Bot Tester · Item

LaT-PFN

LaT-PFN · Isolated zero-shot time-series forecasting candidate (pretrained PyTorch model)

full botdebuggingno reproduced evidence
Stage
debugging
Licence
review required
Blockers
1
Leakage audit
pending

Claims

What the upstream repository, author or a static read asserts — never evidence
unverified
Emits forecast, uncertainty, direction/confidence and regime from a pretrained LaT-PFN model.
claimed by upstream code reading (static audit)
unverified
The wrapper's lagged context mode corrects an older misaligned legacy mode.
claimed by upstream code/comments
Exact temporal alignment is a critical audit target.
unverified
Author live-record evidence.
claimed by upstream author
Whether tuning occurred after observing reported live outcomes is an open audit question.

Reproduced by TradeRep

Only findings with an openable evidence artifact count

Nothing reproduced yet, so no performance figure may be attributed to this item. Promotion requires independently reproduced evidence — a README claim is not a result.

Source & licence

Repository
meatham07-design/latpfn-trading
Upstream
fork of an external author's repository
Immutable SHA
Not pinned yet — reproduction should record one.
Licence status
review required — Model weights provenance is itself an audit target; license review precedes reuse.
Disposition
Reproduce and challenge as an isolated zero-shot forecasting candidate.
Notes
Test model inference independently from its execution stack (EXTERNAL_FUTURES_BOT_VALIDATION_PLAN.md).

Audit targets, data & blockers

Audit targets
  • · exact temporal alignment
  • · pretrained weights provenance
  • · model/input normalization
  • · yfinance/data quality
  • · forecast threshold calibration
  • · author live-record evidence
  • · post-hoc tuning after observing live outcomes
Test data
yfinance-sourced series in the upstream stack; data quality is an explicit audit target.
Blockers
  • latpfn-baseline-blocker: An open implementation blocker prevents a reproducible baseline run; owned by the external-bot lane.
    docs/status/TRADE_REP_MASTER_TRACKER.json

Statistics

Baseline · stress · OOS · funded simulation
baseline
No recorded statistics.
stress
No recorded statistics.
oos
No recorded statistics.
funded simulation
No recorded statistics.

Reuse decisions

What is worth keeping, and what was rejected
Reusable pieces
  • · model-inference-isolated-from-execution harness pattern
Rejected pieces
None recorded.
Gate outcomes
No scientific gate has produced an outcome for this item.

Read from research-output/registry/bot-testers.json · docs/status/TRADE_REP_MASTER_TRACKER.json