Trust and transparency
How the RightSignal Algorithm Works
This page describes the outer layer of the RightSignal Algorithm, which is everything that RightSignal publishes about it. It covers what the Algorithm decides, the outcomes it can reach, the purposes and legal profiles it covers, the kinds of signals it reads, and how its versions and receipts are signed.
01The Algorithm
What it is: The RightSignal Algorithm is the signed and versioned body of rules under which RightSignal reads what a website publishes about automated access to it and use of its content, and reaches a decision. Manraj Singh Chandpuri conceived, designed, built and developed it. Each signed version has a number, and the version in service at the date of this page is RightSignal Algorithm 1.3.6.
What it decides: A client names a website or page, a crawler, a purpose and a legal profile. The Algorithm reads the signals that the website publishes and returns one outcome for that request. Each purpose is decided separately, and the signed rules reach every outcome while a language model never does.
02Outcomes
The five outcomes: Every decision is one of the five outcomes in the table below. Each outcome is named with its profile, for example "Do not collect under the EU profile".
| Outcome | What it means |
|---|---|
| Collect | RightSignal read the site's signals under the chosen profile and found nothing that refuses or conditions the proposed collection for that purpose. |
| Collect on conditions | The proposed collection may go ahead under the chosen profile only while the conditions that the decision lists are met. |
| Hold for review | What the site says is unclear, conflicting or incomplete in a way that needs a person's judgement, so collection should wait until the client has reviewed it with its own legal adviser. |
| Information needed | Something on which the decision depends could not be read or was not supplied, so collection should wait until it has been. |
| Do not collect | RightSignal found an access barrier, a robots.txt disallow addressed to the crawler or a clear refusal, or the chosen profile otherwise stops this purpose, so the client should not collect for it. |
Outcome groups: Every decision also falls in one of three groups. Collect and Collect on conditions form the collect group, Hold for review and Information needed form the hold group, and Do not collect is a stop.
What a collect outcome is not: A collect outcome is not a licence or permission from the website's owner. It is not legal advice and is not a statement that the collection is lawful, as the Not Legal Advice Notice explains.
03Purposes and profiles
The five purposes: The Algorithm covers the five purposes in the table below, and it decides each one separately. The service offers a purpose only where the app lists it.
| Purpose | What it covers |
|---|---|
| Search indexing | Reading pages to build a search index that points people to the source. |
| AI training | Using content to train or fine-tune an AI model. |
| AI answer use | Retrieving content when a question is asked and using it to generate the answer. |
| Text and data mining | Analysing content by automated means to find patterns, trends and other information. |
| Dataset building | Assembling content into a dataset to sell or supply to others. |
The legal profiles: A profile sets the legal frame under which the signals are read, and the client chooses it. The table below describes the three profiles.
| Profile | What it is built on |
|---|---|
| India | Indian law, including the Copyright Act, 1957, the Information Technology Act, 2000 and the interim ruling of the Delhi High Court in ANI Media v OpenAI of 24 July 2026. The profile relies on that ruling only as interim authority, because it is under appeal. |
| EU | The law of the European Union, including the text and data mining exception and the reservation of rights in Article 4 of Directive (EU) 2019/790. |
| India and EU Bridge | Both bodies of law. At each step it applies the stricter of the India profile and the EU profile, for those who operate in India and intend to serve the European Union. |
Stricter variants and changes in the law: Where the service offers a stricter-caution variant of a profile, that variant only ever makes outcomes stricter, and every decision made under it says so. The EU Artificial Intelligence Act places duties on providers of general-purpose AI models, including a policy to identify and respect reservations of rights, and using RightSignal does not by itself meet any such duty. RightSignal will issue a new signed version of the Algorithm if the law changes, and a decision already made does not change.
04Signals
Kinds of signal: The Algorithm reads the three kinds of signal in the table below. RightSignal's own readers read human-readable statements, and in live use these may include an AI reader, in which case the decision says so.
| Kind | Examples |
|---|---|
| Machine-readable signals | Rules in robots.txt, AI and content-use files such as ai.txt and llms.txt, text and data mining reservations, licence files, HTTP headers and meta tags. |
| Human-readable statements | What the site says in its terms of use, legal notices, AI or content-use policies and copyright or licensing pages. |
| Access barriers | A login page, a paywall, a bot challenge or a refusal by the server. |
How a site is read: RightSignal's crawler reads a site under its own declared identity and never passes a barrier. The Crawler and Site Owner Policy explains what the crawler reads, the limits it keeps and what it never does.
05Signed versions and receipts
Signing and sealing: RightSignal signs each version of the Algorithm, and the service does not run a version whose signature fails. Each decision is sealed in a receipt with a record of what was read and when, the profile, the outcome and the version of the Algorithm. A receipt verifies against the version that made it and keeps verifying after later versions are released.
Verification: Anyone who holds a receipt can check it with RightSignal's standalone verifier, which needs no connection to RightSignal. Verification shows that the receipt has not changed since it was sealed and which key sealed it. It does not show that a decision is correct or that what a site published was true.
06Caution and limits
A hold or a stop is preferred to a wrong collect: Where what a site says is unclear, missing, unreadable or in conflict, the Algorithm is designed to return a hold or a stop. Clients will therefore see holds and stops for some sites from which collection might be permitted on fuller facts or on a different view of the law, and that is intended.
Limits of a decision: A decision speaks only for the site or page, the crawler, the purpose, the profile, the version and the time that it states, and it carries an expiry time. Websites change, can publish a signal wrongly and can block RightSignal's crawler, and the AI reader can miss wording. RightSignal does not promise that any decision, reading or item of evidence is correct, complete or current, and the Terms of Service set out what RightSignal is and is not responsible for.
07What is not published
Confidential rules: This page is the whole of what RightSignal publishes about the Algorithm. Everything else is RightSignal's confidential information and know-how, including the full rules, grades, marks, tables, amendments, mappings, precedence and combination rules, silence tests, thresholds, weights, the scenario bank, measurement data, the readers, prompts, model routing, the policy-page assistant, capture and retry methods and all source code.
Disclosure and restrictions: RightSignal discloses the rules of a specific version only to a client's counsel, an auditor or a court, under a non-disclosure agreement or a court order and for a specific decision, as the IP and Confidentiality Schedule sets out. That Schedule prohibits reverse engineering the service, reconstructing the Algorithm from its outputs, benchmarking it for a competing product, scraping the service and training models on its outputs.