# JoinQuant TargetPackage execution evidence — 2026-07-26

This is a conservative record of one real JoinQuant hosted execution smoke.
It proves that the frozen TargetPackage consumer reached the real hosted
account/price/order/fill boundary. It does not prove that JoinQuant recomputed
the upstream Universe, Alpha, Portfolio, or Risk layers, and it is not an
investment-performance claim.

## Frozen identity

- Run:
  [JoinQuant backtest detail](https://www.joinquant.com/algorithm/backtest/detail?backtestId=2457a7c39a276e09e0fabf99e28978e1)
- Run ID: `2457a7c39a276e09e0fabf99e28978e1`
- Observed at: `2026-07-26T08:50:15Z`
- Bundle SHA-256:
  `0586930ddc483f4489f09b562f4d4232b458b703e6a1194af809b707211be8c4`
- Bundle manifest SHA-256:
  `91f4b4dc0bdd0d9f99b3466c830600e2678be78dd991d4499a4154bffec8597a`
- Portable core SHA-256:
  `f9de9f18e4008bb0fc80edb0b21cae6ea06a61828d89e9b7c11057f98225fc19`
- Wrapper SHA-256:
  `0c672b64665b7c1d4409ba288b00d1608e976800af3a4f9a6467767ce4def592`
- TargetPackage ID: `TP-20240311-795dcfba`
- Package SHA-256:
  `baa15c8156aa6cb54d3775d0cd562613a9d0a4f15b0c39f2a60e0f8b8c6f1f3a`
- Tape SHA-256:
  `5383c376951637b8d4b1ca42fc6db80347abda74c0181c0f70d6c411fb97ea64`
- Evidence class: `synthetic-research`

The uploaded browser snapshot and the frozen bundle had the same SHA-256.
The machine record distinguishes
`platform_evidence_class=real_platform_runtime` from
`input_evidence_class=synthetic-research` and fixes
`performance_claim=false`.
The tracked, redacted machine record is
[`run.json`](../evidence/joinquant/TP-20240311-795dcfba/2457a7c39a276e09e0fabf99e28978e1/run.json).
It contains no account identifier, cookie, token, credentials, screenshots, or
personal information.

## Runtime and observed sessions

- Period: 2024-03-11 through 2024-03-13
- Initial capital: CNY 1,000,000
- Frequency/runtime: daily, Python 3
- Benchmark: CSI 500 (`000905.XSHG`)
- Decision mode: `target_package`

| Session | Hosted observation | Orders | Fills |
| --- | --- | ---: | ---: |
| 2024-03-11 | Exact-clock miss; `TARGET_PACKAGE_NOOP`; no fallback | 0 | 0 |
| 2024-03-12 | Package hit; prices bound from `CURRENT_DATA_DAY_OPEN` | 3 | 3 |
| 2024-03-13 | Exact-clock miss; `TARGET_PACKAGE_NOOP`; no fallback | 0 | 0 |

The hosted plan preserved the package weights exactly: 10% each for
`600519.XSHG`, `601012.XSHG`, `601318.XSHG`, and `688981.XSHG`; all other
weights were zero.

The resulting targets and fills were:

| Symbol | Weight | Target quantity | Filled quantity | Fill price |
| --- | ---: | ---: | ---: | ---: |
| `600519.XSHG` | 0.10 | 0 | 0 | — |
| `601012.XSHG` | 0.10 | 4,500 | 4,500 | 22.22 |
| `601318.XSHG` | 0.10 | 2,300 | 2,300 | 43.18 |
| `688981.XSHG` | 0.10 | 2,000 | 2,000 | 47.87 |

`600519.XSHG` opened at CNY 1,693.94. One 100-share board lot therefore
required CNY 169,394, above the CNY 100,000 target notional. The deterministic
lot rule correctly produced a zero-share target; this was not a rejected or
skipped order. The STAR Market request for `688981.XSHG` used the configured
hosted limit-price protection.

## What this evidence proves

For the `execution` layer only, this run observes:

- exact bundle/package identity verification;
- exact signal-close/next-session clock matching with no momentum fallback;
- real hosted equity and `day_open` price binding;
- deterministic target-quantity and order construction;
- three accepted orders and three real hosted fills;
- clean no-op behavior on the surrounding sessions.

The release reachability manifest may therefore mark only
`execution.real_platform_observed=true`, backed by the tracked machine record.
Universe, Alpha, Portfolio, and Risk remain
`real_platform_observed=false`: those layers were computed locally and carried
to JoinQuant as a frozen TargetPackage rather than recomputed inside JoinQuant.

## What this evidence does not prove

The package still has evidence class `synthetic-research`. This three-session,
single-package run provides no:

- authorized real-data long-sample or out-of-sample validation;
- real-data calibration of the upstream Alpha, Portfolio, or Risk layers;
- multi-period same-input layer-by-layer local/JoinQuant parity;
- real QMT peer run or 20-trading-day broker shadow;
- G9 cross-engine acceptance, `BASELINE_60`, release verification, live
  readiness, or investment-value evidence.

The hosted performance fields are retained in the machine record solely to
identify the completed smoke. They are economically meaningless over this
three-session window and must not be presented as strategy performance.