Candidate names become detached
An agent can summarize a part number while losing the dimensions, ratings, or family fields that made the candidate worth review.
Agent-readable result schema
AI-assisted replacement review needs structured results, not pasted catalog fragments.
When Codex or Claude Code helps with industrial part search, the result should preserve the source part, candidate paths, evidence fields, missing checks, and history. Partglyph gives agents a cleaner structure to summarize and humans a cleaner result to review.
Unstructured result risk
A loose AI answer can lose the evidence that made the candidate appear useful.
The problem is not only search quality. It is preservation. If candidate fields, missing checks, and review context disappear, the user has to rebuild the decision trail. In replacement work, the useful answer is not just a candidate name; it is the reason that candidate surfaced, the fields that support it, the fields that weaken it, and the next questions that keep the review controlled.
Missing checks get buried
Open questions need to be visible as next actions, not hidden inside a paragraph that sounds more confident than the evidence supports.
History is hard to reuse
Without structured results and history, the next maintenance search repeats the same work under the same pressure.
Public schema concept
The useful result structure keeps every review layer visible.
This page explains the public concept, not a fixed internal product contract. The practical value is that agents and humans can follow the same review trail. A structured result makes the agent a better assistant because it can summarize evidence without detaching a part number from the comparison fields that matter.
Source identity reference manufacturer, part number, product family, known fields
Candidate paths ranked options, evidence fields, visible differences, weaker paths
Review metadata missing checks, next questions, history, credit-tracked result detail
Agent workflow
A structured result lets agents explain without inventing certainty.
The agent can summarize, compare, and route the result because Partglyph keeps the review evidence organized.
Read the source
The agent starts from the reference part identity and product family rather than a generic search phrase.
Summarize candidates
It lists the candidate paths and the fields that explain why those paths appeared.
Name gaps
It keeps missing checks visible so the user can request better supplier or engineering evidence.
Preserve context
It points back to the result detail and history instead of turning the review into disposable chat text.
Readable result layers
Agents need the same evidence layers human reviewers need.
A good result schema makes the candidate easier to inspect, not easier to over-trust.
Reference source part identity and family context stay visible Open page Evidence dimensions, ratings, materials, interface, and family checks stay attached Open page Candidate ranked paths can be summarized without hiding differences Open examples Gaps missing fields become next checks, not silent assumptions Open page Workflow history and MCP access help agents continue the review Open MCP page
Structured trust
Agent-readable does not mean agent-approved.
The value is that Codex or Claude Code can work from a cleaner result and preserve evidence. The review still belongs to the user, the team, and the plant's engineering controls. This is why the schema keeps source identity, candidate evidence, missing checks, and history visible instead of compressing the review into one confident paragraph.
Use structured results
Run Partglyph where agent-assisted search needs evidence that survives the chat.
Use the app or MCP workflow to return candidate evidence, missing checks, and result history in a structure Codex or Claude Code can summarize without replacing the review.