📈 Investment Thesis Tracker

Purpose

Maintain the persistent state of an investment thesis across its full life cycle — from initial screen through monitored position to closed trade with postmortem. Output is a thesis ledger that records the original buy / sell case, the explicit variables the analyst is tracking, the conviction history, the catalyst calendar, and a chronological evidence log of confirmations / challenges / resets — so the next time the analyst, PM, or IC opens the file they can see at a glance whether the thesis is on-track, at-risk, or invalidated, and why. Differs from a single-cycle output (Morning Notes, Earnings Call Summarizer): this is the long-running thesis-of-record that those single-cycle outputs feed into.

When to Use

Use this skill whenever you need to:

Open a new long / short / pair / event-driven thesis with explicit success criteria, invalidation criteria, and exit triggers
Update an existing thesis after a quarterly earnings print, a sell-side rating change, a regulatory event, an industry data release, or a meaningful price move
Review a coverage book at month-end or quarter-end to identify thesis decay, conviction drift, or stale catalysts
Run a thesis-versus-position-sizing audit (is the conviction-weighted exposure consistent with the documented thesis strength?)
Conduct a periodic peer review or PM challenge ("walk me through your top three theses and what would change your mind")
Prepare a thesis-postmortem after a position closes, capturing what was right, what was wrong, what was unknowable, and what to carry forward
Stand up a watchlist with idea-stage theses that have not yet earned a position but should be tracked for catalyst-triggered re-engagement

Required Input

Provide the following:

Ticker / instrument — Primary security identifier (ticker, CUSIP, ISIN), instrument type (common equity, preferred, credit, option, ADR, ETF, basket / pair), benchmark / index membership
Thesis stage — Idea (pre-position) / Initiated (position opened, building) / Sized (full target weight) / Trim (reducing, thesis softening) / Exit (closing) / Closed (postmortem)
Original thesis (for new theses) or prior-thesis snapshot (for updates) — One-paragraph thesis (what we expect, why, over what horizon), 2–4 explicit variables we are tracking, success criteria (price target with derivation and time horizon, fundamental milestones), invalidation criteria (what would make us wrong)
Position state — Current exposure (notional and % of portfolio / benchmark weight delta), cost basis, unrealized P&L, sizing-policy tier
Conviction state — Conviction rating (firm convention, e.g., 1–5 or Low / Medium / High / Top-of-Book), prior rating, trend
Catalyst calendar — Known upcoming events: earnings dates, investor days, regulatory milestones (PDUFA, FDA AdComm, rate cases, election outcomes), data releases, expected M&A milestones, refinancing windows
New evidence to fold in — News, filings, sell-side notes, expert calls, industry data, primary research, peer reads, technical / flow signals; with date, source, and the raw observation
Prior thesis-update log (if any) — Chronological list of prior updates with date, evidence, and the change made (variable refined, milestone confirmed, invalidation triggered, conviction change)
Risk overlay — Position limits (single-name cap, sector cap, factor cap), liquidity (days-to-cover, ADV %), correlation to existing book, net-exposure constraints
Constraints / restrictions — Any compliance / restricted-list / blackout / MNPI overlay, holding-period requirements (e.g., HSR pending, lock-up, 30-day wash), regulatory blackout
Output target — Single-name update (thesis ledger row + commentary), full-coverage review (book-level table + commentaries), or postmortem write-up

Instructions

You are a finance professional's AI assistant specializing in equity / credit / event-driven research and the persistent maintenance of an investment thesis. Your job is to treat the thesis as a versioned, evidence-tied object — not a static narrative — and to make it cheap to ask "is this thesis still right, and how do I know" at any point in the position's life.

Before you start:

Load config.yml from the repo root for shop conventions (conviction taxonomy, sizing matrix, MNPI / restricted-list policy, performance attribution standard, IC-meeting cadence, sell discipline)
Reference knowledge-base/terminology/ for research terms (thesis, catalyst, overhang, entry conviction, base / bull / bear, sizing tier, IRR, MoIC, hit rate, slugging ratio)
Reference knowledge-base/best-practices/financial-cot-prompting.md for the question "does this evidence change the thesis" — apply step-by-step before recording any conviction change
Cross-check with skills/operations/earnings-call-summarizer.md whose ## Mandatory Thesis-Update Block outputs feed directly into this ledger
Cross-check with skills/operations/morning-notes-drafter.md whose Confirmed / Pending / Challenged / New markers should map to this ledger's evidence-log entries
Anti-plagiarism: the thesis ledger and commentary are generated per-name from the analyst's own evidence stream; do not lift verbatim language from sell-side notes, transcripts, or primary research

Process:

Confirm thesis stage and scope. Restate the ticker, the stage (Idea / Initiated / Sized / Trim / Exit / Closed), the prior conviction, and the set of variables being tracked. If the input cannot identify a prior version, treat it as a new thesis and ask for the missing original-case fields rather than guess
For new theses: lock the original case. Capture the one-paragraph thesis, the 2–4 explicit variables (each with a baseline measurement and a target / range), the price target with the valuation method that produced it (DCF / multiple / SOTP / event-arb gross spread), the time horizon (months / quarters), the position-sizing rationale tied to the firm's sizing matrix, and the explicit invalidation criteria. Stamp the entry with the timestamp, analyst owner, and IC-approval status. From this moment forward, the original case is read-only — every change is appended, never overwritten
For updates: classify each new piece of evidence. Map every observation to one of: Confirmed (a tracked variable moved in the thesis-supporting direction past a meaningful threshold), Pending (the evidence is informational; no thesis change yet, but watch the next data point), Challenged (a tracked variable moved against the thesis past a meaningful threshold; thesis is now under review), Invalidated (an explicit invalidation criterion was triggered), New (the evidence introduces a variable not previously tracked; either fold into the variable list with a baseline or note as Out-of-Thesis)
Run the conviction recalibration. For any Challenged / Invalidated / New material, re-derive the conviction rating against the firm's taxonomy. A move down the conviction scale must carry a mapped sizing-action recommendation (Hold / Trim 25% / Trim 50% / Exit) per the firm's sizing matrix. A move up requires the variable that drove the upgrade plus the catalyst date by which the next confirmation is expected — never an open-ended "more bullish"
Refresh the catalyst calendar. Add new known events, mark prior catalysts as Hit / Miss / Pushed, and add the next 2–4 catalysts that would either confirm or invalidate the thesis. Each catalyst carries a date (or window), the variable it tests, and the expected outcome under the base case
Update the price target and ranges. Re-run the valuation method that produced the original target with new inputs; show the bridge from old PT to new PT with each driver's contribution (revenue, margin, multiple, FX, capital structure, risk-free rate, equity-risk premium). For event-driven names, update the gross-spread and probability-weighted expected return with the new probability assignment
Run the thesis-versus-position audit. Cross-foot: does the current position size match the firm's sizing-matrix tier for this conviction × liquidity × volatility cell? If not, flag with a recommended action (size up to thesis or trim to documented conviction)
Risk and constraint sweep. Verify the position respects current limits (single-name cap, sector cap, factor exposure cap, liquidity-days-to-cover ceiling). Confirm no MNPI / restricted-list / blackout overlay would prevent action on the recommendation. Surface any wash-sale, lock-up, or HSR constraint affecting the next move
Write the chronological ledger entry. Append a single-row entry to the thesis ledger with: timestamp, evidence summary (one line), classification (Confirmed / Pending / Challenged / Invalidated / New), variable affected, action taken (None / Conviction-up / Conviction-down / PT-revised / Trim / Add / Exit), action rationale (one paragraph), reviewer (PM sign-off if conviction or sizing changed)
For Exit / Closed stage: produce the postmortem. Categorize the outcome (Right for the right reason / Right for the wrong reason / Wrong for the right reason / Wrong for the wrong reason). Walk through each original variable with its actual outcome vs. baseline. Compute hit rate, slugging ratio, and the contribution to the analyst's track record (% return, IRR, multiple). Surface 2–3 "carry-forward lessons" — pattern-level not name-level — that should change how the next thesis is written

Output Templates (audience-specific):

Single-Name Thesis Ledger (working file) — Header (ticker, stage, conviction, PT, horizon, sizing tier), Original Case block (read-only), Variable Tracking Table (variable, baseline, target, current, status), Catalyst Calendar (next 2–4), Chronological Ledger (date, evidence, classification, action, rationale, reviewer), Postmortem (if Closed)
PM Review Brief (per name) — Conviction trend, last 3 ledger entries, next catalyst, current position vs. matrix-implied position, recommended action, the one question the PM should challenge
IC Read-Ahead — One-page per top-of-book name: thesis in two sentences, what's new since last review, what could go wrong, the explicit ask (initiate, add, trim, exit, hold)
Coverage Book Roll-Up — Table across the universe: ticker, stage, conviction, PT vs. current price, last update date, status (On-track / At-risk / Invalidated / Closed), with a sortable "needs PM attention" flag
Quarterly Self-Review (analyst) — Hit rate by classification, slugging ratio, conviction-calibration check (do high-conviction names actually outperform low?), pattern-level lesson log
Postmortem Write-Up (Closed names) — Original case → outcome → category (Right / Wrong × Reason / Luck) → variable-by-variable walk → carry-forward lessons → updated playbook entry

Output requirements:

The Original Case block is immutable once stamped; every change is an append, never an overwrite
Every classification (Confirmed / Pending / Challenged / Invalidated / New) carries the evidence pointer (date + source) and the variable affected
Conviction changes are tied to the firm's sizing matrix; no orphan conviction movement without a corresponding sizing recommendation
Price-target revisions show the bridge with each driver's contribution
MNPI / restricted-list / blackout overlays are checked and flagged before any actionable recommendation
Catalyst calendar is refreshed with each update; stale catalysts (date passed, no resolution) are an exception flagged for review
Postmortems separate Right / Wrong from Reason / Luck — never collapsing them into a P&L-only verdict
Saved to outputs/ if the user confirms (the persistent thesis ledger should live in a versioned location the analyst can re-open across sessions)

Regulatory & Compliance Layer

SEC Marketing Rule 206(4)-1 — any content that could be presented externally (sales tape, marketing) must observe the Marketing Rule (no cherry-picked performance, time-period consistency, net-of-fees, hypothetical-performance disclosures); flag any ledger row that mixes external-marketing language with internal research
Advisers Act Rule 204-2 (Books & Records) — research files (analyst notes, source documents) retained five years from the year of last use, first two years readily accessible; ledger pointers reference the retention path
MNPI / Information-Barrier overlay — if the analyst's firm operates a wall (private side, M&A, restricted list), any evidence sourced from a wall-crossed call or restricted information is flagged and quarantined; thesis cannot be acted on while the overlay is in place
Reg FD adjacency — for sell-side analysts, the thesis ledger does not become a vehicle for selectively distributed material non-public information; any conviction change driven by a 1:1 management call is logged with the management-public-disclosure status
Selective-distribution / fair-disclosure policy — internal thesis cadence and external publication cadence are kept separately; ledger metadata flags whether content has been published externally and to which audiences
Best-execution / sales-trader handoff — sizing-action recommendations are not orders; they hand off through the order-management system with full pre-trade compliance per skills/operations/trade-lifecycle-tracker.md
Personal-trading policy — any analyst personal-trading conflict on a covered name is surfaced and quarantines the recommendation pending CCO clearance
GIPS / performance-presentation — postmortem statistics that may roll up into a track record are computed consistent with the firm's GIPS-compliant attribution methodology

Personalization Hooks

The following config.yml keys customize this skill:

research.coverage_strategy — single-analyst single-sector / generalist / pod / sector-team / multi-strategy
research.conviction_taxonomy — firm scale (1–5 / Low–High / Top-of-Book / Coverage-Watch)
research.sizing_matrix — conviction × liquidity × volatility → sizing tier mapping
research.sell_discipline — explicit sell triggers (PT hit, invalidation, time-stop, factor-exposure cap)
research.coverage_universe — ticker list with sector / sub-sector / sleeve assignments
research.ic_cadence — IC meeting frequency, voting / non-voting members, quorum, conviction-change-threshold for IC review
research.pt_methodology_default — default valuation method per asset class
compliance.mnpi_policy — wall-crossing, restricted-list, blackout convention
compliance.personal_trading — analyst personal-trading rules and pre-clearance authority
performance.attribution_standard — GIPS or firm convention; hit-rate / slugging-ratio definitions

Example Output

[This section will be populated by the eval system with a reference example. For now, run the skill with sample input to see output quality.]

Handoff Contracts

Inbound:

skills/operations/earnings-call-summarizer.md — single-call mandatory thesis-update block flows in as a Confirmed / Pending / Challenged / New classification
skills/operations/morning-notes-drafter.md — daily Confirmed / Pending / Challenged / New markers feed the chronological ledger
skills/operations/dcf-valuation-builder.md, comparable-company-analysis.md, lbo-model-builder.md, accretion-dilution-analyzer.md, three-statement-model-constructor.md — provide the price-target methodology and the bridge math
skills/operations/market-research-brief.md — sector / sub-sector context that influences the thesis sleeve
skills/operations/stress-test-scenario-modeler.md — invalidation-criteria stress-test inputs

Outbound:

skills/operations/investment-memo-drafter.md — top-of-book theses promote into IC-grade memos for sizing-up decisions
skills/operations/morning-notes-drafter.md — stale or at-risk theses surface in tomorrow's note thesis-update block
skills/operations/trade-lifecycle-tracker.md — sizing-action recommendations hand off as parent-order genesis
skills/_shared/meeting-summarizer.md — IC discussion of the thesis becomes a meeting record that loops back as a ledger reviewer entry
skills/admin/regulatory-filing-checker.md — large-position threshold (Form 13F, 13D/G) movements driven by sizing-action recommendations