Update to Equity Council

{
  "description": "Use when revisiting a saved Equity Council industry ranking, updating an investment thesis after new evidence, or assessing recorded forecasts against resolved outcomes.",
  "included_files": [
    {
      "relative_path": "agents/openai.yaml",
      "size_in_bytes": 236
    }
  ],
  "name": "equity-update",
  "skill_md_contents": "---\nname: equity-update\ndescription: Use when revisiting a saved Equity Council industry ranking, updating an investment thesis after new evidence, or assessing recorded forecasts against resolved outcomes.\n---\n\n# Thesis update and forecast accountability\n\nRead the prior run's mandate, plan, source cutoff, assumptions, scenarios, report and forecast ledger. Read the [operating contract](../../references/operating-contract.md) and [evidence standard](../../references/evidence-standard.md). Preserve the original artifacts. Create a new dated run or version; do not rewrite past predictions to fit observed results.\n\n## Procedure\n\n1. State the update plan: new evidence to collect, changed prices/corporate actions, affected assumptions, relevant specialists, recalculation and verification steps. Retain the prior mandate unless the user changes it. Explain any changed industry boundaries, benchmark or ranking policy.\n2. Refresh current prices, latest filings and subsequent material events. Identify evidence revisions separately from new events. Compare the information available at the old cutoff with what became available later; later restatements must not leak into a claimed historical forecast.\n3. Build a change table: old assumption, new evidence, revised assumption, affected cash flows/per-share value, and rank impact. Distinguish business change, price change, modeling correction and user-preference change.\n4. Route only affected work to the relevant specialist skills. Reassess previously rejected companies if their economics or evidence changed. Keep an unchanged thesis unchanged only after checking its key dependencies.\n5. Rerun current-price scenarios and meaningful sensitivity sets. Reaudit changed inputs and calculations; do not rerun unchanged work for ceremony. Show old vs new rank, confidence, entry valuation and thesis-break conditions.\n\n## Forecast scoring\n\nUse immutable event definitions: metric, entity, period, threshold, units, accounting basis, deadline, resolution source and original probability. Score only resolved events with comparable observed data. For a binary outcome y in {0,1}, Brier score is `(p-y)^2`; report the mean across a disclosed sample. Keep unresolved or ambiguous outcomes separate and explain missing coverage.\n\nDo not use a later period, different metric or revised event definition to turn a miss into a success. Record revisions explicitly. One outcome cannot establish calibration; dependent events and small samples limit interpretation. Evaluate probability bins only with adequate observations and uncertainty. Keep forecast skill, accounting accuracy and investment returns distinct.\n\nFor actual or paper investment performance, account for splits, distributions, delistings, entry timing, costs, benchmark and horizon. Preserve point-in-time universe membership and all tested strategy variants if backtesting; a selected attractive historical result is not validation.\n\n## Deliver\n\nProvide the refreshed ranking or conditional status; what changed and why; updated model/evidence links; resolved forecast scores with sample limitations; original errors and their corrections; new thesis tests; and the next evidence that could change the decision. A monitoring checklist is a deliverable, not an automation. Create a scheduled follow-up only when the user separately asks and the host supports it.\n"
}