Earnings postmortem

Earnings Postmortem (Post-Report Breakdown)

Stage 0 - Decision Memory (load BEFORE forming any verdict)

Before fetching market data, load prior decisions on this ticker so the verdict stays consistent across sessions:

• mcp__aifolimizer__get_ticker_decision_history with ticker=TICKER, max_decisions=5 - prior actions, outcomes, reflections
• mcp__aifolimizer__get_ticker_reflection with symbol=TICKER, n=3 - prior recs + realized alpha
• mcp__aifolimizer__get_cross_ticker_lessons with max_lessons=3 - portfolio-level win/loss patterns

Reconciliation rule: if a prior decision exists and your new read flips it, state explicitly WHY it changed (new data / catalyst / price move). Never silently contradict a logged decision - that drift is exactly what this prevents.

How to run

1. Call mcp__aifolimizer__get_profile - account types and capital. Frame any post-report trim/add decision and tax impact
2. Confirm ticker via mcp__aifolimizer__get_portfolio - current weight + cost basis (or note "not held" if researching watchlist name)
3. Call mcp__aifolimizer__get_earnings_results with symbols=[ticker], quarters=4 - last 4 quarters EPS actual vs estimate, surprise %, beat/meet/miss outcome
4. Call mcp__aifolimizer__get_fundamentals with symbols=[ticker] - current P/E, forward P/E, EPS TTM, analyst target, recommendation, profit margin, revenue growth
5. Call mcp__aifolimizer__get_news_headlines with ticker=ticker - post-report news + analyst reactions
6. Call mcp__aifolimizer__get_recent_filings with ticker=ticker, forms=["8-K","10-Q","10-K"] - the actual results 8-K + the 10-Q/10-K with a direct EDGAR link; confirms the filing exists and timestamps it (primary source for section 2 numbers). US-listed only
7. WebSearch only for: full earnings call transcript quotes, segment revenue breakdown, forward guidance text, options-implied move that already played out, sell-side rating changes post-report

Investor profile

• Canadian retail investor
• Time horizons: short-term trading + long-term (10yr+) holding
• Account types and capital: always read from get_profile - never hardcode

Output structure

1. Verdict (top, one paragraph)

• Beat / Meet / Miss on EPS (from get_earnings_results most recent quarter)
• Stock reaction summary (from news + price move if available)
• Thesis change: confirmed / weakened / broken
• Action: Hold / Add / Trim / Exit - with tax-account framing

2. Headline numbers

• This quarter EPS actual vs estimate, surprise %
• Revenue actual vs estimate (from WebSearch if not in MCP - clearly label source)
• YoY growth on revenue and EPS

3. Last 4 quarters trend

Render markdown table. Columns: Quarter | EPS Estimate | EPS Actual | Surprise % | Outcome. Below table: one sentence on pattern (improving beats, deteriorating, choppy).

4. Management commentary signal

• Call mcp__aifolimizer__get_earnings_commentary with ticker=<TICKER> FIRST. Lead with mgmt_tone_trend (improving/deteriorating/stable vs prior quarters) - this is the predictive field. IGNORE the absolute mgmt_tone/mgmt_tone_signal in isolation: prepared remarks are always promotional, so absolute tone is near-1.0 every quarter. primary_metric shows the trend source (av_sentiment = AV's LLM model, preferred; lexicon_tone = EDGAR-only). Probabilistic drift read, NOT a verdict: improving trend agreeing with a beat strengthens the upward-drift thesis; deteriorating after a beat (tone worse despite good numbers) is a strong caution flag. If relative is false (only one quarter / non-US) or source is None, fall back to WebSearch.
• Guidance: raised / maintained / lowered (commentary tool gives language tone; WebSearch for the exact guidance numbers - yfinance doesn't carry guidance)
• Key segment commentary (e.g., cloud growth, ad revenue, subscriber net adds)
• New risks called out by management
• Capital allocation changes (buybacks, dividend, capex)

5. Analyst reaction

• Upgrades/downgrades since report (from headlines + WebSearch)
• Price target revisions: pre vs post
• Consensus shift

6. Valuation re-rate

• Forward P/E now vs pre-report
• Did multiple expand or compress on print?
• Is current price implying guidance is credible?

7. Recommendation (Canadian tax aware)

• Before emitting any Add/BUY verdict, call mcp__aifolimizer__get_positioning_signals with symbols=[ticker] - a beat often pulls the crowd into the post-print reaction. If crowding_score>=70 (consensus-crowded), downgrade conviction or defer the add (negative expected alpha for late entries)
• If user holds: hold / add / trim / exit with reasoning tied to cost basis
• If user doesn't hold: initiate / wait / pass
• Account placement (TFSA / RRSP / Non-Reg) - same framework as stock-analysis
• For Non-Reg trims, flag capital gains realization explicitly

After output - log decision

Call mcp__aifolimizer__log_recommendation with action (BUY/HOLD/SELL/ADD/TRIM/PASS), conviction (HIGH/MED/LOW), target_pct + stop_pct (% from entry - entry is captured live), 1-line rationale citing thesis-change verdict (confirmed / weakened / broken), skill="earnings-postmortem". Feeds forward win-rate / track-record loop.

Rules

• Decision summary at very top
• Always render 4-quarter trend table - verbal-only not acceptable
• Under 600 words
• Cite cost basis if user holds ticker
• Never invent EPS or revenue figures - quote from get_earnings_results or state "WebSearch: <source, date>"
• Forward guidance MUST come from WebSearch (yfinance has no guidance field) - do not fabricate

Gotchas

• get_earnings_results returns yfinance earnings_history - EPS only, no revenue. Revenue beats/misses require WebSearch (earnings press release or transcript)
• yfinance surprisePercent is decimal (0.05 = 5%) - service already multiplies by 100, but verify sign on negative surprises
• For .TO tickers, yfinance earnings_history sparse - many TSX names return empty. Note "TSX coverage gap", rely on WebSearch with company's IR release
• "outcome" field is strict EPS-only beat/miss - company can beat on EPS via buybacks while missing on revenue. Always look at revenue separately
• "Stock reaction" requires post-report price move - get_technicals cached 1h, may not reflect fresh print. Note timestamp
• Pre-earnings consensus revisions matter: "beat" against lowered estimate weaker than beat against raised estimate. Flag if WebSearch shows estimates cut in 2 weeks before report
• Guidance shift is dominant signal, not headline beat. Beat with guide-down is sell catalyst; small miss with guide-up is buy
• Crypto holdings: skill inapplicable - redirect to crypto-specific analysis