Decision Rehearsal · Enterprise Negotiation

Practice against
the person,not the scenario.

ArenaSynth builds synthetic decision-makers from a psychometric profile you can inspect. They hold their position under pressure and change it when you give them good reasons.

Rehearse it free →Enter the arena

// We start from the human on the other side of the table: a profile fixed before the conversation.

MC
Marcus Chen
VP PROCUREMENT · ENTERPRISE SAAS
SESSION
MarcusYour pricing is 40% over our current vendor. Give me one reason I put this in front of the committee.
YouWhat's a quarter of manual reconciliation costing your team right now?
MarcusI've heard that pitch before. Show me a concrete ROI model, not a hypothetical — or we're done here.

Pressure, fatigue and held-position run as internal signals — surfaced in the post-session analysis, in language. No live gauges, no scores.

The difference

Know who you are practicing against.

A serious rehearsal needs a counterpart you can understand: who they are, what they value, and why they answer the way they do. In ArenaSynth that profile is set before the first word and it is yours to read.

Generic roleplay

A personality filled in on the fly.

A generic roleplay fills in the personality on the fly: you don't see who it settled on or why it answered what it did.

Session 1→Session 8→Session 20
no profile to read

ArenaSynth

A profile you can read.

The profile is fixed before the conversation begins and you can open it. The situation changes; the profile you are facing stays the same.

Same profile→new situation→you can see why
one profile, open to inspect
Why it isn't a generic-AI wrapper

Same knowledge. Three people. Three behaviors.

Give three counterparts the exact same fact. Each starts from a different psychometric profile, and you can see which.

Identical input →“The CFO is pushing for a cost cut.”
MethodicalRisk-averseGuarded
“I'd need two years of P&L before any commitment. What's your pricing comparable for similar verticals?”
Reads as: methodical, asks for data, won't commit without evidence.
AssertiveStatus-drivenComposed
“We could move on this tomorrow. What's your discount if I commit to five years? My board wants a partner I can champion.”
Reads as: assertive, pushes for commitment, leverages authority.
Risk-anxiousReassurance-seekingAccommodating
“I'm worried about implementation risk. We tried this with another vendor and it got messy. Can you walk me through a case like ours?”
Reads as: anxious, seeks reassurance through concrete cases.

One fact. Three psychologies. And you can see the profile behind each one.

Identity before the conversation

The person exists before you say a word.

The counterpart isn't generated by your message. A psychological profile is built first (temperament and values) and the conversation runs on top of it.

Change the situation and you are still facing the same profile, now under a different kind of pressure.

Try switching the situation →

Elena Vargas
FOUNDER · SERIES B · EU TECH
From the profile
Protects team autonomy over deal speed
Distrusts aggressive acquisition framing
Emotionally controlled under pressure
◈Situation · Budget freeze
“I won't sign anything this quarter that touches headcount. Bring me something that doesn't, and we can talk.”
// Same values, higher price sensitivity. The situation sharpens the behavior — it doesn't replace the person.
The certainty layer

Practice against what you think you know.

Everything the counterpart knows carries a level of certainty: verified facts, estimates they operate on, and what you merely assume. Practice separating the three before the real meeting.

Public

Verified fact

Their funding, public stance, reported numbers. The counterpart stands firmly on them.

Estimated

Their working reality

Likely budget, internal pressure, their boss's expectations. Treated as true-for-them, the way a real person operates on incomplete information.

Assumption

What you suppose

The things you walked in believing. The conversation can put them to the test, which is where real deals are won or lost.

For example, you open with: “I know your contract with your current vendor is up in March…”
Counterpart: “It's not, we renewed in January.” If your prep is off, this is a good place to find out: before it costs you the real room.
Not an LLM wrapper

Half of your debrief is not decided by an AI.

“I'll pay you as soon as I can” and “I'll pay you on March 15” look almost identical in a transcript. Only one of them gets collected. So that difference is not left to a model: we count it.

Counted in code

What can be checked, is

Was a figure with a currency stated. Was a concrete date committed: 'March 15' counts, 'soon' does not. Was a number used that is not in the figures you declared. No model, no latency, counted the same way every time.

Judged by the model

What needs judgment, gets it

Why the counterpart hardened after turn nine. Whether that concession was a trade or a reflex. Which sentence, rewritten, changes the trajectory. That work needs a model — and every line of the report says which half produced it.

The question worth asking any vendor in this category:
“Which parts of your report were not decided by a language model, and how would I check?” Read how we answer it →
From setup to live session

Built in minutes. Ready to rehearse.

01

Build the counterpart

From an archetype, a pasted CV/LinkedIn, or a multi-source dossier of a real public figure (beta). The psychological identity is derived — and yours to inspect.

02

Set the situation

Pick the scenario and an optional snapshot: budget freeze, board pressure, mid-cycle review. The situation reshapes behavior without changing the profile.

03

Run the session

The counterpart answers from their profile, holds its position under pressure, and changes it when you give good reasons.

04

Read the analysis

Afterwards: outcome, relational trend, key moments and what drove them — in language, not a score. Re-run and compare attempts.

One engine, many arenas

The hard part is the engine. Arenas are the packaging.

StrataSynth composes the synthetic human and runs the behavioral engine underneath. PsycheGraph is the identity it produces. Each arena wraps that engine in the context of one kind of high-stakes conversation. One is live today; the rest are scoped.

Enterprise NegotiationLive
Executive AlignmentSoon
Board PresentationSoon
M&A Due DiligenceSoon
Crisis CommunicationSoon
Investor RelationsSoon
Regulatory NegotiationSoon
Hiring & RetentionSoon
PsycheGraph

The profile comes first.

The profile is decided first

Who the counterpart is (temperament and values) is set before a single word is exchanged. The conversation runs on top of that profile.

The model gives it a voice

The profile comes first. The language model turns it into conversation; the profile is what you can read and check.

Signals stay internal

The counterpart feels pressure, fatigue and risk — but you read them the way you would in a real room: in how they respond. Surfaced afterwards in language, never as a live score to game.

// You don't get a chatbot with a personality prompt. You get a counterpart built from a profile you can inspect.

What's actually real

Claims we can show our work for.

No invented consistency score, no hand-wavy “AI magic.” What stands behind a counterpart is established personality science and a profile you can open and read.

Grounded
Counterparts are built on established personality and values science, drawn from published norms.
decades of psychometric research
Inspectable
The psychological profile isn't a black box you have to trust. You can open it, see who you're up against, and delete it whenever you want.
open the profile · delete on demand
EU first
Built and hosted in the EU. Source material is processed and discarded — only the derived profile stays.
EU infrastructure · source data never stored

// Our position isn't “our synthetics are right.” It's “we tell you exactly what each one demonstrates — and what it doesn't.”

Privacy by design

What you share simulates the person. Nothing more.

Never stored
The CV or notes you provide are processed in memory to derive the profile, then discarded. They never reach our database or logs.
Only the profile
What exists in the system is the derived psychometric vector — which you can inspect and delete at any time.
Yours to delete
Counterparts expire automatically, and a delete removes the profile and everything derived from your text.
EU infrastructure
Hosted in the EU. We use Google Analytics for aggregate usage to improve the product — no ad retargeting, no data brokering. Full policy →
⚠ Honest limit — Dossier (beta)
The dossier path, which simulates real public figures from multi-source evidence, currently routes some processing through models without EU residency, and the evidence engine retains raw material briefly with an aggressive TTL. It's beta, not for regulated enterprise until that gap is closed. The archetype and CV paths stay fully inside the EU guarantees above.
Where this goes

Every important conversation deserves rehearsal.

Today, the arena is enterprise negotiation. The engine doesn't care about the domain.

Negotiations today
→ every strategic human interaction tomorrow
Enter the arena →

// Create a decision-maker, set the situation, and practice until the pressure feels familiar.