HodlerKitBuilding in public

Explore · Reproduce

Same inputs. Fresh process. Same bytes.

The current evidence can be inspected and replayed, but it cannot be promoted by changing a flag. Canonical publication still requires the open quality, owner, legal and signing gates.

Pinned identity

The candidate you are reproducing

Config ID
trendfollower-2.0.0-eur-trailing-off-canonical-candidate-v1
Candidate decision SHA-256
d46a2694dedb0df8658acb91e070d6a1ab57148599328e52ed0d5f8a036a9799
Exit profile
trendfollower-2.0.0-sidecar-trailing-disabled-approved-v1
Exit profile SHA-256
e1fd0677c832ed043e16594a159e421640a35a435d4af3e5b327ae8dfc0d59b7
Dataset
binance-spot-eur-4h-20200103t080000-20260715-v2
Dataset manifest SHA-256
494b925f0f898cd8809ca43c47a4d806215b51dc7f827c36edc9800d05c1862f
Engine
Freqtrade 2025.11.2
OCI image
freqtradeorg/freqtrade@sha256:0dc30badce8c7523a187f333ba37cddec314f7ef1988e73710d730197c4bb91c
Platform
linux/amd64
Repeated receipt
canonical-replay-execution-67a5c73164e5a7b1ef94f6f22bf7f8c45c7b5819d2e11b50841bcda7ba2c9d81

Repository root

Two fixed launchers

Neither command accepts a replacement strategy, pair list, dataset, runtime image or friendlier gate selection.

Completed evidence

Repeat the proposed EUR candidate twice

uv run --project research --frozen --no-sync python -m research.runner.canonical_eur_replay

Requires the reviewed local EUR v2 bundle, exact locked wheel, cached OCI image and a local Unix Docker socket. The output remains ignored and non-publishable.

Open runtime gate

Run the fixed 6×2 quality matrix

uv run --project research --frozen --no-sync python -m research.runner.canonical_release_quality

Runs combined lookahead, per-pair recursion and three timezone cases, each twice in a fresh networkless container. A measured failure remains evidence; infrastructure uncertainty does not.

Acceptance checks

What a successful local replay proves

  1. Inputs are unchanged. Config, strategy, data, adapter, image and verifier identities match their code-owned registry.
  2. Containers are fresh and offline. Each attempt has a distinct owned container with no network and confirmed cleanup.
  3. Output is deterministic. Raw result, signal projection and normalized result match byte for byte across both attempts.
  4. Terminal positions stay open. End-of-period engine bookkeeping is not presented as a completed strategy exit.