the-high-council 0.6.0

package/chains/plan-unanimous.json

@@ -0,0 +1,82 @@
+{
+  "name": "plan-unanimous",
+  "description": "Same seats as plan-cheap, but every critic must independently sign off on the SAME draft before it's done (his call, 2026-09-07) - not 'first lab to pass it, stop'. Each round is a full panel pass (all 5 critics), so 3 rounds here means up to 15 critic calls, not 5. A round that isn't unanimous revises against the union of every objection any lab raised, then the whole panel reviews again.",
+  "maxRounds": 3,
+  "signoff": "unanimous",
+  "estimate": {
+    "promptTokens": 16000,
+    "draftTokens": 16000,
+    "critiqueTokens": 2000
+  },
+  "seats": {
+    "criteria": {
+      "provider": "anthropic",
+      "model": "claude-sonnet-5",
+      "maxTokens": 8000,
+      "extra": {
+        "thinking": {
+          "type": "disabled"
+        }
+      }
+    },
+    "builder": {
+      "provider": "anthropic",
+      "model": "claude-sonnet-5",
+      "maxTokens": 40000,
+      "extra": {
+        "thinking": {
+          "type": "disabled"
+        }
+      }
+    },
+    "reviser": {
+      "provider": "anthropic",
+      "model": "claude-sonnet-5",
+      "maxTokens": 40000,
+      "extra": {
+        "thinking": {
+          "type": "disabled"
+        }
+      }
+    },
+    "critics": [
+      {
+        "provider": "together",
+        "model": "Qwen/Qwen3.5-9B",
+        "maxTokens": 20000,
+        "extra": {
+          "chat_template_kwargs": {
+            "enable_thinking": false
+          }
+        },
+        "lab": "qwen"
+      },
+      {
+        "provider": "zai",
+        "model": "glm-4.7-flash",
+        "maxTokens": 20000,
+        "extra": {
+          "thinking": {
+            "type": "disabled"
+          }
+        }
+      },
+      {
+        "provider": "cohere",
+        "model": "command-r7b-12-2024",
+        "maxTokens": 4096
+      },
+      {
+        "provider": "google",
+        "model": "gemini-3.6-flash",
+        "maxTokens": 20000
+      },
+      {
+        "provider": "openrouter",
+        "model": "meta-llama/llama-3.3-70b-instruct",
+        "maxTokens": 20000,
+        "lab": "meta"
+      }
+    ]
+  }
+}

package/chains/seven-cheap.json

@@ -0,0 +1,67 @@
+{
+  "name": "seven-cheap",
+  "description": "Six labs, each lab's cheap model, one seat per lab: Qwen, GLM, Cohere, Google and Llama on the panel, Anthropic in the builder seat. (Named for the seven-lab shape it was cut down from - the roster below is the real one.) Builder and reviser stay on the cheap Anthropic seat because the deliverable here is a plan/prompt, not code - the actual build happens afterward in Claude Code on the existing subscription, not through this chain. No finalist seat for the same reason: the human takes the plan to Claude Code themselves.",
+  "maxRounds": 5,
+  "stopOnPass": true,
+  "estimate": {
+    "promptTokens": 4000,
+    "draftTokens": 6000,
+    "critiqueTokens": 1200
+  },
+  "seats": {
+    "criteria": {
+      "provider": "anthropic",
+      "model": "claude-haiku-4-5-20251001",
+      "maxTokens": 2000
+    },
+    "builder": {
+      "provider": "anthropic",
+      "model": "claude-haiku-4-5-20251001",
+      "maxTokens": 8000
+    },
+    "reviser": {
+      "provider": "anthropic",
+      "model": "claude-haiku-4-5-20251001",
+      "maxTokens": 8000
+    },
+    "critics": [
+      {
+        "provider": "together",
+        "model": "Qwen/Qwen3.5-9B",
+        "maxTokens": 20000,
+        "extra": {
+          "chat_template_kwargs": {
+            "enable_thinking": false
+          }
+        },
+        "lab": "qwen"
+      },
+      {
+        "provider": "zai",
+        "model": "glm-4.7-flash",
+        "maxTokens": 20000,
+        "extra": {
+          "thinking": {
+            "type": "disabled"
+          }
+        }
+      },
+      {
+        "provider": "cohere",
+        "model": "command-r7b-12-2024",
+        "maxTokens": 4096
+      },
+      {
+        "provider": "google",
+        "model": "gemini-3.6-flash",
+        "maxTokens": 20000
+      },
+      {
+        "provider": "openrouter",
+        "model": "meta-llama/llama-3.3-70b-instruct",
+        "maxTokens": 20000,
+        "lab": "meta"
+      }
+    ]
+  }
+}

package/chains/seven.json

@@ -0,0 +1,60 @@
+{
+  "name": "seven",
+  "description": "Seven frontier labs, one seat each, as originally sketched. Kept so the cost and the benefit can be measured against the two-lab chain rather than argued about.",
+  "maxRounds": 5,
+  "stopOnPass": true,
+  "estimate": {
+    "promptTokens": 4000,
+    "draftTokens": 6000,
+    "critiqueTokens": 1200
+  },
+  "seats": {
+    "criteria": {
+      "provider": "anthropic",
+      "model": "claude-opus-5",
+      "maxTokens": 2000
+    },
+    "builder": {
+      "provider": "anthropic",
+      "model": "claude-opus-5",
+      "maxTokens": 16000
+    },
+    "reviser": {
+      "provider": "anthropic",
+      "model": "claude-opus-5",
+      "maxTokens": 16000
+    },
+    "critics": [
+      {
+        "provider": "openai",
+        "model": "gpt-5",
+        "maxTokens": 20000
+      },
+      {
+        "provider": "google",
+        "model": "gemini-2.5-pro",
+        "maxTokens": 20000
+      },
+      {
+        "provider": "xai",
+        "model": "grok-4",
+        "maxTokens": 20000
+      },
+      {
+        "provider": "deepseek",
+        "model": "deepseek-chat",
+        "maxTokens": 20000
+      },
+      {
+        "provider": "groq",
+        "model": "llama-3.3-70b-versatile",
+        "maxTokens": 20000
+      }
+    ],
+    "finalist": {
+      "provider": "anthropic",
+      "model": "claude-opus-5",
+      "maxTokens": 16000
+    }
+  }
+}

package/chains/smo-v2-cheap.json

@@ -0,0 +1,67 @@
+{
+  "name": "smo-v2-cheap",
+  "description": "Long-form building-plan pass for SMO v2. Builder/reviser/criteria upgraded from Haiku to Sonnet 5 (his call - this is the seat that actually writes the plan, worth paying for). 5 critic seats, cheap-tier, no xAI/Grok. Mistral parked (2026-09-07: account-side 429 across three keys/workspaces, unresolved on their end) - add back once it clears, nothing else needs to change. Criteria are generated fresh, not hand-written, because this is a new plan, not a description of existing SMO.",
+  "maxRounds": 5,
+  "stopOnPass": true,
+  "estimate": {
+    "promptTokens": 16000,
+    "draftTokens": 12000,
+    "critiqueTokens": 2000
+  },
+  "seats": {
+    "criteria": {
+      "provider": "anthropic",
+      "model": "claude-sonnet-5",
+      "maxTokens": 2000
+    },
+    "builder": {
+      "provider": "anthropic",
+      "model": "claude-sonnet-5",
+      "maxTokens": 24000
+    },
+    "reviser": {
+      "provider": "anthropic",
+      "model": "claude-sonnet-5",
+      "maxTokens": 24000
+    },
+    "critics": [
+      {
+        "provider": "together",
+        "model": "Qwen/Qwen3.5-9B",
+        "maxTokens": 20000,
+        "extra": {
+          "chat_template_kwargs": {
+            "enable_thinking": false
+          }
+        },
+        "lab": "qwen"
+      },
+      {
+        "provider": "zai",
+        "model": "glm-4.7-flash",
+        "maxTokens": 20000,
+        "extra": {
+          "thinking": {
+            "type": "disabled"
+          }
+        }
+      },
+      {
+        "provider": "cohere",
+        "model": "command-r7b-12-2024",
+        "maxTokens": 4096
+      },
+      {
+        "provider": "google",
+        "model": "gemini-3.6-flash",
+        "maxTokens": 20000
+      },
+      {
+        "provider": "openrouter",
+        "model": "meta-llama/llama-3.3-70b-instruct",
+        "maxTokens": 20000,
+        "lab": "meta"
+      }
+    ]
+  }
+}

package/chains/smo.json

@@ -0,0 +1,50 @@
+{
+  "name": "smo",
+  "description": "The SMO vertical-slice test run. Same seats as 'verify', but the acceptance criteria are hand-written rather than generated - see Docs/SmoTestRun.md section 3.",
+  "maxRounds": 2,
+  "stopOnPass": true,
+  "estimate": {
+    "promptTokens": 16000,
+    "draftTokens": 9000,
+    "critiqueTokens": 1500
+  },
+  "seats": {
+    "criteria": {
+      "provider": "anthropic",
+      "model": "claude-haiku-4-5-20251001",
+      "maxTokens": 1000
+    },
+    "builder": {
+      "provider": "anthropic",
+      "model": "claude-opus-5",
+      "maxTokens": 24000
+    },
+    "reviser": {
+      "provider": "anthropic",
+      "model": "claude-opus-5",
+      "maxTokens": 24000
+    },
+    "critics": [
+      {
+        "provider": "openai",
+        "model": "gpt-5",
+        "maxTokens": 20000
+      },
+      {
+        "provider": "google",
+        "model": "gemini-2.5-pro",
+        "maxTokens": 20000
+      }
+    ]
+  },
+  "criteria": [
+    "The deliverable is the production bible itself, not a plan or outline of one.",
+    "It covers exactly the named slice, and explicitly states what is out of scope.",
+    "Every factual claim about SMO is traceable to the source description, and gaps are marked 'not specified in source' rather than filled in.",
+    "Every design decision originating in the deliverable is marked as a proposal with a stated reason.",
+    "It contains an itemised, countable asset list.",
+    "It contains acceptance criteria that decide the slice is done.",
+    "It contains no schedule, timeline, headcount or cost estimate.",
+    "It contains no praise of the project and no pitch language."
+  ]
+}

package/chains/verify.json

@@ -0,0 +1,40 @@
+{
+  "name": "verify",
+  "description": "The recommended default. Two labs, a hard 2-round cap. Most of the measurable gain, a fraction of the cost.",
+  "maxRounds": 2,
+  "stopOnPass": true,
+  "estimate": {
+    "promptTokens": 4000,
+    "draftTokens": 6000,
+    "critiqueTokens": 1200
+  },
+  "seats": {
+    "criteria": {
+      "provider": "anthropic",
+      "model": "claude-opus-5",
+      "maxTokens": 2000
+    },
+    "builder": {
+      "provider": "anthropic",
+      "model": "claude-opus-5",
+      "maxTokens": 16000
+    },
+    "reviser": {
+      "provider": "anthropic",
+      "model": "claude-opus-5",
+      "maxTokens": 16000
+    },
+    "critics": [
+      {
+        "provider": "openai",
+        "model": "gpt-5",
+        "maxTokens": 20000
+      },
+      {
+        "provider": "google",
+        "model": "gemini-2.5-pro",
+        "maxTokens": 20000
+      }
+    ]
+  }
+}

package/package.json

@@ -0,0 +1,55 @@
+{
+  "name": "the-high-council",
+  "version": "0.6.0",
+  "description": "The High Council: a multi-model planning harness. One builder, independent critics from other labs, checkable acceptance criteria, a debate on the record, and a hard stop.",
+  "type": "module",
+  "license": "MIT",
+  "bin": {
+    "council": "src/cli.js"
+  },
+  "scripts": {
+    "run": "node src/cli.js",
+    "cost": "node src/cost.js",
+    "dry": "node src/cli.js --dry-run",
+    "ui": "node src/ui/server.js",
+    "mcp": "node src/mcp/server.js",
+    "http-server": "node src/http-server.js",
+    "test": "node --test"
+  },
+  "engines": {
+    "node": ">=20"
+  },
+  "dependencies": {
+    "@modelcontextprotocol/sdk": "^1.30.0",
+    "express": "^4.22.2",
+    "zod": "^4.6.4"
+  },
+  "files": [
+    "src/",
+    "chains/",
+    "README.md",
+    "TROUBLESHOOTING.md",
+    "CONTRIBUTING.md",
+    "PERSONAS.md",
+    "LICENSE",
+    ".env.example"
+  ],
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/muad-yasin/the-high-council-mcp.git"
+  },
+  "homepage": "https://muad-yasin.github.io/the-high-council-mcp/",
+  "bugs": {
+    "url": "https://github.com/muad-yasin/the-high-council-mcp/issues"
+  },
+  "keywords": [
+    "mcp",
+    "model-context-protocol",
+    "llm",
+    "multi-agent",
+    "planning",
+    "claude",
+    "ai",
+    "cli"
+  ]
+}

package/src/board-export.js

@@ -0,0 +1,77 @@
+// v5 §1 candidate 8: a run's proposals, debate, replies and verdict as one
+// self-contained HTML file - no external assets, no template-engine
+// dependency, no server. Same "interactivity without JavaScript" discipline
+// docs/demo.html already uses: collapsible sections are plain
+// <details>/<summary>, inline CSS only.
+const esc = s => String(s ?? '')
+  .replace(/&/g, '&amp;').replace(/</g, '&lt;').replace(/>/g, '&gt;').replace(/"/g, '&quot;');
+
+function proposalSection(p, posts, replies) {
+  const on = posts.filter(x => x.on === p.id);
+  const re = replies.filter(r => r.id === p.id);
+  const status = p.withdrawn
+    ? `WITHDRAWN by ${esc(p.lab)}${p.replaced_by ? ` in favour of ${esc(p.replaced_by)}` : ''}`
+    : p.amended ? 'AMENDED by its author after debate' : 'stands';
+  const boardItems = on.map(x => `<li>${esc(x.by)} - ${esc(x.stance)}${x.merge_with ? ` with ${esc(x.merge_with)}` : ''}: ${esc(x.text)}</li>`)
+    .concat(re.map(r => `<li>${esc(p.lab)} (author) - ${esc(r.action)}: ${esc(r.text)}</li>`))
+    .join('\n') || '<li>(no posts)</li>';
+
+  return `<details>
+<summary>${esc(p.id)} (${esc(p.lab)}) - ${status}</summary>
+<p><strong>Title:</strong> ${esc(p.title)}</p>
+<p><strong>Serves:</strong> ${esc(p.serves)}</p>
+<p><strong>What:</strong> ${esc(p.what)}</p>
+<p><strong>Why:</strong> ${esc(p.why)}</p>
+<p><strong>How:</strong> ${esc(p.how)}</p>
+<p><strong>Acceptance test:</strong> ${esc(p.acceptance_test)}</p>
+<p><strong>Board:</strong></p>
+<ul>
+${boardItems}
+</ul>
+</details>`;
+}
+
+/**
+ * A run's report.json as one self-contained HTML string: no external
+ * assets, no <script>, no server. Never throws on a partial report -
+ * missing proposals/debate/signoff just render as empty sections.
+ */
+export function renderBoardHtml(report) {
+  const proposals = report.proposals || [];
+  const posts = report.debate?.posts || [];
+  const replies = report.debate?.replies || [];
+  const signoff = report.signoff || [];
+
+  const proposalsHtml = proposals.length
+    ? proposals.map(p => proposalSection(p, posts, replies)).join('\n')
+    : '<p>No proposal debate ran this run.</p>';
+
+  const verdictItems = signoff.map(s =>
+    `<li>${esc(s.provider)}: ${s.signedOff === null ? 'abstained' : s.signedOff ? 'signed off' : 'objected'}</li>`).join('\n') || '<li>(no panel signoff recorded)</li>';
+
+  return `<!doctype html>
+<html lang="en">
+<head>
+<meta charset="utf-8">
+<title>Debate board${report.runId ? ` - ${esc(report.runId)}` : ''}</title>
+<style>
+  body { font-family: system-ui, sans-serif; max-width: 900px; margin: 2rem auto; padding: 0 1rem; color: #1a1a1a; }
+  summary { cursor: pointer; font-weight: 600; margin: 0.5rem 0; }
+  details { border: 1px solid #ddd; border-radius: 6px; padding: 0.5rem 1rem; margin-bottom: 0.75rem; }
+  h1, h2 { border-bottom: 1px solid #eee; padding-bottom: 0.25rem; }
+</style>
+</head>
+<body>
+<h1>Debate board${report.runId ? ` - ${esc(report.runId)}` : ''}</h1>
+<p>Chain: ${esc(report.chain || 'unknown')}</p>
+<h2>Verdict</h2>
+<p>${report.passed ? 'PASSED - every lab signed off' : 'OPEN OBJECTIONS'}</p>
+<ul>
+${verdictItems}
+</ul>
+<h2>Proposals</h2>
+${proposalsHtml}
+</body>
+</html>
+`;
+}

package/src/budget.js

@@ -0,0 +1,126 @@
+#!/usr/bin/env node
+// Prices a whole-run token budget across a roster of models.
+//
+//   node src/budget.js --per-model 10000000 --roster frontier --split 80
+//
+// Unlike --dry-run, which prices one pass through a chain from per-stage
+// assumptions, this answers the blunt question: if each seat chews through
+// N tokens over a whole run, what does the run cost?
+
+import { readFileSync } from 'node:fs';
+import { fileURLToPath } from 'node:url';
+import { dirname, join } from 'node:path';
+
+const here = dirname(fileURLToPath(import.meta.url));
+const PRICES = JSON.parse(readFileSync(join(here, 'pricing.json'), 'utf8'));
+
+export const ROSTERS = {
+  // Seven labs, top model from each. The seven-frontier-lab harness as asked for.
+  frontier: [
+    'anthropic/claude-opus-5',
+    'openai/gpt-5',
+    'google/gemini-2.5-pro',
+    'xai/grok-4',
+    'mistral/mistral-large-latest',
+    'deepseek/deepseek-reasoner',
+    'together/Qwen/Qwen2.5-72B-Instruct-Turbo',
+  ],
+  // The frontier roster with the Anthropic seat dropped from Opus 5 to
+  // Sonnet 5. Same seven labs; only the most expensive seat changes.
+  'frontier-sonnet': [
+    'anthropic/claude-sonnet-5',
+    'openai/gpt-5',
+    'google/gemini-2.5-pro',
+    'xai/grok-4',
+    'mistral/mistral-large-latest',
+    'deepseek/deepseek-reasoner',
+    'together/Qwen/Qwen2.5-72B-Instruct-Turbo',
+  ],
+  // Seven labs again, but each lab's small model. Same structure, same
+  // independence between labs, roughly a tenth of the bill.
+  cheap: [
+    'anthropic/claude-haiku-4-5-20251001',
+    'openai/gpt-5-mini',
+    'google/gemini-2.5-flash',
+    'deepseek/deepseek-chat',
+    'groq/llama-3.3-70b-versatile',
+    'mistral/mistral-small-latest',
+    'together/meta-llama/Llama-3.3-70B-Instruct-Turbo',
+  ],
+  // The shape worth actually running: one expensive builder, cheap critics.
+  // Output tokens are what cost money, and only the builder writes at length.
+  mixed: [
+    'anthropic/claude-sonnet-5',
+    'openai/gpt-5-mini',
+    'google/gemini-2.5-flash',
+    'deepseek/deepseek-chat',
+    'groq/llama-3.3-70b-versatile',
+    'mistral/mistral-small-latest',
+    'together/meta-llama/Llama-3.3-70B-Instruct-Turbo',
+  ],
+};
+
+// Repeated input is billed at a fraction of base price when a provider's
+// prompt cache hits. The multiplier differs per lab (Anthropic reads at 0.1x
+// and writes at 1.25x, OpenAI and Google land near 0.1x-0.25x). One number
+// here is a deliberate simplification; treat cached figures as a floor.
+const CACHE_READ_MULTIPLIER = 0.1;
+
+export function priceRoster({ perModel, roster, inputShare, cacheHitRate = 0 }) {
+  const rows = roster.map(key => {
+    const p = PRICES[key];
+    const input = perModel * inputShare;
+    const output = perModel * (1 - inputShare);
+    if (!p) return { key, input, output, usd: null };
+    const cached = input * cacheHitRate;
+    const fresh = input - cached;
+    const usd = (fresh / 1e6) * p.in
+              + (cached / 1e6) * p.in * CACHE_READ_MULTIPLIER
+              + (output / 1e6) * p.out;
+    return { key, input, output, usd, rate: p };
+  });
+  const total = rows.reduce((s, r) => s + (r.usd || 0), 0);
+  return { rows, total, tokens: perModel * roster.length };
+}
+
+const usd = n => n >= 100 ? `$${n.toFixed(0)}` : n >= 1 ? `$${n.toFixed(2)}` : `$${n.toFixed(4)}`;
+const M = n => `${(n / 1e6).toFixed(1)}M`;
+
+function table({ perModel, roster, inputShare, cacheHitRate, label }) {
+  const { rows, total, tokens } = priceRoster({ perModel, roster, inputShare, cacheHitRate });
+  const w = Math.max(...rows.map(r => r.key.length));
+  console.log(`\n${label}`);
+  console.log(`  ${M(perModel)} tokens per model, ${roster.length} models, ${M(tokens)} total`);
+  console.log(`  split ${Math.round(inputShare * 100)}% input / ${Math.round((1 - inputShare) * 100)}% output` +
+              (cacheHitRate ? `, ${Math.round(cacheHitRate * 100)}% of input served from prompt cache` : ', no prompt caching'));
+  console.log('');
+  for (const r of rows.sort((a, b) => (b.usd || 0) - (a.usd || 0))) {
+    const price = r.rate ? `${r.rate.in}/${r.rate.out}` : 'unpriced';
+    console.log(`  ${r.key.padEnd(w)}  ${price.padStart(12)} per Mtok   ${(r.usd === null ? 'unpriced' : usd(r.usd)).padStart(9)}`);
+  }
+  console.log(`  ${''.padEnd(w)}  ${''.padStart(12)}              ${usd(total).padStart(9)}  per run`);
+  return total;
+}
+
+if (import.meta.url === `file://${process.argv[1]}`) {
+  const argv = process.argv.slice(2);
+  const arg = (n, d) => { const i = argv.indexOf(`--${n}`); return i === -1 ? d : argv[i + 1]; };
+  const perModel = Number(arg('per-model', 10_000_000));
+  const share = Number(arg('split', 80)) / 100;
+  const cache = Number(arg('cache', 0)) / 100;
+  const which = arg('roster', null);
+
+  const names = which ? [which] : Object.keys(ROSTERS);
+  const totals = {};
+  for (const n of names) {
+    totals[n] = table({
+      perModel, roster: ROSTERS[n], inputShare: share, cacheHitRate: cache,
+      label: `ROSTER: ${n}`,
+    });
+  }
+  if (names.length > 1) {
+    console.log('\nRatio to the cheapest roster:');
+    const min = Math.min(...Object.values(totals));
+    for (const [n, t] of Object.entries(totals)) console.log(`  ${n.padEnd(9)} ${usd(t).padStart(9)}   ${(t / min).toFixed(1)}x`);
+  }
+}

package/src/buys.js

@@ -0,0 +1,44 @@
+#!/usr/bin/env node
+// "What does $N actually buy?" - the inverse of budget.js.
+//
+// budget.js answers: given a token budget, what does it cost.
+// This answers: given money, how much real work is that.
+//
+//   node src/buys.js --usd 43 --roster cheap --split 90
+
+import { ROSTERS, priceRoster } from './budget.js';
+
+// Concrete units, so a token count means something.
+const UNITS = {
+  'page of prose (~500 words)': 670,
+  'long design document': 12_000,
+  'one deliverable through the 2-lab verify chain (worst case)': 54_400,
+  'one SMO-style production bible through the verify chain': 130_000,
+  'a full 10-request blind benchmark (4 outputs each)': 130_000 * 40,
+};
+
+const argv = process.argv.slice(2);
+const arg = (n, d) => { const i = argv.indexOf(`--${n}`); return i === -1 ? d : argv[i + 1]; };
+const usdBudget = Number(arg('usd', 43));
+const rosterName = arg('roster', 'cheap');
+const share = Number(arg('split', 90)) / 100;
+const roster = ROSTERS[rosterName];
+
+// Price 1M tokens per seat to get the roster's blended rate.
+const probe = priceRoster({ perModel: 1e6, roster, inputShare: share });
+const usdPerMtokTotal = probe.total / (probe.tokens / 1e6);
+
+const tokens = (usdBudget / usdPerMtokTotal) * 1e6;
+
+console.log(`\nRoster: ${rosterName} (${roster.length} models, ${Math.round(share * 100)}% input / ${Math.round((1 - share) * 100)}% output)`);
+console.log(`Blended rate: $${usdPerMtokTotal.toFixed(3)} per million tokens across the whole roster\n`);
+console.log(`$${usdBudget} buys ${(tokens / 1e6).toFixed(1)}M tokens of total traffic.\n`);
+console.log(`That is, in units you can picture:\n`);
+const w = Math.max(...Object.keys(UNITS).map(k => k.length));
+for (const [label, cost] of Object.entries(UNITS)) {
+  const n = tokens / cost;
+  // Force en-US grouping; the default locale turns 5823 into "5.823".
+  const shown = n >= 100 ? Math.round(n).toLocaleString('en-US') : n >= 1 ? n.toFixed(1) : n.toFixed(2);
+  console.log(`  ${label.padEnd(w)}  ${String(shown).padStart(9)}`);
+}
+console.log(`\nOr per unit: one deliverable through the verify chain costs $${(usdPerMtokTotal * 54_400 / 1e6).toFixed(3)} on this roster.`);