Skip to content

NFI Backtest Engine v1.6.0

Choose a tag to compare

View all tags
@github-actions github-actions released this 13 Aug 21:41
v1.6.0
87bc580
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

NFI Backtest Engine v1.6.0

v1.6.0 is the Full Native Strategy release. The Native lane compiles supplied NFI
Python into typed programs, then executes indicators, Signal and Tag mutation,
stateful callbacks, orders, wallets, and result assembly in Rust. Strategy Python is
not imported or executed during the Native run.

Current X7 hotfix

The X7 Grind-5 compiler and runtime now represent source-defined
slice_profit_exit, Boolean protection columns, bare Derisk state, Futures-mode
guards, liquidation-price presence, and liquidation-distance comparisons. These are
generic expression operands and predicates; there is no v536/v537, Signal, pair,
timerange, SHA, or expected-result runtime branch.

The engine also pins the relevant Freqtrade candle order: adjustment evaluation and
its filled order happen before stop/exit evaluation. We did not reverse this order in
response to the external report because pinned Freqtrade 2026.5.1 uses the same
sequence.

The upstream v17.4.536 and v17.4.537 sources pass the supported Spot/Futures static
Native checks. The reporter's sealed 2022 Futures input is not present in this
repository, so this release does not claim its reported 180-trade result as an exact
certificate. Unknown or newly changed active behavior still fails closed.

The source compiler also normalizes the NaN-tolerant Chaikin helper used by the
captured release fixtures, including its explicit NumPy float64 zero buffer. The
same generic Rust kernel therefore covers both the current helper and older captured
X7 sources without a strategy-version branch.

Full Native runtime

  • Complete Indicator, Signal, Tag, callback, order, wallet, and state execution is
    source-compiled and runs without strategy Python at runtime.
  • Pair-local DAG work is parallel; shared wallet and order mutation stays in one
    chronological event loop.
  • Raw frames and programs are path- and SHA-bound before decode. Legacy sealed
    Feather replay remains available for evidence reproduction.
  • Temporary row spools are capacity-admitted before execution and use
    delete-on-close ownership. Completed and failed runs do not accumulate spool files.

Evidence boundary

The latest-X7 Spot/Futures qualification and captured regressions have zero-tolerance
trade-surface and full-state equality. A five-year Full Native Spot development
workload also produced three byte-identical 1,162-trade results in 63:17, 62:51, and
62:44, with the temporary spool remaining inside its 38.47 GB admitted bound.

That performance result is single-host evidence. v1.6.0 keeps
combined_full_x7_certified=false: it does not relabel the v1.0.0 Spot or v1.1.0
Futures certificates, claim new official-Freqtrade five-year parity, or claim
cross-platform performance.

Upgrade

Run the normal installer again after v1.6.0 is published. It selects the matching
wheel, verifies the release SHA-256 manifest, and replaces the isolated nfi-bte
installation.

Assets 14
Loading