Job Search profile review

Status: Implemented in #2373. This roadmap page is retained as the original design description.

Job Search is being extended with a guided profile-review flow for cases where technically valid results still do not match what the user actually wants.

The goal is to make the search profile easier to improve without requiring the user to reverse-engineer why a result appeared.

Implemented flow

The profile editor will be able to review the current profile together with a bounded snapshot of recent search outcomes.

The review is intended to:

1. Summarize what the saved profile currently appears to prioritize.
2. Point out likely gaps, conflicting signals or areas that are too broad.
3. Use recent returned and dismissed listings as evidence of where the search may be drifting.
4. Ask focused questions such as what the user wants more of, what should be deprioritized and what the search appears to have missed.
5. Feed the user's answers into the existing AI instruction helper.
6. Return richer instruction drafts that can be reviewed before they are saved.

The assistant remains proposal-only. A review, answer or AI suggestion must never silently rewrite the search profile.

Better instruction proposals

The existing Job Search instruction helper will produce longer, more actionable alternatives rather than short generic guidance.

Each proposal should clearly express:

• target role families and credible adjacent roles;
• strongest evidence-backed skills and responsibilities;
• hard requirements versus preferences;
• remote, hybrid, local and commuting priorities when configured;
• technologies or duties that should increase relevance;
• roles, employers, sources or patterns that should reduce relevance;
• how broadly the search should explore beyond literal titles and keywords;
• what kinds of mismatches the search should avoid repeating.

Saved keywords, locations, exclusions and relevance rules remain authoritative inputs. CV and cover-letter material may be used as private candidate evidence under the existing Job Search privacy rules.

Search-result analysis

The review is not intended to claim that a returned job is objectively bad. It should explain why the current profile could reasonably have matched it and identify which profile signal could be refined.

Recent data is bounded and profile-scoped. The review may use safe stored listing fields and dismissal state, but it must not expose another profile, another user or raw provider payloads.

Instruction history and restore

Changing instructions will create a private revision history for the selected search profile.

The owner should be able to:

• inspect recent instruction revisions;
• see when a revision was saved, cleared or restored;
• restore an earlier snapshot explicitly;
• keep the restored value editable before making further changes.

Restoring a revision is a real profile configuration change and therefore follows the same Job Search configuration-revision and audit rules as an ordinary instruction edit.

Privacy and authorization

The review and history are owner-scoped.

AI request logs keep only safe request metadata such as the request mode and profile identifier. Raw CV text, cover-letter text, saved instructions, user answers, review output and generated instruction drafts are not copied into ordinary provider request/response logs.

Administrator editing remains a separate authorized surface and must not weaken the owner API.

Internal implementation tracking: #2372.