# DCZion

> DCZion is team decision-making software (SaaS) for management, organizational and strategic calls: team members score options blindly with confidence ranges, and a Monte Carlo + Condorcet engine turns those judgments into one clear, mathematically verified recommendation with win odds, margin of victory, and disagreement diagnostics.

## What teams decide with it

DCZion is used for any decision that needs more than one person's judgment, and that has to still look defensible later:

- **Hiring and promotion:** executive and lead hiring, final-round candidates, internal promotion, restructuring. Interviewers score candidates independently against the same criteria.
- **Performance evaluation and remuneration:** quarterly and annual reviews, calibration across teams, promotion, raises and bonuses. Everyone on the list — the person running the review included — is scored blind on the same criteria, and a member's own row is their self-evaluation.
- **OKRs and planning:** OKR prioritization, what to fund and what to cut, headcount and budget allocation. Competing objectives are weighted by reversibility, urgency and confidence.
- **Strategy and direction:** market entry, positioning, build versus buy or partner, pivots, go / no-go on a large bet. Pairwise (Condorcet) comparison shows which direction beats every alternative head-to-head.
- **Product and engineering:** roadmap bets, architecture and stack choices, platform migrations, technical debt. Disagreement diagnostics isolate the criterion the team actually disagrees on.
- **Operations, vendors and risk:** vendor and tool selection, procurement, deal approvals, security and compliance sign-off. Cross-functional stakeholders score independently, and the decision is recorded with an auditable trail.

[Worked examples](/use-cases) walks through 22 of these across the six families, each showing the inputs, the result, and the arithmetic that connects them. The wizard ships 15 pre-calibrated templates covering the same ground.

## What DCZion is

- **Blind scoring:** every member rates alternatives privately (as ranges, e.g. 7–9) before seeing anyone else's scores, eliminating anchoring, rank, and loudest-voice bias.
- **Monte Carlo simulation:** 1k–100k draws propagate each member's uncertainty into the result — you get probabilities, not false precision.
- **Condorcet winner:** every option is compared pairwise against every rival; the alternative that beats all others in a majority of simulations is the recommendation.
- **Disagreement diagnostics:** per-criterion dispersion, agreement %, and score spread pinpoint exactly where the team diverges, so meetings focus on the one or two contested criteria.
- **Sensitivity / fragility analysis:** test what-if score shifts live to see what could flip the winner.
- **Exports:** reproducible PDF briefs, JSON data, and decision journals for audit.

## Key pages

Paths below are relative to this file's own origin, so they stay correct on
every deployment without carrying a domain in the repository.

- [Home](/): product overview, live result preview, templates, methodology, and a knowledge base on team decision-making research.
- [Pricing](/pricing): Free (10 solo + 10 team decisions/month per workspace, 5 workspaces, 5 seats), Team seat classes — 10 / 50 / 100 / 250 seats per workspace at $200 / $1,000 / $2,000 / $5,000 monthly (17% off billed annually), each with 100 decisions a month and workspaces to match (15 on Team 10, then one per seat) — Enterprise (custom, unlimited).
- [Learn — the science](/app/learn): how peer-reviewed research (affect asymmetry, shared mental models, organizational decision frameworks) maps to GroupConsensus features, plus a curated source library.
- [Start a decision](/app/start): entry point for creating a team decision or a solo decision.
- [Use cases](/use-cases): 22 worked examples of what teams decide with DCZion — hiring and promotion, performance and remuneration review, OKRs and planning, strategy, product and engineering, operations and vendors — with the inputs, the numbers, and how each conclusion was reached.

## API & MCP

An AI agent can analyze decisions directly without an account:

- **MCP server (recommended):** connect any MCP client to `/mcp` (Streamable HTTP, no install). Exposed tools:
  - `run_decision_analysis` — given alternatives and weighted criteria (the payload calls this array `factors`; performances as numbers or uncertainty ranges), returns the recommended choice, full ranking, pairwise win matrix, per-criterion averages, and a plain-language recommendation.
  - `run_sensitivity_analysis` — what-if analysis forcing one criterion's scores to see how the winner changes.
- **REST API:** `POST /api/analyze` with `{ decisionName, alternatives, factors, numSimulations?, seed?, framework? }` returns the same analysis as JSON. `POST /api/sensitivity` for what-if runs. `numSimulations` is optional and bounded to 1000–100000 (default 10000). See [openapi.json](/openapi.json).
- **Health:** `GET /api/health`.

Example `run_decision_analysis` input:

```json
{
  "decisionName": "Cloud provider",
  "alternatives": ["AWS", "GCP", "Azure"],
  "factors": [
    {
      "name": "Cost",
      "importance": 9,
      "performances": { "AWS": 6, "GCP": 8, "Azure": 7 }
    },
    {
      "name": "Reliability",
      "importance": 8,
      "performances": { "AWS": { "min": 8, "max": 10 }, "GCP": { "min": 6, "max": 9 }, "Azure": { "min": 6, "max": 8 } }
    }
  ],
  "numSimulations": 10000
}
```

## Notes for agents

- The MCP tools and `/api/analyze` / `/api/sensitivity` are stateless — they run the engine and return results without persisting anything or requiring an account. They are reachable without a session (no auth header); everything else under `/api/` requires signing in while the app is in private beta.
- Persisting a decision to a user account requires signing up (email/password via Supabase Auth) and is not exposed as an agent-friendly API today.
- `robots.txt` blocks general crawlers from `/api/` and `/admin146`; the MCP endpoint itself is a programmatic interface, not a web page.