Safety claims need tests of the whole user journey

Original research record

We connect the OTR data on Instagram Instants risk checks, TikTok safety-control allegations and hidden harmful link destinations. Ofcom opened an investigation; its launch does not establish a breach. Meta separately describes checking where an advertisement leads. The common mechanism is that a button, disappearing image or clean-looking advertisement does not by itself show what happens next. We see a useful early test for product teams: examine the path from the screen to the recommendation, recipient or external destination.

We would specify the control being tested and its expected effect. We would compare actual outcomes with what the screen promises. We would record blocked destinations and mistaken removals without opening or reproducing illegal material. We expect one of the named public regulator or platform records to add a concrete assessment detail, testing method or product response within one week. We do not forecast a fine or legal finding.

The announced protections may already work well. Regulatory inquiry and allegations may remain unresolved. Independent testing that confirms the promised user outcomes would reduce our concern.

We would ask whether the protection works through the full journey, rather than using its presence on screen as proof.

Bayesian · what changed the estimate? We give the stated testable claim a 61% subjective estimate over 1 week. It is uncalibrated and is not a market-direction probability. We raise the weight of dated original statements and concrete events, while keeping repeated OTR files in the same evidence family. de Finetti · are the comparisons fair? We judge the exact event and sample stated here; broader audience belief, adoption and price returns are separate outcomes. Kolmogorov · do the numbers agree? The event estimate and its opposite add to 100%; no measured global denominator or validated base rate is available. The three checks organise the reasoning and do not produce three independent probabilities.

We compare the OTR Database Data across the five-desk editions, research-intelligence records and global-signal verification records. The linked original documents support the underlying facts. The new connection and testable expectation are our inference, rather than a reported result.

Not applicable. This is a social information signal, without an asset-price forecast.

Not applicable to a market direction. We are recording a new synthesis. Earlier versions of the OTR data are used to preserve what was already known; repeated reporting is not counted as a new discovery.

We distinguish discussion, promotion and accusation from original documents, measurements and official decisions. Public comments show that a view was expressed; they do not establish whole-community belief or a measured real-world effect.

We have no frozen, version-matched social-forecast outcome series for this exact claim. We cannot report a historical accuracy percentage. The proposed sample and deadline make this forecast checkable.

How likely is this call? · Three frameworks

Bayesian · what changed the estimate? We give the stated testable claim a 61% subjective estimate over 1 week. It is uncalibrated and is not a market-direction probability. We raise the weight of dated original statements and concrete events, while keeping repeated OTR files in the same evidence family. de Finetti · are the comparisons fair? We judge the exact event and sample stated here; broader audience belief, adoption and price returns are separate outcomes. Kolmogorov · do the numbers agree? The event estimate and its opposite add to 100%; no measured global denominator or validated base rate is available. The three checks organise the reasoning and do not produce three independent probabilities.

Bayesian · what changed the estimate?

We give the stated testable claim a 61% subjective estimate over 1 week. It is uncalibrated and is not a market-direction probability. We raise the weight of dated original statements and concrete events, while keeping repeated OTR files in the same evidence family.

de Finetti · are the comparisons fair?

We judge the exact event and sample stated here; broader audience belief, adoption and price returns are separate outcomes.

Kolmogorov · do the numbers agree?

The event estimate and its opposite add to 100%; no measured global denominator or validated base rate is available. The three checks organise the reasoning and do not produce three independent probabilities.

Evidence behind this view

We compare the OTR Database Data across the five-desk editions, research-intelligence records and global-signal verification records. The linked original documents support the underlying facts. The new connection and testable expectation are our inference, rather than a reported result.

What past price moves show

Not applicable. This is a social information signal, without an asset-price forecast.

Evidence against this view

The announced protections may already work well. Regulatory inquiry and allegations may remain unresolved. Independent testing that confirms the promised user outcomes would reduce our concern.

Why the direction changed

Not applicable to a market direction. We are recording a new synthesis. Earlier versions of the OTR data are used to preserve what was already known; repeated reporting is not counted as a new discovery.

What people say vs the facts

We distinguish discussion, promotion and accusation from original documents, measurements and official decisions. Public comments show that a view was expressed; they do not establish whole-community belief or a measured real-world effect.

How past forecasts performed

We have no frozen, version-matched social-forecast outcome series for this exact claim. We cannot report a historical accuracy percentage. The proposed sample and deadline make this forecast checkable.