Preliminary analysis base
No confirmed analysis report yet
Documentation · solidarity · independent review
Rettsspeil
When more people ask the same questions, we can examine them together.
A workspace for understanding decisions and examining patterns in administrative procedures. The aim is a verifiable basis for professional review, access to records and possible legal assistance.
This is a prototype. You can try local text review and a simulated study. Real study submissions, contact registration, external AI analysis and payments are closed. The analysis examples and account demonstration use fictional test data. A separate section displays only a confirmed account report, if one is available from the server.
Build the evidence base and funding in parallel
Three phases – legal assessment when the basis is ready
- 1. Collect and analyse
Decisions receive preliminary analysis while contributions are collected by the lawyer responsible for the account. Each decision can count as machine-analysed without prior legal assessment. Account administration and reconciliation are a separate engagement.
- 2. Invite an assessment
Once enough distinct decisions have received preliminary analysis and a current account report confirms sufficient available funds, a lawyer can be invited to assess the material. No invitation is sent automatically.
- 3. Independent legal assessment
The lawyer decides whether there are grounds for an engagement and whether to accept it. Later use of funds requires a mandate, agreed terms and a budget. A full progress indicator is not a legal conclusion.
Two conditions must be documented
Available funding
No confirmed account report yet
Real analysis collection has not opened. The test figures below never populate this section and cannot be combined with a real account report to indicate that the project is ready.
Progress · controlled report releases
See the work move forward
More participants. More decisions preliminarily analysed. A clearer basis for a later legal assessment.
Readers should be able to follow the material before the lawyer-assessment phase: decisions received, decisions machine-analysed and flags for possible review needs. A machine flag is neither a probability nor an established breach of law.
All figures here are fictional test data. No real submissions are collected. Try the three releases to see how a controlled update can be displayed.
People: Each participant with an accepted decision is counted once.
Decisions: Resubmitting a file must not make the same decision count again.
Checkpoints: One legal provision assessed in one decision. Several checkpoints may concern the same person and decision.
Where does the review stand?
Machine analysis, human checking and legal assessment are shown separately. A decision counts as fully machine-analysed when every planned checkpoint has machine-analysis provenance. Prior legal assessment is not required. Missing documents may remain unresolved, and a flag only means that something should be checked. Retained machine flags may later concern the same points that receive human or legal review; these counts must not be added together.
Legal provisions assessed
The table counts checkpoints per provision. Its example references illustrate the method; they are not a completed check of current law. One decision may appear in several rows.
Next working milestones
Milestones describe workload in this test. No milestone opens collection, releases money or starts legal proceedings.
How updates should be checked
A public report must have checked underlying material and a visible date. Small groups and small changes are withheld where they might reveal participation. Figures for priority or lawyer-reviewed issues require material from at least ten distinct people before display; their relationship to earlier releases must also be checked. This does not guarantee anonymity.
Duplicate checking is demonstrated here with test keys. Verification of distinct real people and decisions, analysis provenance and central approval of reports must be established before real submissions. A local text review or MAITE draft does not automatically enter the overview.
This example has its own underlying data. The institution breakdown and lawyer-held account below are separate demonstrations with different units and test figures.
Step 1 · in this browser tab
Understand a decision
Fixed search rules look for Norwegian wording and find potentially relevant excerpts and questions. The interface is translated; your source text stays unchanged. This is not an AI analysis or legal assessment, and a missing match does not establish non-compliance. The tool does not determine guilt, motive, appeal deadlines or compensation claims.
The note may contain your excerpts and comments. It is not anonymised and must not be published. Downloading sends nothing.
What about automatic AI analysis?
Open the analysis workshop for multiple documents, dated legal texts and checks of structured model results against exact quotations. A separate local runner has been built for a selected AI provider, but requires an API key and explicit approval of the content before transmission. No real cases have been sent and no model account is connected. Preliminary machine analysis does not require prior legal assessment. Source checking does not establish legal correctness, and private results do not automatically enter public statistics. “Maximum effort” or a long processing time does not guarantee a correct answer.
Step 2 · synthetic test data only
Example institution breakdown
This is a separate method exercise with its own synthetic cases, not a breakdown of the progress figures above. Statistics can show which checkpoints need further work. They cannot alone establish that anyone deliberately broke the law. Test material is grouped by institution, broad function and year, without names or prisoner numbers.
Each case counts once
Several decisions or uploads from one case must be reconciled before counting. Duplicate checking is demonstrated with fixed test IDs. Linking real cases requires a separate, controlled process.
Small counts are withheld
At least 10 cases per visible cell. The entire category breakdown is withheld where a small cell could otherwise be calculated. Fixed releases and restrictions on small changes limit inference between reports. This does not guarantee anonymity.
Counterevidence matters too
Completeness, document versions, appeals and information supporting the decision must accompany private review. A self-selected sample cannot show how common an error is across the correctional service.
Funding alongside analysis
Legal assistance – funds held by the lawyer
The plan is for contributions to go directly to a suitable account controlled only by the project's account lawyer, if the lawyer and bank accept the arrangement. The lawyer may take on account administration and reconciliation before a later engagement to assess litigation. The website, project owner and AI tool must not be able to withdraw or transfer money.
No fundraising is open. Account administration, authority over funds and contribution terms must be settled separately. There is no account number or payment button here. The server report and local demonstration below are clearly separated.
Funds held by the lawyer
The lawyer confirms receipts against the bank statement and keeps records under the agreed arrangement. A pledge does not count as money received.
A defined engagement
Funds may only pay for agreed legal assistance. Once sufficient analysed material and confirmed funding are available, a lawyer is invited to assess the material, a possible engagement and its budget. A case count or AI assessment cannot start proceedings or trigger payment.
Returning unused funds
Terms must specify deadlines, costs and when funds must be returned. The system can prepare a refund list when the conditions are met. The lawyer makes the transfers and confirms them against the bank.
Confirmed account report
This section retrieves only dated aggregate amounts from the server's reporting service. The designated account reporter confirms the figures; the website has no bank access. Donor names, account numbers, payment references and individual cases are not published.
No confirmed account report yet
Try the account workflow with fictional amounts
The example has a six-month collection period, seven days for assessment, then seven days to register any disbursement. Its test goals are NOK 5,000 in reconciled available funds and 100 preliminarily machine-analysed decisions. These are demonstration values, not agreed terms for real fundraising. The analysis button represents the analysis operator; account buttons represent the account lawyer. The separate legal assessment is tried only after both goals are met. The buttons confer no login or authority over funds and perform no AI analysis.
Reconciled test report · fictional amounts only
An approval to use funds is not a bank transfer. A refund list is not a completed refund. Confirmed totals change only when a corresponding bank event is registered in the demonstration. Real bank controls must follow the agreement with the lawyer and bank.
What you should be able to consider before participating
The first real pilot must be limited to adults aged 18 or over. Participation must be voluntary and separate from decisions about imprisonment, treatment or benefits. Declining or withdrawing must have no adverse consequences. Other people's information, particularly children's, requires separate assessment even where the submitter is an adult.
- Understand the purpose. The study must test questions using an open method and also record findings that do not support the suspicion.
- Check the documents. Keep your originals. Distinguish personal experience, the document's wording and assessment. Appeal deadlines must be handled independently of the project.
- Understand what may be shared. Private review is separate from permission to enter a study or publish material.
- Choose any later contact separately. A contact link makes information pseudonymised, not anonymous. An independent custodian must control that link. Contact information must not appear in public statistics.
- A new request if a case becomes possible. A lawyer must check originals and seek fresh consent and an engagement before information is used in a specific case. Nobody automatically becomes a party to litigation.
- Understand withdrawal and retention. It must be clear what can be withdrawn, what has already been aggregated and published, and how long each type of information is retained.
Participation is not open. No checkbox on this page activates real collection. A legal basis for processing health information and criminal-offence data, among other types, must be assessed separately. Consent alone does not resolve every requirement.
Before a public launch
The project owner must approve the working version. A verifiable method, security review and defined operational responsibility must precede real collection.