skills/thesis-tracker/references/quality-guardrails.md

# Quality Guardrails

## PM-Grade Checklist

Before delivery, answer:

- What did we originally believe, what did the market believe, and what happened?
- What is the company-thesis status, and separately is the security thesis ready for an investment decision?
- Is aggregate company-thesis status reconciled to core-pillar statuses, with any override rationale explicitly supported by evidence?
- Which pillar is affected, and is the signal confirmatory, disconfirmatory, mixed, irrelevant, or untested?
- Is the change temporary, cyclical, structural, timing-related, or already priced in?
- Does the model, valuation, downside, conviction, sizing, hedge, or decision timing change?
- What action follows, what would make us wrong, and what is the next proof point?

## Market Data And Readiness

- For a material security-thesis update, attempt to obtain current public price and basic market context when accessible; capture source and as-of timestamp.
- A missing current price, valuation model, consensus frame, position size, benchmark/active weight, or risk budget must appear as an explicit decision gap when it changes the action.
- Do not turn a weakened company thesis into a valuation-led `trim`, `exit`, or `add` conclusion when the security-thesis inputs required for that call remain missing. Use `Re-underwrite`, `Conditional`, or `Not decision-grade` as appropriate.

## Falsifiability And Drift

Every tracker needs thesis, variant perception, market-implied expectation, prioritized pillars, KPIs, confirm/warning/break thresholds, catalysts, kill criteria, action thresholds, and open diligence. If missing, propose and label them. Retain numeric weights or scores only when inherited from an existing tracking method or explicitly requested; do not manufacture an aggregate score or scored chart for presentation.

Every new action threshold must be labeled `Draft threshold for PM confirmation` until approved. Keep `Inherited threshold` and `Approved monitoring rule` separately labeled with source and status.

An inherited core pillar marked `Impaired` should prevent aggregate `Watch` unless an explicit evidence-supported override rationale is stated. Multiple core pillars marked `Impaired` should normally result in aggregate `Impaired`.

Flag thesis drift when the reason to own/short changes, failed catalysts are replaced by vague qualitative rationale, valuation support replaces the original growth/margin thesis, repeated disconfirming evidence is ignored as "long term," or downside protection becomes the new rationale for an upside position.

Labels: thesis intact, thesis evolved, thesis changed, thesis broken, thesis retired.

## Evidence Quality

- Headline beat/miss is insufficient: parse revenue, mix, price, cost cuts, tax, buyback, FX, one-time items, KPI quality, guidance, FY2/long-term revisions, cash conversion, management credibility, and buyside expectations.
- Management commentary earns weight when quantified, consistent, source-supported, specific, and candid. Discount vague optimism, KPI definition changes, selective disclosure, and unexplained macro blame.
- Price action must be decomposed into fundamentals, estimates, multiple/spread/rates, sector/factor beta, positioning/crowding/short squeeze, liquidity/flows, options/vol, and macro/policy.

## Red Flags

KPI definition/disclosure changes, rising non-GAAP adjustments, revenue recognition changes, receivables/inventory build, weak cash conversion, capitalized cost spike, segment changes, auditor/CFO/controller departure, related-party transactions, aggressive add-backs, repeated one-time charges.

## Compliance And Sensitivity

Flag possible MNPI, wall-crossed or restricted-list material, expert-network sensitivity, broker redistribution limits, rumors, and uncited claims. Do not use sensitive information for public-equity trading recommendations; suggest compliance review or an external-clean version.

## Adjacent Handoffs

Route when primary need is pre/post-earnings analysis, detailed model update, new pitch, sizing/hedging, event scenario tree, Credit Markets handoff work, macro impact, committee memo, or deck/report QC.

## Final QA

Verify original thesis is preserved, evidence is appended, sources/as-of dates are visible, both sides are considered, KPIs are compared to expectations/thresholds where available, company-thesis status and security-thesis readiness are separately stated, proposed thresholds are not presented as approved rules, recommendation has timing/threshold/action, and red-team is credible.

## XLSX Presentation QA

For workbook deliverables:

- Keep the `Cover` tab decision-facing and readable without horizontal scrolling.
- Prefer a compact pillar sheet and move extended thresholds, citations, and model impacts to dedicated tabs.
- Keep full action-rule and diligence/gap registers on their dedicated tabs; `Cover` should show only summarized action posture, top blockers, and next gate.
- Freeze key identity columns and use bounded column widths; render every workbook tab and inspect it at ordinary zoom before delivery.
- Repair clipped headers, overly wide monitoring tables, unreadable source paths, or decision-critical text that requires excessive horizontal navigation.
- Do not include a weighted score chart unless its calculation methodology is inherited from the existing tracker or explicitly requested.

## PM Monitoring Guardrails

Every update should separate company thesis, stock thesis, portfolio role, and market setup. Required decision actions are `add`, `press`, `hold`, `trim`, `exit`, `hedge`, `wait for proof`, or `re-underwrite`. Include benchmark/ETF implications, liquidity, crowding, and what evidence would force a different decision when relevant.