# FinRunbook examples

Eight completed interactive financial reports generated with FinRunbook: **five English and three Chinese**, plus a financial-data preparation demonstration. Three primary use cases lead the gallery; the retrospective thesis checkpoint is an additional walkthrough. Research cutoffs are September 3–4, 2026. Reporting periods vary by example; these are saved snapshots with no live data or background monitoring.

## Start with a task

| Use case | Worked example | Reusable prompt and output |
| --- | --- | --- |
| [Review earnings—what changed?](index.html#earnings-update) | Lululemon: FY2021–FY2025 and Q1/Q2/H1 FY2026, cutoff September 4, 2026 | [Adapted earnings prompt](lululemon-earnings-update/earnings-prompt.txt), [report](lululemon-earnings-update/report/), [earnings table CSV](lululemon-earnings-update/earnings-update.csv) |
| [Prepare comparable financial data](index.html#data-cleaning) | Three worked cases using saved Cloud and Lululemon observations | [Demo](financial-data-cleaning/index.html), [prompt](financial-data-cleaning/prompt.txt), [inputs](financial-data-cleaning/inputs.csv), [cleaned data](financial-data-cleaning/cleaned-data.csv), [dictionary](financial-data-cleaning/metric-dictionary.csv), [exceptions](financial-data-cleaning/exceptions.csv) |
| [Compare companies before deeper research](index.html#company-comparison) | AWS, Azure and Google Cloud over 2021–2025, cutoff September 3, 2026 | [Full comparison prompt](cloud-computing/prompt.txt), [report](cloud-computing/report/), [report data JSON](cloud-computing/report/report-data.json) |
| [Check a thesis against new evidence](index.html#thesis-check) | Illustrative Lululemon checkpoint using the same September 4 evidence | [Adapted thesis prompt](lululemon-earnings-update/thesis-prompt.txt), [illustrative checkpoint](lululemon-earnings-update/thesis-check.html) |

The gallery starts with questions and visible results: a year-over-year quarterly
comparison, worked data transformations, and a peer comparison. Inspect a saved
result, download its data, or adapt a prompt for your own companies. Cutoffs and
material comparison limits stay alongside the results; detailed receipts are
available on demand. Viewing the static gallery does not execute research.
The adapted Lululemon prompts are not
the exact instruction that produced the historical report; its original remains
in `prompt.txt`.

The data-cleaning demonstration starts from existing extracted observations. It
does not claim a new PDF extraction or a generic cleaning engine. Exact input
values, fiscal calendars, source scope and threshold operators are preserved;
new arithmetic is reconciled against directly reported quarters. It is not a
ninth financial report. Its separate transformation review covers this example.

The thesis assumptions were written retrospectively from saved evidence. This
is an instructional example, not a real customer's thesis, a historical
prediction or a monitoring record. It reuses Lululemon's report and does not add
a ninth report. No time savings, customer adoption or investment performance are
established by these examples.

## View online

[Open the example gallery](https://ickma2311.github.io/FinRunbook/).

GitHub Pages publishes only this directory. Updates to `examples/` on `main`
run the packaging checks and deploy automatically through
`.github/workflows/pages.yml`. Working runs and raw source documents are not
included in the deployment.

## View locally

From the FinRunbook repository root:

```sh
python3 -m http.server 8771 --bind 127.0.0.1 --directory examples
```

Open <http://127.0.0.1:8771/>. Serve the directory over HTTP: the report renderer fetches its adjacent JSON file, so opening an HTML file directly with `file://` is not supported.

### Local-only previews

`examples/local/` holds previews whose source-data redistribution review is still
pending. It is ignored by Git and excluded from the public catalog. When that
directory exists locally, open `/local/` on the same preview server.

The earlier local S&P 500 / SPY preview remains separate from the curated public
package below. The public package contains analytical output and a ledger excerpt,
not the full daily-price ledger. It retains its original `PASS_WITH_WARNINGS`
status and all three warnings. Publication does not establish a data license.

The packaging test rejects tracked files under `examples/local/`, including files
added with `git add -f`. To publish a local preview, first resolve its data-use
boundary, then curate it as a regular package and add it to the public catalog.

| Example | Report | Main analytical feature |
| --- | --- | --- |
| Lululemon | [Open report](lululemon-earnings-update/report/) | Earnings deterioration, regional comparisons, financial bridges, peer disclosures and illustrative recovery scenarios |
| Apple | [Open report](apple-business-quality/report/) | Product and geography mix, cash allocation and an EPS bridge separating income and share-count effects |
| Cloud computing | [Open report](cloud-computing/report/) | AWS and Google Cloud comparison, separate Azure disclosures and explicitly scoped parent capex |
| Pandemic growth retention — English | [Open report](pandemic-growth-retention/report/) | Five businesses across FY2019–FY2025; indexed revenue, profitability/cash matrix, margin bridges and acquisition limits |
| AI infrastructure — English | [Open report](ai-infrastructure-capex-supply-chain/report/) | Four hyperscalers, eight suppliers, capital-spending filters and named relationship evidence |
| 六家科技公司的资本配置 — 中文 | [打开报告](big-tech-investment-allocation/report/) | 五年趋势、实际财年日期、支出口径和现金桥接 |
| AI 产业链 — 中文 | [打开报告](ai-industry-value-chain/report/) | 模型、硬件、云和应用；财务筛选、增长桥接、推理成本情景 |
| 美股大盘长期收益与风险 — 中文 | [打开报告](sp500-20year-return-risk/report/) | 复合收益与购买力、回撤与回本、滚动持有窗口及提款顺序；保留 3 项验证警告 |

## Why these examples?

The examples cover company quality, capital allocation, industry value chains,
buyer–supplier relationships and long-term market risk. Selection considered existing validation and
editorial records, first-party sources, interactive controls, comparison limits and
suitability for a static public package. Selection is not a new independent audit or
a guarantee of financial accuracy.

The three Chinese reports retain their original Chinese text and controls. The English
supply-chain example now opens actual evidence and calculation details from its
existing ledger instead of displaying IDs alone; its financial data and conclusions
are unchanged.

## What is in each package?

- `prompt.txt`: the original instruction, in its original language.
- `report/`: portable HTML, CSS, JavaScript and presentation JSON. No third-party browser library or financial API call is needed to view it.
- `report/report-data.json`: downloadable presentation data, linked from every gallery report. Its structure varies by report; it is not a standardized spreadsheet model.
- `research-record.json`: structured facts, precise evidence locators, source URLs and hashes, calculation inputs, and pinned skill revisions. The market-history package contains an explicitly labeled report-evidence excerpt, not the complete original run ledger.
- `validation.json`: FinRunbook schema, reference-integrity and provenance checks.
- `model-audit.json` or `model-tests.json`, where available: original numeric-recomputation receipts. The English supply-chain report has no separate model-audit receipt; none is implied.
- `semantic-validation.json`, where available: original same-agent review and disclosed limitations.
- `editorial-review.json`: language-specific tone review and recorded changes.
- `browser-tests.json`: browser interaction, chart geometry, responsive-layout and local-resource checks.

The same author performed the analysis and editorial/semantic review. Automated validation is not an independent audit and does not guarantee that issuer disclosures are correct.

## Re-run the prompts

Ask the FinRunbook router to run the contents of a package's `prompt.txt`. Normal working data belongs in ignored `runs/<run-id>/` directories. These curated example packages are deliberately outside `runs/` so they can be version-controlled and published. Review new packages for data rights and sensitive information before adding them here.

The original validation and model receipts retain their original scope and timestamps.
New `publication-review.json` receipts describe packaging and browser checks only.
Original raw-file names and hashes may remain in evidence locators for traceability;
those source archives are not included. Use the publisher URLs to consult documents.

## Sources and limitations

Lululemon uses thirteen SEC, issuer and investor sources. It covers FY2021–FY2025
and matching Q1/Q2/H1 periods for FY2026 and the prior year. Peer disclosures keep
their own reporting periods. Actuals, guidance and recovery assumptions remain
separate; the refund exclusion is a narrow analyst adjustment, not a fully
normalized earnings series. The CSV exports recorded facts and calculation
results with evidence references; it does not include executable spreadsheet
formulas. The accompanying thesis checkpoint is a retrospective illustration.
See its [package notes](lululemon-earnings-update/README.md).

The S&P 500 / SPY example covers 2006–2025, with a separate 1993–2026 rolling-window
sample and a September 4, 2026 research cutoff. It uses Yahoo adjusted prices as a
return proxy, not an independently reconstructed total-return index. Its public
ledger has 20 facts, 16 evidence entries, 9 sources and 8 calculations. Original
receipts remain scoped to the full local run; a separate publication review checks
the packaged report. Raw snapshots, caches and the bulk daily-price ledger are
excluded. Source-use limitations remain visible. See its
[package notes](sp500-20year-return-risk/README.md).

The pandemic-growth report uses 14 SEC annual filings and Shopify's FY2022 IR
release, whose annual financial tables are unaudited. It covers annual fiscal
periods ending in 2019–2025, without calendarization. Etsy FCF subtracts separately
reported capitalized software; acquisitions, impairments and sparse workforce
checkpoints remain visible. See its [package notes](pandemic-growth-retention/README.md).

Apple uses its FY2023 and FY2025 Forms 10-K, whose comparative financial statements together cover all five fiscal years. Cloud uses Amazon and Alphabet's 2023 and 2025 Forms 10-K plus Microsoft's five annual reports for FY2021–FY2025. Overlapping financial years are reconciled.

Apple's FY2023 has 53 weeks. AWS and Google Cloud use calendar years; Microsoft's fiscal years end in June. Google Cloud includes Workspace, while Azure is not the Intelligent Cloud segment. Missing Azure revenue and operating margins are not estimated. Microsoft's FY2025 Azure revenue disclosure is a threshold, not an exact value. Parent-company cash capex is not cloud-only or AI-only investment.

The Chinese capital-allocation report uses issuer-specific five-year windows (ending
in FY2026 for NVIDIA and Microsoft and FY2025 for the other companies). The Chinese
AI report separates annual and quarterly observations, and its cost simulator is
illustrative rather than a company forecast. The English supply-chain report spans
2023 through the research cutoff; supplier revenue is not additive across layers.

Packages contain attributed factual extracts and analysis, not copies of complete issuer filings. Source documents and third-party materials remain subject to their respective terms. No issuer endorsement, blanket commercial-data license, or independent audit assurance is claimed.