> ## Documentation Index
> Fetch the complete documentation index at: https://docs.shumi.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Evidence

> The published studies, what each one tested, and how to check the walk-forward record yourself.

Five studies are published at [shumi.ai/studies](https://shumi.ai/studies). The strategy studies list every outcome and the benchmark lists every score. The figures stay in the studies, next to the sample sizes that go with them.

## What each study tested

| Study | Published | What it tested |
| - | - | - |
| [How Shumi Tracked the Spring 2026 Rally](https://shumi.ai/studies/regime-engine-april-may-rally) | May 13, 2026 | One live window in spring 2026 on Hyperliquid prices, with every closed outcome listed |
| [400 Experiments. 3 Survived.](https://shumi.ai/studies/400-experiments) | Mar 25, 2026 | Hundreds of strategy variants, 2021 to 2025, and what real Binance funding costs did to a strong in-sample result |
| [Walk-Forward Performance, 2024-2025](https://shumi.ai/studies/walk-forward-performance-2024-2025) | Mar 10, 2026 | Trained on 2021 to 2023, tested on 2024 to 2025, Binance funding charged per entry, weak spots listed |
| [Crypto AI Benchmark v2](https://shumi.ai/studies/crypto-ai-benchmark-v2) | Mar 11, 2026 | Several AI tools given the same trader questions, scored on a weighted rubric. Shumi was left out of the scored field |
| [Context Engineering Under Recording](https://shumi.ai/studies/live-session-context-engineering) | Mar 14, 2026 | One unedited working session, mistakes and corrections left in |

Three of these studies carry a dated update from September 28, 2026. The engine behind the spring rally study did not hold up in later forward testing. A stricter retest found the engines behind the 400-experiments and walk-forward studies no better than chance, and they lost money on 2026 data. Each update sits at the top of its study; the original numbers stay below it as the record of that period.

Read a study's limits before its headline.

## Read the walk-forward record

`shumi walkforward outcomes` lists the tracked walk-forward outcomes: each entry, each exit and the reason it closed. On 2026-09-23 it held one closed outcome (LINK, long, closed 2026-05-28), so its `summary` is a sample of one. `shumi walkforward signals` lists the entry and exit signals themselves (50 per call).

<Tabs>
  <Tab title="Terminal">
    ```bash theme={"dark"}
    shumi walkforward outcomes
    ```
  </Tab>

  <Tab title="API">
    ```bash theme={"dark"}
    curl "https://api.shumi.ai/api/cli/walkforward?action=outcomes" \
      -H "Authorization: Bearer shumi_sk_your_key_here"
    ```

    `action` takes `signals`, `positions` or `outcomes`.
  </Tab>
</Tabs>

Each outcome has these fields:

| Field | Meaning |
| - | - |
| `symbol` | The coin |
| `direction` | `long` or `short` |
| `entryPrice`, `exitPrice` | Price at entry and at exit |
| `entryDate`, `exitDate` | When it opened and closed |
| `returnPct` | Return in percent |
| `holdDays` | Days held |
| `exitReason` | Why it closed |

A `summary` block counts the total, wins and losses.

## Ask whether a signal has a validated record

```bash theme={"dark"}
shumi signal-quality --asset BTC
```

Over the API, `GET /api/cli/signal-quality?asset=BTC`. From an AI agent, the `get_signal_quality` tool. `--signal-type` defaults to `mean_reversion`.

The answer says whether a validated record exists for that signal on that asset. On 2026-09-23 BTC came back with `reliability_tier` `INSUFFICIENT_DATA` and the message "No validated edge exists for mean\_reversion BTC. Raw signal only." That reading means the signal has no validated record behind it.

## Recheck the results yourself

Every claim on this page can be checked.

* **Outcomes.** Every closed outcome in a strategy study is listed with its dates and prices. Pull the same dates from any exchange's price history and recompute the return.
* **The walk-forward record.** `shumi walkforward outcomes` carries the same fields. Recompute any return from its entry and exit prices.
* **Benchmark answers.** The raw answers from every tool are published. Score them with your own rubric.
* **Limits.** The walk-forward study and the benchmark each list their own gaps.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.