ic-memo

Load the us-equity-ic-rigor skill and begin the IC memo build pipeline on the ticker provided.

If a ticker is provided as the first argument, use it. Otherwise ask the user which name and what thesis direction (long, short, or pair-trade structure).

Workflow per SKILL.md Phases 0-3:

Phase 0 — delegate to us-equity-research (run /us-equity-research:research {TICKER} first if those outputs don't already exist in outputs/). Phase 0 produces raw research; do not skip.

Phase 1 — source stratification gate. Every specific number must be classified S1 (audited 10-K/20-F), S2 (unaudited filing: 10-Q/8-K/DEF 14A/S-1/13D/13F/Form 4/6-K), S3 (earnings call transcript / mgmt guidance / Investor Day), S4 (sell-side consensus median + dispersion — not mean), S5 (alt-data/expert network with provider + methodology + sample + freshness), or Pending (do NOT promote to headline anchor). G7 hard rule: if any of the top-3 anchors is S3 or weaker, headline must be source-conditional.

Phase 2 — five-scenario probabilistic framework (strong_bear / bear / base / bull / strong_bull). Probabilities must sum to 1.00 (G4 enforces). Bear bridge must reconcile to base via line-item delta walk (G5).

Phase 3 — three-method valuation reconcile: DCF + Comps + multi-multiple bear floor + SOTP (per D8 sector branching — P/E mature, EV/EBITDA leveraged, EV/ARR + Rule of 40 SaaS, P/B + ROE banks, P/AFFO + NAV REITs, NPV pipeline biotech, P/AUM asset managers). EPS × multiple rows must multiply exactly (G1). Segment GM must reconcile to consolidated within tolerance per gm-taxonomy-us.md (G2 — ±50bp industrials, ±5bp banks NIM per D24). SOTP NI ≤ Pre-Tax ≤ PPNR ≤ Total Revenue (G3 monotonicity). Mandatory quant overlay per D13 — Barra factor exposure (Value/Quality/Momentum/Growth/Size/Low-Vol/Liquidity) per G13, capacity/days-to-exit at 10/20/30% ADV per G14. v0.2.0+ also requires consensus variance disclosure (G15), bank discipline if sector=Banks (G16), and revision velocity (G17). v0.3.0+ requires quant cross-doc consistency (G18), provenance manifest (G19), and view defensibility (G20) for any score claim above 8.5.

Output (v0.5.0+ consolidated convention): at the end of Phase 3, write these files to outputs/:

• {TICKER}_IC_memo.md — institutional Markdown memo, English-only per D4
• {TICKER}_structured.json — consolidated structured representation containing memo_metadata, recommendation, source_tags, scenarios, consensus_variance, quant_overlay, gm_taxonomy, bear_eps_bridge, what_would_reverse, position_sizing, and any other domain-specific blocks (banks, biotech, REITs). All sub-objects live inline in this single JSON file rather than in separate sidecar files.
• {TICKER}_manifest.json — provenance manifest written by ${CLAUDE_PLUGIN_ROOT}/scripts/write_manifest.py containing run_id, phase_timing, agent_provenance (≥15 agents per D9), web_search_log (≥12 entries per D9), verification_calls_count, output file SHA-256 hashes, attacker_model + isolation flag for R-v2.

(The prior v0.4.x convention of separate _source_tags.json and _scenarios.json sidecar files is grandfathered for backward compatibility; v0.5.0+ memos use the consolidated structure.)

After the Markdown + structured JSON + manifest are written, run all 20 verification gate scripts (${CLAUDE_PLUGIN_ROOT}/scripts/verify_*.py) against the structured JSON. Write the gate-run audit trail to outputs/{TICKER}_verification_gates.json conforming to ${CLAUDE_PLUGIN_ROOT}/schemas/verification_gates.json. Report pass/fail per gate. If any gate fails, fix and re-run before declaring the memo done.

If a verifier script cannot be reached (file not found, permission denied, import error), DO NOT degrade to LLM-analytical gate evaluation. STOP, report the error to the user with the exact command and exact error message, and follow the SKILL.md anti-degradation preamble protocol.

Then prompt the user: "Run /us-equity-ic-rigor:red-team {TICKER} to score this memo against the PM rubric?"