All stories
Work NotesSeptember 14, 20266 min read

Why “missing” is a dangerous research word.

Hermes once surfaced an email as a useful gap even though the prospect’s own website displayed it. That single word hid the difference between not found, not verified, and actually absent.

Editorial representation of Remy reviewing notes at a research desk
Remy reviewing notes at a research deskVisual · editorial representation

A prospect record said the email was missing. The operator opened the website and saw the contact information sitting there in plain text.

That sounds like a small extraction miss. It is more important than that, because the word “missing” can change downstream behavior. Remy may treat the prospect as not send-ready. Research may be repeated unnecessarily. A good lead can sit still because one provider or extractor returned nothing.

Not returned by one extractor is not the same thing as absent from the world.
Editorial representation of Remy arriving at an office
Remy arriving at an officeEditorial representation

Research needs states, not confidence theatre.

For basic contact evidence, Ariy now needs to distinguish at least four conditions: found, not found after a verified search, not verified, and inaccessible. Those states describe what the research actually established instead of collapsing uncertainty into a clean-looking answer.

First-party evidence matters here. Before Hermes concludes that a contact field is absent, the research procedure should inspect visible page content, common contact and footer locations, relevant navigation, mailto links, and ordinary obfuscations such as “help-[at]-example.com.”

The correction has to travel downstream.

Fixing the label in the research UI would not be enough. If the old conclusion already made Remy's opportunity not send-ready, the correction must return to Hermes, produce a new canonical research result, and cause downstream readiness to be recalculated.

This is one reason Ariy separates research from commercial follow-through. Hermes owns evidence and research truth candidates. Remy owns what to do commercially with accepted context. When the evidence changes, the handoff should change too.

Editorial representation of Remy beside an office window
Remy beside an office windowEditorial representation

Operator feedback should be part of the system.

The useful product response is not to make the operator file an engineering ticket every time research is obviously wrong. The operator should be able to attach the correction, point to the evidence, ask for re-research, and let the owning system recover the workflow.

Aletheia can supervise that correction path. She should not silently replace Hermes by conducting research herself. Clear ownership is what makes the correction inspectable instead of magical.

Grounded in

INC-002/Hermes research contract/Remy/Hermes responsibility boundary

Publication note

Story truth: factual. Generated teammate imagery is editorial representation; operational claims resolve to system behavior, evidence, outcomes, or identified interpretation.