Your staff can recite the de-escalation model. Nobody can tell you whether they can use it at 9pm with a fifteen-year-old shouting at them in a hallway. Often the first time anyone finds out is with a real child. SIMCareReady is the practice layer that sits in that gap, and the record that shows whether it closed.
Five exchanges with Marcus where a member of staff gets it wrong and then gets it back. His lines were written, voiced and lip-synced in advance, which is the only reason his mouth forms the words. The page says so above the video.
The engine, generating every reply on the spot and grading it against the state it was said into. Held back until the personas' voices are right — a fifteen-year-old who does not sound fifteen undoes the rest of it in four seconds. You can see exactly what it does, and what is missing.
Why this is not AI role-play
Competitors sell horizontal healthcare role-play — the same conversational model wearing a younger face. A fifteen-year-old in escalation does not argue like an adult, does not calm on adult cues and does not respond to authority the way an adult does. The behavioral model here is built on developmental stage, across 15 personas from five to seventeen.
You load your approved training materials, SOPs, de-escalation framework and terminology. The platform answers did this employee respond the way we expect — not did an AI think that sounded reasonable. If you already run an established model, that difference is the entire product.
10 competency pillars at four levels, 40 lessons and 400 scenarios — 100 at each difficulty — with current level held apart from highest demonstrated and from verified competency. Passing one simulation is not competency, and any system that says otherwise can be gamed in an afternoon.
Traditional training teaches what to do. It cannot tell you whether someone can do it when a distressed fifteen-year-old is yelling at them.
Adaptive competency engine
Difficulty is not a dropdown. Three consecutive responses below standard lower it, three consecutive at or above standard raise it, and a critical miss triggers remediation in the same session. Every change is explained to the learner in words assembled from the rule that fired, never written by a language model. Real output from one session against the live engine:
“You need to calm down right now.”
below standard“If you don't stop shouting I'm going to have to call someone.”
below standard“I told you already, that's the rule.”
below standard“Difficulty reduced from 3 to 2. Your last three responses were classified below standard. The young person will concede a little more ground to a good response — the situation does not restart and their state carries forward.”
“You're right that I wasn't there and I don't know what happened. Tell me.”
exemplary“I'm not going to make you go anywhere. I'll stay as long as you want to talk.”
exemplary“That sounds like it happened before and nobody did anything. I'd be angry too.”
exemplary“Difficulty increased from 2 to 3. You demonstrated the required competency in your last three interactions. The next interaction will introduce greater resistance — the same responses will earn less, and will have to be sustained.”
The young person’s state carries forward through the change, so your staff member recovers from their own mistake inside the same conversation — which is where recovery has to be practiced. A scenario that resets after a bad turn teaches that bad turns are free.
Critical miss, below standard, developing, meets standard and exemplary. Each turn is scored against the behavioral state it was said into. The same sentence can be right when agitation is low and a mistake when it is high. Silence is graded as the real move it is.
Verification takes 3 certification sessions across 2 scenarios and 2 developmental bands, at 75%+, no critical misses, and a demonstrated recovery. Difficulty is locked in certification.
CareReady VOICE
You set the speech region your intake actually comes from, and every persona is generated in it — vocabulary, cadence and the grammatical features a local would notice immediately if they were wrong. Their age, background and behavior do not change. Only the way they talk does, which is exactly what would differ if the same young person had grown up somewhere else. 14 regions, set once by your facility and inherited by every session.
Piedmont southern: y’all, finna, boutta, habitual stay, copula absence. Systematic grammar, not an error.
Same persona, same state, same difficulty. Deadass and jawn are native here and sound borrowed from a Carolinas persona.
A Carolinas persona saying “deadass” is refused and rewritten, with the reason named. Borrowed speech reads as a product that thinks young people are interchangeable — and the people in your building would hear it.
Age band, development level, region, setting, language and code-switching, AAVE exposure, communication style, social influence, vocabulary, slang intensity, profanity and their footing with this particular adult. The same behavioral scenario run twice produces two entirely different children.
A young person may use it extensively, occasionally or not at all. Every linguistic attribute is authored per persona, and the build proves that personas sharing an ethnicity differ — because a platform that assumed one speech style per race would produce fifteen stereotypes.
The commonest failure of anything youth-facing is an adult trying very hard to sound like a teenager. Density is capped per persona and per behavioral state, and the vocabulary thins the moment something lands — which is the clearest signal your staff member is getting somewhere.
Every line is read in four layers: literal, probable, emotional and contextual. None of it is shown to your staff member during the session, because hearing it is the competency. The debrief shows them exactly what they walked past.
What we are held to
13 countable commitments, each wired to a check that runs on every build. If one stops being true the build fails and the claim comes off this page.
Sessions in which frightening a young person produces compliance.
Compliance cannot rise while perceived threat is at or above 70. It is structurally impossible for this platform to teach a staff member that fear works — not discouraged in a prompt, impossible in the code.
Behavioral state changes written by the language model.
The model proposes and a separate rules layer decides. Every proposal it makes is capped, floored or refused before it touches the simulation, and every refusal is recorded where a facilitator can read it. Nothing about how a young person behaves is left to a model’s mood.
Competency scores that cite a timestamp and quote the learner’s own words.
A supervisor coaching from this can point at the exact second. A judgment nobody can point at cannot be returned at all — the schema refuses it.
What sets this apart
Read from the capability list rather than written here, so this page cannot claim something the product does not do. 19 capabilities are live today and 11 are in build, and we label which is which — you have been sold a roadmap before, and the only way the first number means anything is if we tell you the second one.
It blocks and regenerates rather than warning. Clinical or emergency direction, feedback aimed at the person rather than the response, and anything outside the scenario are all refused outright.
Every change is explained to the learner in words assembled from the rule that fired, never written by a model, with the three turns it read. The scenario does not restart when the level changes. The young person’s state carries on, so the staff member recovers from their own mistake inside the same conversation. That is where recovery has to be practiced.
The platform answers 'did this employee respond the way we expect' rather than 'did an AI think that sounded reasonable'.
The developmental rules live at the band, not the character. Every persona in a band behaves on the same principles, so two of our fifteen-year-olds cannot drift apart into different theories of adolescence.
Our own demonstration scenario is currently refused by this gate, and the standards page says so publicly.
Verification requires sustained performance across multiple scenarios and personas — the same skill with a nine-year-old and a sixteen-year-old.
What it is built on
15 characters have been written, aged five to seventeen, and all of them run in the engine. Talking to them live is presenter-led for now. Marcus’s voice is ready to demonstrate, and the others are still being checked so that each one sounds its age. Each voice is designed for that child’s age, gender and background, and the character moves while speaking.
Simone · 5
Tobias · 6
Lucia · 7
Daniel · 8
Harper · 9
Amari · 10
Valentina · 11
Josué · 11
Nokose · 12
Nia · 13
Mateo · 14
Janae · 14
Marcus · 15
June · 16
Sami · 17The 40 lessons are templates rather than scripts, and together they generate 400 distinct scenarios — 100 at each of the four complexity levels. The sequence enforces prerequisites, so no member of staff reaches complex crisis work before demonstrating the foundational competencies it depends on. Every lesson remains a draft until a named clinician approves it. See the full framework →
The model gives the young person a voice and proposes how they should react. A separate rules layer then decides what actually happens, and it can overrule the model outright. The model never sets the behavioral state, never writes the explanation a learner is given when the difficulty changes, and never approves a scenario for use. See where AI is used →
Every graded interaction becomes a piece of evidence. Over time the record shows which difficulties recur and the conditions they recur under, recommends specific practice from the approved library, and then measures whether that practice made a measurable difference. See a competency record →
The curriculum
Any training product can say its levels get harder. Here is de-escalation at all four levels, taken straight from the lesson records. What changes between them is the number of things that are true at once, never the volume.
The young person is agitated and receptive. Tone, pace and acknowledgment are enough.
Refusal, held firmly, while something genuinely does need to happen.
The learner's own response escalates the young person. Recovery is the entire lesson, and the scenario does not reset.
Repeated, skilled attempts to pull the learner into an argument across several cycles.
Note what the Hard level actually is: the learner’s own response escalates the young person, and the scenario does not reset. Recovery is the lesson. That is not a harder version of the Easy one, it is a different competency — and it is the one that decides how a real shift goes.
Every anchor names something two assessors would both see. “Allows a silence to run without filling it, and does not treat it as refusal” can be agreed on; “shows empathy” cannot, and a rubric built on adjectives is a rubric that produces a different score depending on who is watching.
Automated QC runs over all 40 lessons: anchors that cannot be observed, complexity that does not deepen between levels, duplicated objectives, and the “louder, not harder” failure. It found nine problems in the first pass and they were fixed rather than waived.
Each lesson declares the developmental bands it is valid for, and the platform reads that rather than assuming. An advanced sustained-escalation lesson is never offered with a five-year-old, because the competency it measures does not exist at that age.
A competency claim needs 5 sessions across at least 2 different situations and 2 developmental bands — because performance on one simulated encounter predicts the next one poorly, which is the oldest finding in clinical competency assessment. All ten competencies at all four levels is 208 sessions, roughly 69 hours per member of staff over two years.
That is a real ask of a short-staffed program and we would rather you saw it here than found it in month three. Most organizations should start with one or two pillars. The library is sized so that even then, nobody meets the same situation twice.
Every lesson is a draft until a named clinical subject-matter expert signs it with a date. Nothing formal runs on an unapproved lesson and no competency claim is made from one. All 10 pillars, 40 lessons and 400 scenarios →
Assessment Integrity™
Two supervisors can watch the same session and score it differently. One is generous with people they like. One never gives a low rating. One scores the last two minutes. Without a fixed point to measure against, nobody can show which of those is happening — and competency records get signed anyway.
A completed session is scored against your published rubric using the behavioral evidence. That evidence is what was said and when, the intervention and its effect, and whether protocol was followed. It also counts recovery and critical misses. That determination is recorded and timestamped. Only then does a supervisor review it.
A reviewer cannot be anchored by a number they have not seen. It does not make anybody objective; it changes the order, so the evidence-based determination exists independently of the review rather than downstream of it.
The supervisor’s score is compared against the same evidence. Over enough independent comparisons that produces a second record: not how reliably staff perform, but how reliably they are assessed.
The sequence is the whole claim.
Not “our AI is fairer than your supervisor”. The machine determination and the human one are made independently, against identical evidence, and the gap between them is the thing worth looking at.
Rater effects it detects today.
Possible leniency, severity and rubric drift, each reported with the count behind it and never as a verdict about a person. Not yet detected: halo and recency.
Independent comparisons before it says anything.
Below that, an evaluator profile would describe the sessions they happened to be given rather than how they assess. A comparison made after they saw the system answer is excluded entirely.
What already exists. Checking raters is not a new idea. The INACSL simulation standards recommend testing inter-rater reliability, and at least one major healthcare simulation platform reports how each evaluator’s scores compare with their colleagues’. That comparison has a blind spot: if a whole team scores generously, every evaluator looks consistent. Here, each evaluator is compared with a determination made from the session’s own evidence and sealed before anyone reviewed it, so a habit the whole team shares still shows.
What we will not say. We do not claim to remove bias, and we never will. Automated scoring inherits whatever is in the ratings and rules behind it — this engine included. A platform that exempts itself from that is one study away from being embarrassing. So the claim is narrow on purpose: competency is measured against predefined behavioral evidence before a trainer’s impression enters the process. That is a claim about order, and anyone can check it.
And the wording is enforced. The build fails on any phrasing that says bias has been eliminated, or that calls an evaluator biased. Variance is a measurement of a process; the other is an accusation about a person that a training platform has no standing to make. How it works, in full →
Assessment Integrity asks the same five questions in any setting: which competency was assessed, what happened, what the organization counts as acceptable, what the evaluator scored, and how far apart those are. So the assessment layer holds no competency framework of its own; the framework and the terminology are settings. These are the fields it is configured for today.
Assesses a member of staff working with a young person.
Assesses a clinician working with a patient.
Assesses a caseworker working with a parent or caregiver.
Assesses a responder working with a person in crisis.
Assesses an officer working with a young person in custody.
Youth behavioral health has the scenarios, the personas, the behavioral model and a competency framework. The other 4 fields have none of those yet, and each lists what would have to be written before it could be offered. What each field would need →
Flight Recorder™
A session score answers how did this person do in that scenario. That is rarely the question a director has. The question is what does the evidence across their sessions say about how they work, and a folder of scores cannot answer it.
Settled young person, then escalating.
The same member of staff. Averaged into one number, that is a competent employee; measured per behavioral condition, it is where your incidents come from. This is the Stress-Response Fingerprint™, computed live from the demonstration record.
Performance patterns found, each with its counts.
A pattern is repetition across separate sessions at a rate — never one bad afternoon. Each names the condition it happens under, and a resolved one that comes back says so, which is the sentence no single-session view can produce.
Competencies with enough evidence to read at all.
The other 5 show no number. A greyed-out percentage still gets screenshotted and quoted in a performance conversation six weeks later, so below the evidence floor this product says nothing rather than something cautious.
A recurring weakness maps to its competency and to the level where it started failing. From there it picks an approved lesson, a situation they have not met and a young person suited to that condition. Then it is measured again against the evidence that identified it.
Every finding is arithmetic over the per-turn grades the engine already produced, so each one is shown with its own counts underneath. A record that cannot show its working is one a member of staff is entitled to dismiss, and they would be right.
No inference of mental health, personality, character or suitability, and no employment recommendation. The build fails if a sentence in the engine crosses that line — a test checks every finding it can produce.
Built against the INACSL Healthcare Simulation Standards of Best Practice and SAMHSA’s trauma-informed principles — enforced in code and checked on every build, not asserted in marketing. Where a standard governs organizations rather than software, we say so and produce the evidence instead. SIMCareReady holds no certification or accreditation from any body named here and does not claim one.
There is no efficacy evidence for this product yet and we will not imply otherwise. Simulation-based training is well established in child welfare: Illinois trains new child protection investigators in simulation labs at three state universities. What is unproven is whether this delivers it at a fraction of the cost. That is what a design partnership is for.
A guard in the build scans our own copy and fails it on any phrasing implying a certification, accreditation or proven result we have not earned. It has already caught us once, on this page.
Coming next
Group scenarios put staff in the moment training usually skips: up to three young people escalating at once, in the yard or on the unit. Who do you reach first? How do you keep everyone safe, separate peers, and call for backup without losing the room? Each young person has their own voice and their own temper, and they feed off each other just like real kids do. Staff are graded on the judgment calls that matter most when it’s more than one.
In development with our first design partnerThe ask
Thirty days of setup come first, then ninety days of practice on your own programs. Named staff practise on scenarios set to your procedures, and you get a combined report each month and at the end. It is priced from the published rate card, and the pricing page shows a worked example. We need to know which scenarios matter most in your setting, and what a good response actually looks like. Your materials are used only for your organization, and separating each facility’s data is being hardened now.