# AI use cases for Finathink (ceyyy427/finathink)

Repo: [ceyyy427/finathink](https://github.com/ceyyy427/finathink) | Use cases: 5 | Page: https://dedupedai.com/use-cases/ceyyy427-finathink/

Finathink is a local-first Python workbench that turns user data, evidence, factors, backtests, risk checks and paper portfolios into a traceable research record with offline HTML reports.

This file holds every use case of this repo. Each one starts with a "## " heading and ends with its URL.

- Run the offline research sample and open its workbench: https://dedupedai.com/use-cases/ceyyy427-finathink/run-offline-research-sample-and-open-report/
- Normalize your own data API into point-in-time batches: https://dedupedai.com/use-cases/ceyyy427-finathink/connect-own-data-api-with-point-in-time-checks/
- Trace why a paper decision card cannot be executed: https://dedupedai.com/use-cases/ceyyy427-finathink/audit-why-paper-decision-card-is-blocked/
- Backtest a factor idea with a hidden out-of-sample test: https://dedupedai.com/use-cases/ceyyy427-finathink/backtest-a-factor-with-hidden-oos/
- Run five analyst agents and read the manager's summary: https://dedupedai.com/use-cases/ceyyy427-finathink/run-multi-agent-analyst-research/

## How to run the offline research sample and open its workbench with Finathink

Repo: [ceyyy427/finathink](https://github.com/ceyyy427/finathink) | Difficulty: Intermediate | Time: 30 minutes | Basis: From the README

Start the local app on bundled fixtures and inspect the staged research workbench without any external services.

**Who it is for:** A new user or developer who wants to see how a full research run is recorded before using their own data.

**The problem:** Research tools often hide how a result was produced, so it is hard to trust or reproduce a backtest or a decision.

**The outcome:** A running local app serving the /workbench page, showing staged analyst, research, quant, risk and paper decision records from the sample fixtures.

### Steps

1. Run `python3 -m venv .venv`
2. Run `source .venv/bin/activate`
3. Run `python3 -m pip install -e '.[dev]'`
4. Run `PYTHONPATH=src python3 -m pytest -q` to confirm the test suite passes
5. Run `python3 scripts/run_local_app.py --sample --port 8765`
6. Open the /workbench page on port 8765 and review the stages, checkpoints and manifest fingerprints

### Prompt to paste into your agent

```text
In the finathink repo, set up the environment as described in its Quick Start, then run the local app with --sample on port 8765. Report the /workbench stage list, each stage status, and the manifest fingerprint shown. Verify that the sample run reports the stages from the research flow without requiring any external credentials.
```

### Caveat

The README marks the offline modules as OFFLINE_PASS, but real vendors, models, browsers and production deployment are EXTERNAL_UNVERIFIED and not covered by this sample.

Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/

URL: https://dedupedai.com/use-cases/ceyyy427-finathink/run-offline-research-sample-and-open-report/

## How to normalize your own data API into point-in-time batches with Finathink

Repo: [ceyyy427/finathink](https://github.com/ceyyy427/finathink) | Difficulty: Advanced | Time: 60 minutes | Basis: From the README

Save a data connection, map its fields, run a trial pull and check PIT status and fingerprints before research uses it.

**Who it is for:** A researcher with an existing market data API who wants its data checked for point-in-time correctness before running any analysis.

**The problem:** Data pulled from APIs can leak future information or silently change fields, which invalidates backtests.

**The outcome:** A normalized DataBatch with recorded data fingerprint, field quality, source notes and point-in-time status, with credential values kept out of research records.

### Steps

1. Read docs/USER_DATA_API_GUIDE.md to see the connection contract and field mapping format
2. Save your data API connection config and field mapping in the app's provider settings
3. Store the credential as an environment variable or system keychain reference, not as a value in the config
4. Run an explicit trial connection to pull JSON data and standardize it into a DataBatch
5. Review the recorded fingerprint, field quality and PIT status before using the batch in any research run

### Prompt to paste into your agent

```text
In the finathink repo, follow docs/USER_DATA_API_GUIDE.md to map the fields of <data API name> and run one explicit trial connection. Report the DataBatch fingerprint, field quality summary and PIT status. Verify that no credential value appears in any saved research record or log.
```

### Caveat

The README states that real vendors are EXTERNAL_UNVERIFIED and the repo has no real credentials, so the connection step only works with an API you supply and configure yourself.

Risk, credentials: Your data vendor key must stay in an environment variable or system keychain reference; never paste it into a config file, prompt or research record.

Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/

URL: https://dedupedai.com/use-cases/ceyyy427-finathink/connect-own-data-api-with-point-in-time-checks/

## How to trace why a paper decision card cannot be executed with Finathink

Repo: [ceyyy427/finathink](https://github.com/ceyyy427/finathink) | Difficulty: Intermediate | Time: 30 minutes | Basis: From the README

Follow the risk, portfolio and quant gates to see which stage kept a paper decision card in a non-executable state.

**Who it is for:** A strategy reviewer or AI agent that has received a paper decision card and must explain whether it is actionable.

**The problem:** A paper decision can look ready even when an upstream risk or quant stage did not pass, and the reason is hard to find.

**The outcome:** A written audit that names each failed or passed stage (PIT, liquidity, concentration, drawdown, stress, quant) and states the decision card status.

### Steps

1. Run the offline sample steps from the Quick Start to produce a complete run
2. Open the HTML report for the run and locate the risk, paper decision and quant sections
3. For each risk check (PIT, liquidity, concentration, drawdown, stress scenarios), record pass or fail
4. Confirm the paper decision card is marked non-executable whenever risk, portfolio or quant stages did not pass
5. Write the audit with stage names, status and evidence references

### Prompt to paste into your agent

```text
In the finathink repo, open the HTML report from a local sample run and list each risk, portfolio and quant stage with its status and evidence reference. Report whether the paper decision card is executable. Verify that the card is non-executable if any upstream stage failed, and state that the repo does not place broker orders.
```

### Caveat

Broker orders, live trading and arbitrary execution are NOT_IN_SCOPE in this repo, so the paper decision card is a research output only.

Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/

URL: https://dedupedai.com/use-cases/ceyyy427-finathink/audit-why-paper-decision-card-is-blocked/

## How to backtest a factor idea with a hidden out-of-sample test with Finathink

Repo: [ceyyy427/finathink](https://github.com/ceyyy427/finathink) | Difficulty: Advanced | Time: 1 hour | Basis: From the README

Turn a strategy hypothesis into a controlled factor proposal and test it on training, validation and hidden OOS data.

**Who it is for:** A quant learner who wants to test a factor idea without fooling themselves with overfitting.

**The problem:** Backtests that reuse the same data for tuning and testing look great and fail live.

**The outcome:** IC, ICIR, turnover and decay metrics with a one-time hidden OOS test, logged with fingerprints in an offline HTML report.

### Steps

1. Create a venv and run `python3 -m pip install -e '.[dev]'`
2. Start the local app with `python3 scripts/run_local_app.py --sample --port 8765`
3. Describe your hypothesis in the factor workbench so it maps to a fixed DSL template
4. Run training, validation and the one-time hidden OOS test
5. Read IC, ICIR, turnover and decay, and keep the proposal only if OOS holds up

### Prompt to paste into your agent

```text
Using ceyyy427/finathink with the sample data, turn the hypothesis <idea> into a factor proposal, run the backtest with the hidden OOS test, and summarize IC, ICIR, turnover and decay. Paper trade only.
```

### Caveat

Not financial advice: paper trade first, and remember you can lose money. The README is in Chinese.

Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/

URL: https://dedupedai.com/use-cases/ceyyy427-finathink/backtest-a-factor-with-hidden-oos/

## How to run five analyst agents and read the manager's summary with Finathink

Repo: [ceyyy427/finathink](https://github.com/ceyyy427/finathink) | Difficulty: Intermediate | Time: 45 minutes | Basis: From the README

Let fundamental, technical, sentiment, news and learning analysts work in parallel and a research manager merge their evidence.

**Who it is for:** An individual investor or student who wants structured, evidence-linked research rather than one model's opinion.

**The problem:** A single model answer mixes opinion and evidence and hides disagreements.

**The outcome:** Observations from five analyst roles merged by ResearchManager into a plan with evidence references and open hypotheses.

### Steps

1. Install and start the local app as in the quick start
2. Save a low-frequency OHLCV snapshot or pick a saved one in the research task page
3. Choose your own compatible model endpoint and authorize sending role materials, or use the offline driver
4. Run the five analysts in parallel and let ResearchManager summarize
5. Open the HTML report and compare runs

### Prompt to paste into your agent

```text
Run ceyyy427/finathink's multi-agent research on <ticker or dataset> with the offline driver first, then summarize where the five analysts disagree and which hypotheses ResearchManager marks as still to verify. Paper trade only.
```

### Caveat

Not financial advice: paper trade first, and remember you can lose money. The README is in Chinese.

Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/

URL: https://dedupedai.com/use-cases/ceyyy427-finathink/run-multi-agent-analyst-research/