Skip to content

Questions

Frequently asked questions

How to read the record, where evidence stays, what policies you set, and how the first engagement is scoped.

Questions 01

The record

What is a record?

A record is the account of one real thing as a reader can inspect it: its values, the sources behind them, the date each source was read, and the state of the account at an earlier date. The point is to let another reader challenge a particular value without having to reconstruct the whole investigation.

What happens when sources disagree?

Disagreement stays visible. The record shows the value carried, the competing values and their sources, and the stated reason for the choice. A reader can inspect the excerpts and dates rather than receiving a single answer that hides the conflict. The record can change when new evidence changes the defensible account.

What does an empty answer mean?

An unexplained blank is not an answer. The record distinguishes not found, blocked and not applicable. Not found describes what was searched and when. Blocked names a restriction that prevented reading. Not applicable says the property does not arise for that entity. Each is a claim with a reason a reader can inspect.

Can I read an earlier state?

Yes. A reader can ask what the record said at a prior date and inspect the evidence and decisions that belonged to that state. Later corrections do not silently rewrite what someone saw earlier. This matters when a brief, filing or decision has to be defended against the information available when it was made.

Questions 02

Deployment and data

Where does Reliable Context run?

The product runs inside your infrastructure and uses one database for your organization. Evidence is retained in your perimeter. The deployment and its permitted connections are reviewed against your policies before work starts. The perimeter page describes the boundary and the questions a security review should put to it.

Who decides which sources may be read?

Your organization sets the source policy, including licences and restrictions on what may be exported. A source that is outside that policy is not treated as available evidence. Source access can change, so the date and policy context matter when someone later asks why a value appeared or did not appear.

Can different readers see different evidence?

Reader clearance is your policy. A reader who is not cleared for a source does not receive the values resting on that source. The restriction is declared as blocked rather than shown as an unexplained blank. That distinction prevents a permission boundary from being mistaken for a negative finding.

What can leave the perimeter?

Audit export is governed by your policy and by source licences. A permitted export carries the value, its supporting excerpt, source and locator, the relevant dates and the record context needed to read it. The aim is for an authorized recipient to inspect what the organization relied on without access to the running system.

Questions 03

Models

Which model providers see our material?

Your model routing policy decides which engines may see each class of material. That policy is part of scoping and the security review, alongside the deployment's network egress. No provider should be inferred from a generic site description: the permitted route for an engagement is given in writing for that engagement.

Is our material used to train models?

Reliable Context does not train models on your material. Model routing and any provider terms still need to be stated for the particular engagement before material is exchanged. The security review is the place to examine the allowed engines, their handling of material and the controls on what may be sent.

Does a model answer count as evidence?

No answer earns its place merely because a model produced it. A value presented to a reader needs a source excerpt, a locator and a date, or a declared reason why no cited answer is available. The demonstration lets you inspect the visible record, including disagreement and declared negatives, instead of trusting a fluent response.

Can we restrict model routing by source?

Yes. The organization defines which engines may see which class of material, including restrictions attached to private or licensed sources. The routing policy belongs with the source and reader policies, because an allowed source may still be unsuitable for a particular engine. Specific routes are agreed before the first evidence exchange.

Questions 04

Engagement

What should we bring to a first conversation?

Bring one entity your people already describe differently, the sources they use today, and the decision that depends on its description. Say who may read each source and what infrastructure or model restrictions apply. Do not attach confidential material to a first email; the exchange method is agreed during scoping.

What happens after scoping?

The scope, sources, reader clearance, retention, model routing and commercial terms are fixed in writing. Deployment follows inside your perimeter under those policies. The first record is then read together, value by value, before deciding whether to widen the work to further entities and how to operate the record.

Is there a published price or fixed pilot period?

No price or fixed duration is published here. The commercial terms are agreed in writing during scoping, after the entity, sources, access rules and deployment constraints are understood. A first record is a concrete unit to examine together; the site does not promise a timetable or imply a standard package.

How do we request a security review?

Use the structured security review email on the trust page and describe the proposed deployment boundary and the questions your reviewers need answered. We address topology, network egress, model routing, access control, retention, logging, export and any subprocessors for a hosted part in writing for the specific engagement.

Questions 05

Research

What does the research programme study?

The programme concerns identity from evidence, provenance, declared absence and whether a retained source still supports a value over time. These are research questions as well as practical record problems. The methods and reference pages distinguish the proposed approach from established literature, so a reader can inspect both.

Are the demonstration entities real?

No. Halden Therapeutics AG (fictional) and other demonstration entities are written examples. They are used to make a record inspectable without presenting a client file as public evidence. Their names, excerpts and decisions should not be treated as claims about a real organization or a deployed customer's results.

Where are technical terms defined?

The glossary defines terms used across the product, methods and research pages. It is useful when two readers use the same word for different things, especially evidence, provenance, a declared negative and reproducibility. The reference page points to public work for claims about the broader field.

How can I challenge a claim on this site?

Name the passage, the source or assumption you dispute, and the evidence that would change your reading. The position paper states the argument in full; the demonstration exposes a fictional record for inspection. Research collaboration can begin with a specific objection, including a proposed evaluation that might show the claim is wrong.

Continue reading

Inspect the argument and the example.

Enquiry

Bring a question the page has not answered.

Name the decision or record you are trying to assess and the part of the answer you need to test.

Or write to [email protected] in your own words.