Policy safeguards for fair public-interest analysis.
Observed is designed to compare publicly available information with academic and recognised good-practice benchmarks. This page explains the legal, ethical, publication and correction safeguards that sit around that work.
Policy position
Observed publishes accountable comparisons, not legal verdicts.
Observed’s role is to examine lawful public information, classify public signals, apply independent benchmarks and explain what the comparison can responsibly show. It does not decide guilt, liability, motive or legal wrongdoing.
No legal advice
Observed does not provide legal advice, legal representation or legal findings. People and organisations should seek independent legal advice where legal rights, risk or proceedings are involved.
No findings of wrongdoing
Observed does not state that an organisation is guilty, corrupt, unlawful or responsible for misconduct. Outputs identify public signals, benchmark alignment, limitations and accountability questions.
No private investigation
Observed does not conduct surveillance, impersonate people, access private databases, seek covert recordings, use hacked material or obtain information unlawfully.
Core legal and ethical frame
Four safeguards sit at the centre of the model.
The model is designed to be careful, proportionate and transparent. Every named-organisation output must stay within comparison language, public evidence boundaries, methodology disclosure and right-of-response practice.
These safeguards do not remove all legal or ethical risk. They are designed to reduce risk, support fairness and ensure the work does not become unsupported allegation or reputational attack.
| Safeguard | What it means | Why it matters |
|---|---|---|
| Comparison, not allegation | Outputs compare public signals with research frameworks and recognised good-practice standards. | The benchmark generates the standard. Observed does not manufacture allegations or deliver verdicts. |
| Public evidence only | Every signal must be publicly available, lawfully accessible and attributable. | The evidence boundary protects the model from private investigation, unlawful collection and unfair use of material. |
| Methodology disclosure | Outputs explain sources searched, frameworks applied, confidence ratings, limitations and human review. | Readers can understand how the analysis was produced and what it cannot conclude. |
| Right of response | Named organisations receive a fair opportunity to respond before named publication where required. | Relevant responses can correct, balance, update or limit the proposed analysis before publication. |
Publication boundary
What Observed will and will not publish.
Publication is only appropriate where the analysis remains evidence-based, benchmarked, proportionate and within the methodology.
Published work must remain
- Comparative, not accusatory.
- Source-cited and based on lawful public material.
- Benchmarked against disclosed frameworks or standards.
- Confidence-rated and limited by the evidence base.
- Balanced by contradictory or moderating signals where they exist.
- Reviewed by a human before release.
Published work must not become
- Unsupported allegation or reputational attack.
- Revenge content, harassment or personal targeting.
- Inflammatory campaign material.
- A finding of legal wrongdoing.
- A claim that goes beyond what the evidence can support.
- A republication of private, unlawful or unsafe material.
Named-organisation findings are suppressed unless signals are drawn from at least three independent source types. A concentration of signals from a single source type, including multiple anonymous reviews on one platform, does not meet the threshold regardless of volume.
This protects against thinly sourced findings, small-sample identification, single-source patterns and confidence ratings that exceed what the available evidence can support.
Legal review triggers
Some matters require legal review before publication.
Legal review may be required where a matter carries heightened publication risk, relies on sensitive material or involves circumstances where the consequences of publication may be significant.
Legal review is triggered where publication may involve
- Allegations of misconduct or deliberate harm.
- Reliance on anonymous signals without strong corroborating public records.
- Active legal, employment, complaints or regulatory disputes.
- A politically sensitive or publicly funded organisation.
- Legal correspondence from the subject organisation.
- Children, vulnerable people or heightened privacy risk.
Legal review may result in
- Publication proceeding with changes.
- Language being narrowed or qualified.
- Additional right-of-response steps.
- Further evidence checking.
- Suppression of specific findings.
- Publication being paused, declined or withdrawn.
Human review gate
No named-organisation output leaves the system without human sign-off.
AI supports collection, classification, benchmark comparison and draft synthesis. The human reviewer remains accountable for what is published.
No named individuals
Outputs are checked to ensure natural persons are not named as the subject of analysis unless a separate human review and legal pathway requires it.
Source verification
Every signal must be linked to a specific, accessible, attributable and dated source before it can support a finding.
Confidence limits
Confidence ratings must reflect source diversity, source hierarchy and sample strength, not rhetorical strength.
Contradictions included
Contradictory or moderating public signals must be included where they exist, or their absence explained.
Language review
Language must remain comparative, cautious and proportionate. Accusatory or verdict-style wording is not acceptable.
Publication decision
Publication only proceeds once right of response, legal triggers, conflict checks and proportionality have been assessed.
Conflict of interest
Observed must not publish where conflicts compromise the work.
Observed maintains a conflict check for matters assessed. A named-organisation analysis should not be published where a relevant conflict exists unless the conflict position has been assessed, disclosed where appropriate and legally reviewed where required.
The model applies a 24-month conflict window for organisations that are, or have recently been, clients, employers or commercial partners of the founder or associated businesses. Where doubt exists, the matter should be declined or deferred.
| Conflict control | What is recorded | Possible outcome |
|---|---|---|
| Relationship check | Any prior client, employment, commercial, advisory or related business connection. | Proceed, disclose, defer or decline. |
| Confidentiality check | Whether any confidentiality obligations could be relevant to the proposed analysis. | Decline, defer or seek legal advice. |
| Public-source check | Whether the analysis can be based entirely on public information. | Proceed only if no private knowledge is relied on. |
| Disclosure check | Whether a conflict disclosure is required in any published methodology section. | Disclose, narrow, defer or decline. |
Correction and withdrawal process
Published work is not treated as immutable.
Any person or organisation may submit a correction request. Observed assesses whether new evidence, factual error, changed circumstances or methodological concern requires clarification, correction, update or withdrawal.
Correction request received
The request should identify the specific finding challenged, the grounds for the challenge and any supporting evidence or public source material.
Original source register reviewed
The challenge is assessed against the original methodology, source register, confidence rating, limitations and publication decision record.
Effect of new information assessed
The review asks whether the new evidence changes the confidence rating, changes the finding, reveals a factual error or requires additional limitation.
Outcome recorded and applied
The correction outcome is recorded. Published material is left unchanged, clarified, corrected, updated or withdrawn depending on the assessment.
Correction outcomes
There are five possible correction outcomes.
| Outcome | What happens | When it applies |
|---|---|---|
| No change | The finding stands. The request and assessment outcome are recorded. | The new information does not change the source base, confidence rating or finding. |
| Clarification | The published output is updated with a clarification note and date. | The original wording may be misunderstood or requires additional context. |
| Correction | A factual error is corrected and transparently noted. | A source, date, description or factual element was wrong. |
| Update | The output is updated to reflect new evidence or changed circumstances. | New evidence materially affects the finding or confidence rating. |
| Withdrawal | The finding or output is removed and replaced with an appropriate withdrawal notice. | The finding can no longer be responsibly sustained. |
The organisation may be named. The allegation is not made. The benchmark does the work.
What Observed declines
Some matters do not belong in this model.
Observed may decline or refer a matter where the purpose, evidence base or risk profile does not fit the public-interest research model.
Bad-faith or retaliatory matters
Requests motivated by revenge, harassment, reputational attack or personal targeting are outside the model.
Private disputes without public interest
Matters that do not raise an organisational accountability question may be better suited to another pathway.
Insufficient public evidence
Where lawful public sources are too weak, narrow or unavailable, the matter may be held, narrowed or declined.
Unlawful or unsafe material
Hacked, leaked, private, covert, unattributable or unsafe material is excluded from the evidence model.
Personal targeting
Observed analyses organisational patterns and public accountability questions, not campaigns against natural persons.
Better handled elsewhere
Matters may be referred to legal, employment, regulatory, safety, union, complaints or support pathways where more appropriate.
Have a concern that may fit the model?
Review the methodology, evidence standards and research process before submitting a request. The initial assessment will determine whether the matter fits Observed’s public-interest evidence model.
Public evidence. Academic benchmarks. Better accountability.
