You do not need to know exactly what is wrong to start.

Start with a hunch

Your organization does not need a finished problem statement, a chosen survey, or a named culprit. A hunch is enough to begin. For example:

• Rising friction in how teams work together
• Silence around difficult issues
• Unclear expectations causing avoidable churn
• Directness and respect being read in very different ways
• Boundaries that feel unsafe to say out loud
• Misunderstandings that never get repaired
• Uneven manager practices between units
• Turnover, conflict, or avoidance growing without a clear explanation

A hunch is always about how work happens, never about a named person. It is not a complaint, an accusation, a diagnosis, or a risk rating of anyone, and this concept has no place to enter one.

One level earlier than your current tools

Survey tools are strong when you already know what to ask. Case management begins when a formal case exists. Relationship Safety OS begins one level earlier, from an unclear concern.

• Employee listening

  Engagement and experience surveys work when the question is already known. They show how people feel, but a score does not define the problem, start any action, or measure whether anything changed.

• Relationship Safety OS

  Begins from a hunch. It helps make the underlying pattern visible and testable first, then guides a structured, human-approved software response, and measures what changes.

• Reporting and case management

  Hotlines, investigations, and case tools begin only after something has been formally reported. They handle cases; they do not form the problem or verify prevention work.

The first pathway: from implicit expectations to shared clarity

The first concept pathway is deliberately simple. A team and its manager make the invisible explicit: how expectations are given in writing, what directness and respect mean in this team, how much processing time decisions allow, how anything unclear can be checked without stigma, and how a misunderstanding gets repaired.

The pathway uses rule-based structures and ready-made templates. It does not use AI to guess anyone's intent, motives, or neurotype, and it is offered as a testable response to an approved group-level hypothesis, not as a guaranteed fix. What it looks at is observable: was the expectation expressed or assumed, what happened after a misunderstanding was made visible, did behavior change, and did repair succeed.

The same structure also helps separate two things that often get confused: difference in communication style is not automatically read as harm, and harmful behavior is not explained away as difference.

Different minds, shared clarity

Most workplace interaction systems quietly privilege one way of reading tone, social cues, participation, and professionalism. That is how difference gets mistaken for a problem, and how real problems get explained away.

Relationship Safety OS is designed to reduce reliance on unwritten social rules and support multiple communication and processing styles. Expectations are made explicit, clarification is legitimate, processing time is built in, and repair is documented.

This design direction is informed by double empathy research, which shows that misunderstandings between autistic and non-autistic people are often mutual rather than one-sided. We make no claims about effectiveness for any neurotype: stronger claims would require co-development with neurodivergent professionals and real workplace evidence, and that work is planned, not done.

Explicit expectations help more people than any single group: multilingual teams, remote teams, introverts, and anyone who has ever guessed wrong about an unwritten rule.

What it is, and what it is not

Trust and data boundaries

The customer experiences one journey, but it runs on hard boundaries between separate services and companies. If these boundaries do not hold, the product should not exist.

• Employee answers

  Pseudonymous and aggregated by design. The employer would never see individual answers, and group results are designed to appear only for groups of at least 10 respondents. The first product collects no open text.

• Employer data and the hunch

  The hunch is a structured, non-sensitive description of how work happens, never a record about a person. Organization structure and program administration stay separate from employee response data, and a concerning result never triggers an automatic report to HR about anyone.

• Impact evidence service powered by evaluoi.ai

  Phenomenon modeling, Impact Framework design, measurement, and evidence grading are planned as a separate service provided by evaluoi.ai, a separate company, behind a strict company-to-company service boundary. It would receive only pseudonymous, aggregated measurement data, never names, and no core engine is copied between the companies.

• Flag Tracker stays separate

  Flag Tracker is a private consumer product. It is not part of this concept, employers would never see who uses it or what they record, and its data would never be combined with any workplace product.

Why now

Relationship Safety OS is not a compliance product and does not claim to satisfy any legal obligation. The regulatory context still matters, because prevention duties exist while the effectiveness of preventive work is rarely measured.

• California employers in general industry have had enforceable workplace violence prevention duties since July 1, 2024, including a written prevention plan, training, and a violent incident log (Labor Code section 6401.9).

  Cal/OSHA: Workplace violence prevention guidance

• Cal/OSHA rulemaking for a permanent general industry standard is in progress: a revised draft standard was published on July 22, 2026, with comments requested by August 10, 2026, and the standards board must adopt a standard no later than December 31, 2026.

  Cal/OSHA: General industry rulemaking

• Health care is covered separately: California's health care workplace violence prevention standard (Title 8, section 3342) has been in force since 2017 and 2018, and it distinguishes violence by patients or clients from violence between present or former employees, supervisors, or managers.

  Cal/OSHA: Title 8, section 3342

Regulatory facts last checked against the official Cal/OSHA pages on July 29, 2026.

Where this stands

Relationship Safety OS is a concept in customer discovery by Safety in Relationships Oy Ltd, a Finnish company building digital tools for safer relationships. There is no product on the market, there are no customers or pilots yet, and the evaluoi.ai service connection described above is a planned design, not an existing integration. What you see on this page is the honest current state.

The concept builds on our published values: no applications against anyone, no certain interpretations of another person, and no promises we cannot back with evidence.

Read more about the company on the about page and in our values and product principles.

Join the early conversation

We are talking with people who own this problem at work: employee relations, workplace safety, HR, and people leaders. If starting from a hunch, instead of a finished problem statement, sounds like your reality, we would like to hear how it shows up in your organization.

No sales process, no demo, no mailing list. Your message goes directly to the founder.

I'm interested in workplace relational safety

Opens your email app with a prefilled subject.

Frequently asked questions

What if we do not know what the problem is?

That is the intended starting point. You begin with a hunch, chosen from structured options, and the evaluation layer helps make the underlying pattern visible and testable before any intervention is suggested. If the data does not support a hypothesis, the honest answer is no conclusion.

Does the system diagnose our workplace?

No. It does not diagnose your workplace, find "the real problem" for you, or label anyone. It helps you form and test a group-level change hypothesis, with every consequential step reviewed and approved by people in your organization.

How does evaluoi.ai fit into the process?

evaluoi.ai is a separate company whose impact evidence service is planned to power the evaluation steps: scoping the phenomenon, designing the Impact Framework and measurement, and grading the evidence. It would sit behind a strict company-to-company boundary, receive only pseudonymous and aggregated measurement data, and we do not present it as an independent third-party verifier. The integration is a planned design, not something that exists today.

Does AI choose an intervention automatically?

No. A pathway is suggested only against a human-approved, group-level hypothesis, it must pass a safety exclusion check, and a person approves it before anything starts. AI never assigns an app or an intervention to an employee, and the suggestion is a testable response, not a guaranteed fix.

Is this a whistleblowing or reporting channel?

No. Relationship Safety OS works before anything becomes a case, and the hunch intake has no place to report a person. If something needs formal handling, the concept points people to the channels their organization already has.

Would managers or HR see individual answers?

No. The design is pseudonymous and aggregated: individual answers would never be visible to the employer, and group results are designed to appear only for groups of at least 10 respondents.

How is this different from our engagement survey?

A survey works when you already know what to ask, and it ends in a score. This concept starts one level earlier, from an unclear concern, forms a testable hypothesis, runs a structured and human-approved pathway, and measures whether predefined outcomes moved in the expected direction.

Does it diagnose people or detect neurodivergence?

No. It does not diagnose anyone, and it does not infer anyone's neurotype, intent, or motives. It is designed to reduce reliance on unwritten social rules, which helps many kinds of minds and teams.

What happens if something serious comes up?

The concept is not an emergency service and does not monitor individuals. If a hunch or a result points toward formal handling or serious harm, the flow moves to your organization's existing channels instead of a software pathway, without automatic reports to HR about any person.

Is it validated for autistic or ADHD employees?

No, and we do not claim that. Neuroinclusion here is a design principle: reduce reliance on unwritten social rules and support multiple communication and processing styles. Validation claims would require co-development and evidence that does not exist yet.

Can I buy it?

Not yet. This is a concept preview, and we are in early conversations. If you want to follow the work or share your perspective, use the email link on this page.

Sources and dates

Every regulatory and research statement on this page is sourced and dated. Nothing on this page claims customers, pilots, results, existing integrations, or capabilities that do not exist.

• Cal/OSHA: Workplace violence prevention guidance and resources (checked July 29, 2026)
• Cal/OSHA: Workplace violence prevention in general industry, rulemaking page (checked July 29, 2026)
• Cal/OSHA: Workplace violence prevention in health care, Title 8 section 3342 (checked July 29, 2026)
• Crompton et al. 2020: Autistic peer-to-peer information transfer is highly effective (double empathy research)
• Double empathy in the workplace: mutual misunderstanding between autistic and non-autistic colleagues (open access)