOKENS) || 6000),
   });
 
   const url = new URL("https://openrouter.ai/api/v1/chat/completions");

workers/consolidationWorker.js

@@ -220,9 +220,6 @@ Rules:
     model: llmConfig.llmModel || llmConfig.model,
     messages: [{ role: "user", content: prompt }],
     temperature: 0.1,
-    // Unbounded requests get a 402 for reserving the model's whole output
-    // window against the remaining key budget, before running anything.
-    max_tokens: Math.max(512, Number(process.env.OPEN_ROUTER_MAX_TOKENS) || 6000),
   });
 
   const url = new URL("https://openrouter.ai/api/v1/chat/completions");

workers/graphWorker.js

@@ -101,10 +101,6 @@ Reply with just the number of the match, or "none" if none apply. No explanation
     model: llmConfig.cheapModel || llmConfig.llmModel || llmConfig.model,
     messages: [{ role: "user", content: prompt }],
     temperature: 0,
-    // Same reason as the coordinator: an unbounded request gets a 402 for reserving
-    // the whole output window. This one only ever replies with a number or "none",
-    // the headroom is for models that think before answering.
-    max_tokens: Math.max(256, Number(process.env.OPEN_ROUTER_GRAPH_MAX_TOKENS) || 2000),
   });
 
   if (Date.now() < llmCooldownUntil) return null;

workers/signalWorker.js

@@ -297,9 +297,6 @@ async function callLlm(llmConfig, prompt) {
     model: llmConfig.llmModel || llmConfig.model,
     messages: [{ role: "user", content: prompt }],
     temperature: 0.1,
-    // Unbounded requests get a 402 for reserving the model's whole output
-    // window against the remaining key budget, before running anything.
-    max_tokens: Math.max(512, Number(process.env.OPEN_ROUTER_MAX_TOKENS) || 4000),
   });
 
   const url = new URL("https://openrouter.ai/api/v1/chat/completions");