The Problem
The question no one is asking
When a platform removes a post, it answers one question. It never asks the one that matters.
Every major digital platform has a trust and safety function. The job of that function, almost universally, is to decide whether a piece of content is permitted. If it is not, the content is removed. If it is, the content stays. The decision is made in the moment, about the object, the post, the message, the image and the person who created it plays no part in the analysis.
This is content moderation. It is necessary. It’s not sufficient.
The person who posted something alarming yesterday posted something similar three weeks ago. And two weeks before that. The platform saw each post in isolation and made a binary decision about each one. No system connected the dots. No system asked whether the person behind the posts might be in a deteriorating situation that warranted a response before it became a crisis.
That is the gap. Not a technical gap. A conceptual one. The question of how someone is doing over time has simply not been part of how platforms think about safety.
Where the existing tools stop
The trust and safety market is large and growing. Regulatory pressure from the Online Safety Act in the UK, the EU Digital Services Act, and equivalent legislation elsewhere, has accelerated investment in moderation technology. But the vast majority of that investment has produced faster, more accurate versions of the same tool. Content classifiers that answer the allow-or-remove question at scale.
Some tools go further. A small number of platforms monitor for patterns of toxic behaviour or harassment. A handful of child safeguarding tools track grooming-related language in gaming chat. These are meaningful advances. But even the most sophisticated of them share two fundamental limitations:
-They are optimised for detection, not response. The output of the system is a flag or a removal, not a pathway to support.
- They are calibrated to a moment, not a trajectory. The question is always: what is this content? Not: what is happening to this person?
The result is that grooming, coercive control, self-harm escalation, and exploitation can unfold across weeks or months on a platform, with the platform responding, if at all, only at the point where a single piece of content crosses a threshold. By then, significant harm may already have occurred.
Why gaming and educational environments come first
Gaming platforms and educational environments are the two contexts where the gap is most acute and the existing tools most poorly adapted.
Gaming communities have unique harm vectors: voice chat, in-game messaging, guild and clan dynamics, and a culture of anonymity that creates conditions that predatory behaviour exploits systematically. The language of grooming in a gaming environment does not look like the language of grooming anywhere else. Most existing child safety tools were not built for this context and perform poorly in it.
Educational platforms carry a statutory duty of care, under Keeping Children Safe in Education and related guidance. This creates both a compliance motivation and a moral obligation to do better than binary content moderation. A school's safeguarding responsibility does not begin and end at the school gate, and it does not apply only to individual pieces of content.
Both contexts involve significant populations of young people. Both involve trusted adult-child relationships that predatory behaviour can exploit. Both are environments where early detection and proportionate response could meaningfully change outcomes.
The research gap underneath the product gap
The absence of longitudinal, wellbeing-oriented safeguarding tools is not simply a market failure. It reflects a deeper gap in the research base. To build a system that detects escalating harm reliably and responds proportionately, you need:
- A validated taxonomy of harm signals, grounded in clinical psychology, coercive control theory, criminology, and linguistics, that defines what markers are meaningful and why
- An understanding of how those markers manifest in specific platform environments, gaming communities and educational platforms look nothing like the social media contexts where most of the existing research has been conducted
- A dataset constructed under academic ethics approval, annotated by qualified researchers, and validated across diverse communities, because a detection system calibrated on non-representative data does not protect everyone equally
- A clinical governance structure that defines escalation thresholds, not a commercial team setting confidence scores
None of this exists in usable form. That is why Nostra Veridicus is beginning with research, not with a product.
The approach we are taking
Nostra Veridicus is building a research programme before a product. The foundation of which, is a clinically validated harm signal taxonomy, constructed through systematic literature review, tested against real platform environments under academic ethics approval, and validated by an independent Expert Advisory Board spanning clinical psychology, criminology, linguistics, legal expertise, and critically, the lived experience of those who have experienced the harms we are trying to detect. The technology will follow the research. Not the other way around.
We are not building another content moderation tool. We are building the evidence base for a system that asks a different question: not what was posted, but how is this person doing and what, proportionately, should happen next.
Ready to partner with us?
We work with organisations that share our commitment to evidence-based practice.