# How Sessions Work

> Synths, modes, rounds, credits, synthesis — the concepts that explain what is actually happening when you run a SynthBoard session.

**Canonical URL:** https://www.synthboard.ai/docs/how-sessions-work  
**Markdown source:** https://www.synthboard.ai/docs/how-sessions-work.md

## A session in one paragraph

You bring a decision. You pick a mode. You pick 2–12 Synths from the 24-expert library (or Smart Setup picks for you). The board runs across 2–4 rounds: each round, every Synth contributes from their lane and reacts to the others. The synthesis layer maps consensus, surfaces minority opinions, and produces a recommendation. Credits are debited based on the models used and the depth of the run.

## Synths

A Synth is one of 24 expert AI personas in the library. Each runs on a 6-layer persona stack:

1. **Base prompt** — role, lane, and behavior baseline.
2. **7-dimensional DNA** — domain expertise dimensions specific to this Synth.
3. **OCEAN traits** — personality vector (openness, conscientiousness, extraversion, agreeableness, neuroticism) calibrated for this expert type.
4. **Cognitive framework** — first-principles, Bayesian, game-theoretic, systems, scenario-based, etc.
5. **Position-integrity rules** — anti-sycophancy guards. Synths hold positions under pressure rather than collapsing to consensus.
6. **Voice archetype** — register, vocabulary, and rhythm specific to this expert.

Every Synth is assigned to a model (OpenAI, Anthropic, or Google) based on which provider's strengths match the persona. This is multi-LLM routing — done automatically per Synth.

Tier limits on Synths-per-session: Free up to 3, Pro up to 5, Max up to 8, Ultra up to 12. Synth library size: Free sees 8, Pro sees 17, Max and Ultra see all 24.

## Modes

The mode shapes how the board behaves: how aggressive, how convergent, how exploratory, how time-boxed. The 10 modes:

| Mode | Display | What it does | Min tier |
|---|---|---|---|
| consultation | The Boardroom | Collaborative advice from complementary experts | Free |
| brainstorm | Idea Lab | Divergent ideation without early critique | Free |
| quick_take | Lightning Round | Fast, sharp opinions, <100 words each | Free |
| debate | The Arena | Adversarial exploration of opposing positions | Free |
| stress_test | Shark Tank | Ruthless attack on a plan to surface every flaw | Pro |
| decision | Command Center | Structured framework for choosing between options | Pro |
| war_room | War Room | Crisis triage with immediate action plans | Pro |
| deep_dive | Deep Dive | Thorough expert analysis with evidence | Max |
| forecast | Crystal Ball | Scenario planning with probability-weighted outcomes | Max |
| research | Research Lab | Evidence-first synthesis across disciplines | Max |

## Rounds

Most sessions run 2–4 rounds. Round 1 is "every Synth's first take." Round 2 is "respond to what you heard." Round 3+ is convergence and synthesis. You can intervene mid-round — pause, ask a follow-up, send a specific Synth back to debate harder.

Rounds are not chat turns. Each round happens in parallel across Synths, then synthesizes, then proceeds to the next.

## Synthesis

The synthesis layer is the system component that aggregates Synth output into a recommendation. It does not average. It maps:

- **Cluster of agreement** — what every Synth converged on.
- **Genuine disagreement** — where Synths held different positions, with reasoning preserved.
- **Strongest counter-argument** — the best version of the opposing case, named with its strongest defender.
- **Audit trail** — who said what across rounds, where positions shifted.
- **Recommendation** — the call the synthesis layer believes is most defensible given the deliberation.
- **Consensus score** — how aligned the board ended up. High consensus + strong reasoning = strong signal. Low consensus + strong reasoning = a genuine fork worth investigating.

## Credits

Sessions are credit-based and usage-driven. Cost depends on:
- Number of Synths.
- Models in play (standard vs premium vs frontier).
- Number of rounds.
- Length of context.

Typical ranges:
- Lightning Round, 4 Synths, standard models: 30–80 credits.
- The Boardroom, 5 Synths, premium models: 200–400 credits.
- Deep Dive, 6 Synths, frontier models: 600–1,200 credits.

You see the live cost estimate before you run. No metered billing surprises.

## Cross-session memory

Every session contributes to your private memory. Subsequent sessions reference past decisions, captured outcomes, and the assumptions you have been operating on. This means session #10 on a topic is structurally smarter than session #1 — the board knows your history.

Memory is private and tenant-isolated. Never shared, never used for model training.

## Outcome inference

When connected tools are linked (Stripe, Linear, Calendar, Gmail), the system *infers* outcomes from real signal — revenue movements, shipped tickets, follow-through events, conversation threads. These outcomes feed the Synths' evolution: they get better at *your* problems specifically, not generic best practice.

Outcome inference is opt-in per integration and explained on the integrations page.

## Actions

When a session yields an action (an email to send, a ticket to file, a calendar invite to push), it can be executed from inside the session — but only with your approval. Risk-tier actions require explicit confirmation. Every action is reversible inside 30 seconds.

## Related

- [Quickstart](https://www.synthboard.ai/docs/getting-started)
- [Pricing](https://www.synthboard.ai/docs/pricing)
- [Integrations](https://www.synthboard.ai/docs/integrations)
- [Glossary](https://www.synthboard.ai/glossary)

## How to cite this page

> SynthBoard — How sessions work. https://www.synthboard.ai/docs/how-sessions-work

Site: https://www.synthboard.ai