captain23333/find-your-perfect-advisor/advisor-evaluator

Advisor Evaluator

Produce a decision aid and application-ready workbook without collapsing objective application conditions and subjective advisor suitability into one opaque score.

Read:

• outputs/advisor_records.json
• outputs/program_records.json
• outputs/evidence.json
• outputs/candidates.json
• outputs/matching-audit.json
• project.json

If present, read ../advisor-pipeline/references/core-data-contract.md. Read references/workbook-contract.md before building or validating the workbook.

Alignment

Join records by stable advisor_program_id, not fuzzy advisor-name matching. Preserve multiple programs for one advisor as separate application rows.

Check:

• Every row maps to a real advisor and official program.
• Objective facts are current for the requested intake.
• Detective results cover the same selected sections for compared advisors.
• not_checked, not_found, access failure, and conflicting evidence remain

distinct.

Decision model

Show seven separate layers:

1. Research fit: numeric, formula-driven, and sourced from Finder.
2. Profile match: applicant-relative methods, work, prerequisites, and

transferable skills from the real CV.

1. Hard constraints: pass, fail, or unknown; unknown is not pass.
2. Application pathway: supervisor-led, committee-led, advertised

position, structured program, or unknown.

1. Opportunity evidence: verified open, signal only, unknown, or verified

closed.

1. Objective feasibility: eligible, ineligible, or

needs_confirmation, with explicit reasons.

1. Advisor suitability: selected-section findings, supported risks, and

confidence.

Ask for optional weights only for numeric selected dimensions. Normalize weights over dimensions consistently selected for the compared advisors. Do not score an unselected or merely unavailable section as zero.

Anonymous leads cannot directly change a score. Display a supported severe risk separately even when an overall numeric result is high.

Priority

Recommend application priority using transparent rules:

• Do not recommend an objectively ineligible row as a primary application.
• Exclude a verified hard-condition failure or verified-closed opportunity.
• Keep unresolved feasibility visible instead of silently filtering it.
• Route the next action using the official application pathway; a missing

faculty reply is not a rejection for committee-led or structured programs.

• Prefer strong research fit when eligibility is comparable.
• Surface verified severe risks before total scores.
• Explain the recommendation in plain language and list the next verification

action.

Workbook

Generate advisor_application_ready_YYYYMMDD.xlsx using the deterministic builder. It automatically uses the Codex spreadsheet runtime when available and portable OOXML otherwise. Do not install packages or create and patch an ad-hoc workbook builder. The primary sheet must be usable without manually joining other sheets.

Include:

• School, QS edition, program, degree/intake, official link, deadline, tuition,

scholarships, materials, and RP requirement.

• Advisor research/papers, email, homepage, and multiline recruiting/contact

requirements.

• Research fit, profile match, overall match, competitiveness, hard-condition

status/reasons, application pathway, opportunity status, recommended action, objective feasibility, selected backcheck result, supported risks, gaps, official sources, and last verified date.

Use separate sheets for fit, evidence, sources/freshness, and configuration.

Verification

• Scan for duplicate advisor_program_id rows.
• Verify typed numbers, dates, formulas, filters, frozen panes, wrapping, and

conditional formatting.

• Scan formulas for #REF!, #DIV/0!, #VALUE!, #NAME?, and #N/A.
• Confirm every material row contains official application sources and access

dates.

• Stop rather than fabricate missing application facts.