# AI Infrastructure Research Kit

StocksWatch · Version 1.0 · September 24, 2026

A reusable process for organizing company disclosures, sources and assumptions.
Research and education only. Not financial advice, a recommendation or a prediction.

## Start here

Save a copy for each company and reporting period. Open this Markdown (.md) file
in any text editor. Replace bracketed prompts with your notes. Leave unknowns
explicitly marked "Not verified"; an empty cell is not evidence of zero.

Company: [name]
Research question: [one question you want the evidence to answer]
Period under review: [start and end dates]
Prepared by: [your name or initials]
Last reviewed: [date and time zone]
Next review trigger: [new filing, company update or chosen date]

Use current primary sources. A provider snapshot is a starting point and may be
delayed, restated or wrong. A retrieval timestamp and a reporting period describe
different things; record both. Do not present private or licensed data as yours to
redistribute.

## 1. Find and record the evidence

- [ ] Start with the company's investor-relations page and applicable regulator's filings.
- [ ] Record the exact document, publication date, reporting period and source URL.
- [ ] Check whether a later filing, amendment or correction supersedes the source.
- [ ] Distinguish reported results from guidance, estimates and your assumptions.
- [ ] Preserve currency, unit and metric definition (for example dollars versus millions).
- [ ] Record missing disclosures as "Not verified" rather than inferring a number.

| Claim or metric | Value and unit | Evidence type | Document / section | Source URL | Reporting period | Retrieved at | Verification status |
| --- | --- | --- | --- | --- | --- | --- | --- |
| [Claim] | [Value, unit, currency] | [Reported / guidance / provider estimate / assumption] | [Title, page or section] | [URL] | [Period] | [Date + time zone] | [Verified / conflicting / not verified] |

## 2. Map the business before interpreting a metric

- [ ] What does the customer buy, and how is revenue recognized?
- [ ] Which activities generate revenue now, and which are still planned?
- [ ] Separate announced, contracted, installed and revenue-generating capacity.
- [ ] Record whether utilization is actually disclosed; do not invent it.
- [ ] Check disclosed customer concentration and contractual limitations.
- [ ] List power, construction, hardware, networking and financing dependencies.
- [ ] Write down one material question the available documents cannot answer.

Business in three sentences:

1. [Customer and service]
2. [Revenue model and relevant operating measure]
3. [Principal execution dependency]

Open question: [question and source needed]

## 3. Keep financial comparisons consistent

- [ ] Compare the same reporting period, currency and unit.
- [ ] Separate reported metrics from adjusted or non-GAAP metrics.
- [ ] Find the company's reconciliation before using an adjusted measure.
- [ ] Check operating cash flow and capital expenditure definitions before comparing free cash flow.
- [ ] Record cash, debt and relevant financing commitments with their dates.
- [ ] Check share-count definitions and changes; do not mix basic and diluted counts.
- [ ] Distinguish a company target from an achieved result.

| Metric | Current period | Comparison period | Unit / definition | Source | Limitation |
| --- | --- | --- | --- | --- | --- |
| [Metric] | [Value or not verified] | [Value or not verified] | [Exact definition] | [URL / section] | [Known caveat] |

## 4. Maintain an assumption register

An assumption is not a fact because it appears in a spreadsheet. Record what could
disprove it and how you would revise it. Scenarios are conditional illustrations,
not forecasts or price targets.

| Assumption | Why it is needed | Supporting evidence | Contradicting evidence | What would change it | Review trigger |
| --- | --- | --- | --- | --- | --- |
| [Assumption] | [Question or calculation] | [Source or none] | [Source or not checked] | [Observable condition] | [Date or event] |

## 5. Record changes without rewriting history

| Review date | Previous observation | New evidence | What changed | Still unknown | Source |
| --- | --- | --- | --- | --- | --- |
| [Date] | [Prior note] | [Reported update] | [Interpretation, labeled as such] | [Open question] | [URL] |

Before completing a review:

- [ ] Every material number has a source, period and unit.
- [ ] Facts, estimates and assumptions are distinguishable.
- [ ] Conflicting evidence is preserved, not silently removed.
- [ ] Unanswered questions and next review triggers are explicit.
- [ ] The notes do not imply certainty, guaranteed outcomes or a trading instruction.

## Optional AI-assisted summary check

Use this instruction with text you are permitted to share:

> Summarize only the supplied source text. For each material claim, identify the
> supporting passage or section. Separate reported results, management guidance
> and interpretation. Preserve dates, units and caveats. Mark anything the source
> does not establish as "Not established in this source." Do not produce a trading
> recommendation or invent missing figures.

Verify every material claim against the original. AI-generated citations may be
incorrect; a confident answer is not evidence. Avoid uploading confidential or
restricted documents to a tool without permission.

## Further reading

- StocksWatch research desk: https://stockswatch.cc/
- Reading guide: https://stockswatch.cc/guides/nbis-research-guide.html
- SEC filings guide: https://stockswatch.cc/guides/nbis-sec-filings.html

## Use and limitations

You may copy and adapt this template for your own personal or internal business
research. Do not resell it or represent it as your original product. Rights in
third-party sources remain with their owners. No warranty of completeness or
accuracy is provided. Verify material information independently.

This file contains a blank research framework, not current market data. The kit
does not subscribe you to email, create an account or include a paid service.