In legal review — not indexed

This page describes what Otto does and is readable in full. It stays out of search until it has been reviewed, because a product claim about a compliance control is a claim somebody will hold us to.

Rights Radar — detecting when a consumer invokes a right

Rights Radar reads calls, letters and email for the moment a consumer invokes a right — dispute, cease communication, attorney representation, bankruptcy, identity theft, active-duty status — classifies each one with a confidence score, and applies the corresponding hold on the account rather than waiting for a collector to code it correctly after the call.

Key facts

Six invocation types: dispute, cease communication, attorney represented, bankruptcy, identity theft, and SCRA or active duty.
Every detection carries a source, the phrase that triggered it, a confidence score and a status. High-confidence detections auto-action; the rest are held for human confirmation rather than actioned silently.
A hold applied here is enforced by the same engine that blocks calls, and the hold check runs first — before the client rule, the demand gate, quiet hours or the frequency cap.
The detector as shipped is deterministic and seeded. It sits behind a detector interface so a trained model can replace it without a change to the screen. It is a demonstration of the mechanism, not a model.
Detect-to-enforce timings shown in the module are seeded values, not measurements. No timing instrumentation exists in this build.
Complaints are first-class records rather than email: each carries a response clock, a link to the account and links to the evidence used in the response.

What is Rights Radar looking for?

Six things a consumer can say that change what the firm is allowed to do next. A dispute of the debt or its amount. A demand that communication stop. A statement that they are represented by a lawyer. A bankruptcy filing. An assertion that the account is not theirs and was opened in their name. And a statement of active-duty military service. Each of those has a different consequence and a different legal basis, so they are classified separately rather than collapsed into one flag.

What happens when it detects something?

The detection is written with its source, the phrase, the confidence and the account, and then one of two things happens. Above the confidence threshold, the corresponding hold is applied to the account automatically and the detection is marked auto-actioned. Below it, the detection is queued as needing human confirmation and no hold is applied on the strength of the model alone. In the seeded set, the dispute, cease, attorney and bankruptcy detections auto-action; the SCRA and identity-theft detections wait for a person, which is the right default for two claims that carry heavy consequences in both directions.

Does a hold from Rights Radar actually stop anything?

Yes, because it is not a badge — it is the first condition the call path checks. Before Otto considers the client's outbound rules, the demand gate, the time of day or the frequency count, it looks for a collection hold on the account, and if one is present the call is refused and the refusal names the hold and the source recorded for it. The same hold is a scored factor in suit-readiness and suppresses the credit tradeline in the furnishing file, so the detection reaches every place the account appears rather than only the dialer. TODO: verify the source strings recorded for each hold type against their primary texts before publication.

How good is the detection?

Unknown, and this page will not invent a number. What ships is a deterministic detector over a seeded set of transcripts and documents, written behind an interface so that a production model can be dropped in without the screen changing. That is an honest architectural claim and a dishonest performance claim if it is read as one: nothing here has been measured against a labelled corpus, no precision or recall figure exists, and the confidence scores in the demonstration data are authored values. The design consequence that does hold regardless of model quality is the two-track outcome — a model that is unsure produces a queue item, not an action.

Why is this better than coding it on the call?

It is not a replacement for the collector coding it. It is a second reader that does not get tired, does not triage, and reads the letter that went to the fax machine as well as the call that happened at four in the afternoon. The failure mode it addresses is specific: an invocation that is heard but not recorded, so the account keeps moving because nothing in the system knows. Whether a firm's own quality-assurance process already catches that is a fair question to ask before buying anything.

What does it do with complaints?

Treats them as records rather than as correspondence. A complaint from the CFPB portal, the BBB, a state attorney general or a letter is stored with its received date, its response due date, its status, its link to the account, and the evidence attached to the response — call logs, recordings, letters — pulled straight from the audit trail rather than gathered again. They roll up into the regulator reports rather than being counted by hand. TODO: verify which regulator report formats are actually generated in this build before describing the roll-up as automatic.

This page describes Otto, not the law, and is not legal advice. Otto applies rules that have been written down correctly; it does not warrant that using it results in compliance with any statute, regulation or contract. Where a rule is named here, confirm it against the primary source and your own counsel before relying on it.