Clarity: The Decision Laboratory

The verdict

Why this matters

The pivot, the raise, the leap you keep postponing — these aren't information problems. ChatGPT gives you a hedged answer. Friends are biased toward your comfort. Coaches are expensive and slow. What's missing is a framework for thinking: multiple genuine perspectives, forced to take positions, reconciled into one verdict you can act on.

A single AI answer is an average. A deliberation takes positions. Clarity's design bet is that the structure of disagreement — not more data — is what produces conviction. The site says it plainly: conviction, not hedging.

The council — five agents, zero hedging

Six of Clarity's seven modes run a five-agent council in parallel, then a synthesizer reconciles them into a verdict: a conviction score (0–100), a risk inventory, an action plan, and an agent-alignment readout showing where the council agreed — and where it split.

• The Advocate — Builds the strongest honest case for the move. Its prompt explicitly bans the cop-out: "might work if everything goes perfectly" is not a position — it's a hedge dressed as advocacy.
• The Adversary — Precise, specific caution. Vague warnings ("this seems risky") are called out in its instructions as filler, not analysis.
• The Enhancer — Doesn't judge the plan — upgrades it, and flags the execution risks that could sink it during implementation.
• The Researcher — Facts only, with anti-fabrication rules: it may not embellish your own biography. Inflating "in this field since 2010" into "deep domain mastery" is treated as fabrication, same as a fake salary number.
• The Entrepreneur — Reframes the decision itself. Its key question interrogates whether you've even defined the right decision.

The flow — three minutes in, conviction out

• Describe your situation — Your words, no template — the decision, the tension, the thing you keep circling back to.
• Answer five questions — Non-negotiables, your fear, your definition of success in 12 months — so the council argues against your situation, not a generic proxy.
• The council deliberates — Five agents in parallel, each with an adversarial mandate and a banned list of cop-outs.
• Synthesis — One verdict reconciling the five — with a risk inventory and an action plan attached.
• Conviction score — 0–100, with an agent-alignment readout showing exactly where the council agreed and where it split.
• Follow-up — Push back on the verdict — the council answers in thread.

Seven rooms — six deliberations and a Vent

Each mode frames a different class of decision, and each seats its own lead agents: Career Pivot ("where the next chapter requires a different protagonist"), Inflection Point ("a fork that locks in more than it names"), Negotiation ("a single conversation with an unbounded afterlife"), Market Timing ("where waiting is also a position"), plus Side Hustle and Network Play.

The seventh, Vent, isn't a deliberation at all — it's a sequenced Listener → Reframer → Strategist flow for when you need the clearing that honest deliberation requires.

Two registers

The same council, two postures. Therapist mode is calm and contemplative — for decisions that produce anxiety, not clarity; you leave feeling held, not hectored. Coach mode is direct and kinetic — for when you know the direction and need the push and the plan.

Context depth — a council that learns you

• Sketch — First session — advice calibrated to what you shared today.
• Briefed — Intake complete — the council knows your non-negotiables, fears, and success definition, so advice fits the shape of your career, not just this decision.
• Witnessed — You've returned over time — the council has seen how you decide, reads patterns, and surfaces recurring tensions.

Privacy is engineered in

Decisions are confessions — people won't bring honest input to a system they don't trust. So the trust layer is structural, not a policy page:

• Per-user encryption — Every decision is encrypted with per-user keys at the application layer — not just at rest.
• Voice deleted in 60 seconds — Recordings are destroyed within a minute of transcription.
• Vent is sacred — Vent transcripts are treated as the most sensitive data in the system.
• Default-deny database — Row-level security on every table; a crisis-detection safety layer watches for moments that need more than a council.
• Public commitments — The promises are in writing at the Trust Center — not buried in a ToS.

How it's built

Next.js App Router + TypeScript frontend, Supabase Postgres with row-level security and per-user envelope encryption, Vercel behind a WAF, Claude doing the deliberating with prompt caching, server-side voice (Deepgram in, ElevenLabs out) on the top tier, Stripe for billing.

Tier	Price	What you get
Free	$0	3 decisions/mo · full five-agent council · every mode · both registers
Standard	$19/mo	10 decisions · decision history · pattern insights · follow-up threads
Premium	$39/mo	20 decisions · image & document upload · priority deliberation
Elite	$99/mo	150 decisions · voice intake & spoken verdicts · deepest personalization

What I'd research next

• Does the council actually disagree? — Measure the alignment distribution across real deliberations. If alignment always lands above 80%, the Adversary isn't earning its seat — the honest version of this product publishes its disagreement rate.
• Verdict calibration at 6–12 months — Did people who followed the synthesis report better outcomes than those who didn't? Conviction scores need a hit rate, same as any forecast.
• The register effect — Run the same situation through Therapist and Coach — does the register change the decision or just the experience? If verdicts diverge, that's signal.
• Witnessed-depth anchoring — Accumulated context sharpens advice — but when does "knows how you decide" become "anchors on how you used to decide"? Long-term memory needs a forgetting policy.

Where to go deeper

Bring a decision at clarity.rewiredminds.io — 3 free decisions a month, full council, no card required. The editorial home behind it is rewiredminds.io, and more dives like this live at jeremyknox.ai/deep-dives.