Update to Equity Council

{
  "description": "Independently challenge an equity thesis or company ranking with counterevidence, falsifiable tests, rival explanations and economically plausible pairwise rank reversals.",
  "included_files": [
    {
      "relative_path": "agents/openai.yaml",
      "size_in_bytes": 240
    }
  ],
  "name": "equity-red-team",
  "skill_md_contents": "---\nname: equity-red-team\ndescription: Independently challenge an equity thesis or company ranking with counterevidence, falsifiable tests, rival explanations and economically plausible pairwise rank reversals.\n---\n\n# Equity Red Team\n\nTest the strongest available investment case without trying to manufacture a negative verdict. Read the [operating contract](../../references/operating-contract.md), [ranking method](../../references/ranking-method.md), [evidence standard](../../references/evidence-standard.md), and [independence and reconciliation protocol](../../references/agent-protocol.md).\n\n## Independent first pass\n\nWhen available, receive a neutral evidence packet, mandate, cutoff, source locations and bounded questions before seeing the favored company or draft ranking. Record initial causal objections and alternative interpretations before comparing with the draft. If already exposed to the conclusion, disclose that limitation; do not claim a blind review. A single agent can apply the lens sequentially, but it is not an independent agent review. Shared sources remain shared evidence even when several analysts agree.\n\n## Challenge workflow\n\n1. Reconstruct the strongest thesis fairly: what must hold for operating performance, reinvestment, financing, per-share value and market expectations? Separate a good business from a good investment at this price. Identify which assumptions actually drive the ranking rather than criticizing peripheral prose.\n2. Seek decisive counterevidence from original sources and credible rival explanations. Examine selection bias in the candidate universe, customer economics, competitive responses, substitution, profit capture, reinvestment, financing and dilution. Check whether purported corroboration repeats management's claims. Absence of discovered contradiction is not positive confirmation.\n3. Turn each material objection into a falsifiable statement: observation, metric definition, threshold or range, time window, source, and resulting model change. Prefer operational evidence to stock-price movement as a test of business mechanics. Distinguish a thesis failure from a temporary miss or a valuation-only disappointment.\n4. Challenge the leader against the closest credible alternative under the declared objective. Hold definitions, entry dates, benchmark treatment and common external shocks consistent. Find the smallest economically plausible changes in growth, margin, reinvestment, dilution, entry price, terminal valuation or probabilities that reverse adjacent ranks. Show the calculation, not just a harsher adjective. A single-variable threshold is local sensitivity; add a coherent joint case where drivers interact.\n5. Test whether favorable tails dominate expected wealth, survivorship is assumed, failure states are omitted, or confidence scores masquerade as probabilities. Keep return, success probability, downside and evidence strength separate. Compare the industry winner with the broad-market alternative and allow a tie, conditional rank or no qualifier.\n6. Reconcile objections by type: fact, definition, causal mechanism, forecast or preference. Correct factual errors from evidence; retain coherent forecast disagreements as sensitivities. Do not settle disputes by voting or force equal weight on a weak countercase.\n\n## Output contract\n\nReturn the strongest counterthesis, supported objections ranked by decision impact, evidence IDs and lineage, pairwise reversal calculations, falsification tests, and what evidence would defeat the counterthesis itself. Separate resolved corrections from residual dissent; identify affected model inputs and suggested status changes. State review independence and coverage honestly. Stop when decisive objections are resolved or explicitly represented in uncertainty; hand adjudication to the director without rewriting shared models or selecting a winner outside the assignment.\n"
}