Candidate Mainline Ranker

Rank current mainline directions across a candidate professor set. This skill runs after one or more professor-direction-mapper outputs exist. It answers: "Which shared directions are most worth reading first because they recur as current mainlines across the candidate pool?"

It does not decide whether any single professor is a fit. It aggregates direction signals across candidates.

Inputs

Preferred:

• A JSON file following references/input-schema.md.

Accepted:

• A directory of direction maps, if Codex can extract the same fields.
• A Markdown candidate table, if directions and ranks are explicit enough.
• Manual records pasted by the user.

Required per professor:

• Professor/lab name.
• Pool: primary, extended, or historical.
• Current mainline directions with order/rank.

Recommended per direction:

• Normalized direction label.
• Raw direction text.
• Rank within that professor's current mainlines.
• Confidence: high / medium / low.
• Evidence status: complete / sampled / weak / stale.
• Source direction-map path or URL.

Default Weights

Use transparent additive/multiplicative scoring:

• Base occurrence: 1.0.
• Rank-1 current mainline bonus: +0.75.
• Rank-2 current mainline bonus: +0.35.
• Rank-3+ current mainline bonus: +0.0.
• Pool multiplier:
  • primary: 1.0
  • extended: 0.6
  • historical-only: 0.25
• Candidate priority multiplier:
  • P1: 1.0
  • P2: 0.8
  • P3: 0.6
  • E1: 0.6
  • E2: 0.45
  • E3: 0.3
  • unknown: 0.5
• Confidence multiplier:
  • high: 1.0
  • medium: 0.75
  • low: 0.45
• Evidence status multiplier:
  • complete: 1.0
  • sampled: 0.8
  • weak: 0.55
  • stale: 0.35
  • unknown: 0.7

Formula per professor-direction record:

score = (base_occurrence + rank_bonus) * pool_multiplier * priority_multiplier * confidence_multiplier * evidence_multiplier

The default formula is a decision aid, not a truth machine. If the user gives a different weighting policy, use that policy and state it.

Workflow

1. Collect direction records.

   • Prefer current mainlines from professor-direction-mapper outputs.
   • Include extended candidates if requested; label them as lower-priority evidence through pool/priority multipliers.
   • Exclude historical mainlines unless the user asks for historical influence analysis.

2. Normalize direction labels.

   • Merge obvious aliases, such as RAG, retrieval-augmented generation, and LLM retrieval.
   • Keep a raw-label list for audit.
   • Do not over-merge broad fields. For example, LLM agents, RAG, and LLM reasoning may be adjacent but should remain separate unless evidence shows they are the same reading direction.

3. Score directions.

   • Use scripts/rank_mainlines.py when a JSON input is available.
   • Otherwise calculate manually using the default weights and show the scoring table.
   • Report both weighted score and raw professor count.

4. Interpret the ranking.

   • Prioritize directions that have high weighted score, multiple professors, and evidence from primary candidates.
   • Flag directions dominated by one professor, one extended candidate, weak evidence, or aliasing uncertainty.
   • Recommend the first 1-3 directions to read and explain why.

5. Return a durable output.

   • Use references/output-template.md.
   • Include scoring policy, ranked table, raw records, caveats, and recommended next reading actions.

Guardrails

• Do not rank historical mainlines as current priorities unless explicitly requested.
• Do not let a broad umbrella label swallow distinct reading directions.
• Do not hide that a direction is driven by only one professor.
• Do not treat the highest weighted direction as automatically the best personal fit.
• Do not compare directions without showing the weighting policy.
• Do not count a direction twice for the same professor unless the direction appears as two genuinely distinct current mainlines.

Resources

• scripts/rank_mainlines.py: score a JSON input and emit Markdown plus optional JSON.
• references/input-schema.md: input JSON schema and example.
• references/output-template.md: recommended human-readable report structure.