Both DQN and PPO eval paths used old 3-action index matching (0=Buy,
1=Sell, 2=Hold). Now uses FactoredAction.target_exposure() for
exposure-weighted returns and order-type-specific transaction costs.
PPO path had .to_int() which doesn't exist on FactoredAction.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
compute_reward_pnl() took raw action_idx (0=Buy,1=Sell,2=Hold) — wrong
with 45-action FactoredAction encoding. Now takes &FactoredAction and
uses order-type-specific transaction costs. Position tracking uses
action.exposure instead of old 3-way index match.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Hyperopt adapter and PPO benchmark still had num_actions: 3, which would
produce misconfigured PPO models when used with the 45-action FactoredAction
sampling path. Found by spec compliance review.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Use a purely linear network (no ReLU) for the completeness axiom test.
IG on linear functions is mathematically exact, so the test is
deterministic regardless of random weight initialization. Tighten
tolerance from 20% to 1% (f32 rounding only). Keep ReLU network in
the basic test for non-linear verification.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- trading_service: PPO predict() used TradingAction match but act() now
returns FactoredAction. Use target_exposure() mapped to 0-1 range.
- IG completeness axiom test: relax tolerance from 5% to 20% (random
weights with ReLU non-linearity and trapezoidal rule can exceed 5%).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Was using direction=idx/15 (3 groups of 15: Buy/Sell/Hold) — incompatible
with DQN's FactoredAction encoding. Now uses exposure_idx=idx/9 (5 groups
of 9: Short100/Short50/Flat/Long50/Long100) matching FactoredAction.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
sample_action(), act(), act_with_log_prob(), greedy_action() now return
FactoredAction instead of TradingAction. Fixes the architectural disconnect
where num_actions=45 output neurons were sampled through a 3-action bottleneck.
TrajectoryStep.action and TrajectoryBatch.actions now use FactoredAction.
Added FactoredAction::from_legacy() for backward compatibility in tests.
Updated all PPO consumers: trainers/ppo.rs, hyperopt/adapters/ppo.rs,
validation/ppo_adapter.rs, benchmark/ppo_benchmark.rs.
2487 tests pass, 0 clippy warnings.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Consolidates ExposureLevel, Urgency, FactoredAction from dqn/action_space.rs
and ppo/factored_action.rs into common/action.rs. Both dqn:: and ppo::
re-export for backward compatibility. Deletes ppo/factored_action.rs.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Changed from Status::unavailable to Status::unimplemented with path
forward (ml_training_service forwarding). Fixed supports_feature_importance
from true to false.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
calculate_feature_importance() returned hardcoded fabricated scores with
wrong feature names on every inference. Replaced with honest empty map.
Real importance is computed on-demand via integrated gradients.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Implements IntegratedGradients using Candle autograd (Var::from_tensor +
backward). Computes attributions by integrating input gradients along
interpolation path from baseline to input.
Verified via completeness axiom test: sum(attributions) ≈ F(x) - F(baseline).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The real_ prefix was misleading — there is no fake data loader.
Mechanical rename across 18 source files, no logic changes.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- DQN PER: defer td_errors to_vec1() after loss.to_scalar() — piggyback on
existing pipeline flush instead of forcing premature GPU→CPU stall
- PPO trajectories: capacity-hint Vec allocations, extend_flat_states methods,
states_flat field on TrajectoryBatch for zero-copy GPU upload
- TGGN validate(): batch N per-sample losses on GPU → single to_scalar() sync
(was N GPU→CPU syncs)
- Liquid backward(): batch grad-norm per-param sqr().sum_all() on GPU → single
to_scalar() sync (was N GPU→CPU syncs per optimizer step)
- Liquid validate(): same N→1 GPU sync reduction as TGGN
- DQN trainer: restore EpochPrefetcher/DoubleBufferedLoader API (wrongly deleted)
- train_baseline_rl: wire DoubleBuffer GPU pre-upload — after CPU prefetch
completes, immediately upload next fold to GPU via DqnGpuData::upload() so
next fold starts with data already resident on GPU
2478 tests pass, 0 clippy warnings, 0 compile errors.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The remoteWrite block had placeholder values (COCKPIT_METRICS_PUSH_URL,
COCKPIT_PUSH_TOKEN) that were never replaced, causing Prometheus to
spam "Failed to send batch" warnings every minute. Cockpit was replaced
by self-hosted Grafana+Prometheus+Loki+Tempo.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- ibapi handshake: retry once after 1s on first failure (IB Gateway
needs time to release stale client_id slots after disconnection)
- test_load_nonexistent_config → test_load_config_succeeds: remove
default-value assertions that fail when ~/.foxhunt/config.toml exists
(defaults already tested by test_serde_defaults and test_default_config)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Match the actual K8s broker-gateway service port discovered during
live validation of fxt broker check.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The IB Gateway JVM uses non-blocking I/O on port 4002, which causes
EAGAIN errors with ibapi's blocking Client::connect(). The gnzsnz
image provides socat on port 4004 as a stable proxy that bridges
blocking clients to the non-blocking gateway socket.
Port 4004 is the correct client-facing port for paper mode (4003 for
live). The K8s readiness probe stays on 4002 since TCP probes work
fine with non-blocking sockets.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Readiness/liveness probes were checking port 4004 (socat), which
always listens even when the gateway is stuck at login — masking
failures. Now probes check port 4002 (TWS API), which only opens
after successful IBKR authentication.
Deploy strategy changed to Recreate because IBKR allows only one
session per account. RollingUpdate starts the new pod before killing
the old one, causing both to fight over the session in a crash loop.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The gnzsnz/ib-gateway image renders its IBC config.ini from a template
using envsubst. The ExistingSessionDetectedAction field was rendering
as empty, causing IBC to show a GUI dialog when IBKR detected a
competing session — which nobody could dismiss in a headless pod.
Setting primaryoverride tells IBC to automatically take over any
existing session, which is the correct behavior for a K8s deployment.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add Broker variant to Commands enum with BrokerArgs (flatten),
route to execute_broker_command in match block (no JWT required),
exit(1) on check failure. Clone config.api_gateway_url to avoid
partial move. Add two CLI parsing tests for broker check.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Copy broker_gateway.proto from broker_gateway_service into fxt/proto/
and wire it into build.rs compile_protos + lib.rs proto module so the
CLI can call BrokerGatewayService RPCs (health check, account state,
session status).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Adds IbkrConfig (host/port/client_id with defaults for paper trading)
and BrokerConfig (gateway_url + ibkr nested) to the TLI config file.
Both structs derive serde defaults so existing config files remain
backward-compatible.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Two-layer IBKR connectivity validation: direct ibapi handshake +
broker-gateway gRPC health check. Pure-client architecture using
ibapi as optional dep on fxt (not trading_engine).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Kapsule autoscaler already handles GPU node scale-down
(10min idle timeout). This CronJob only printed a message
and served no purpose.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Grafana now stores users, preferences, and state in our existing
PostgreSQL instance (dedicated 'grafana' database). This eliminates
the 2Gi PVC dependency — all persistent state lives in Postgres,
dashboards in ConfigMaps, datasources in Helm values.
Also adds Grafana to Postgres NetworkPolicy ingress allowlist.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The Grafana PVC was lost during cleanup (orphaned PV deletion).
Dashboards survived (ConfigMap-based), but user accounts and
datasources were lost.
This commit captures all Helm values previously set via --set:
- Sidecar config with foldersFromFilesStructure for folder organization
- Loki + Tempo datasources alongside Prometheus
- Platform node toleration for gitlab taint
Also fixed CI/CD folder annotation (slash caused nested directory).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
ci-rl pool was permanently disabled (enable_ci_rl_pool=false), all
training consolidated on ci-training (L40S). foxhunt-binaries S3
bucket was never created — binaries now deploy via shared PVC.
Also: deleted orphaned infra/live/production/ci-runner/ directory,
cleaned up stale Grafana ReplicaSets, recreated missing grafana PVC,
deleted orphaned Released PV, synced runner Helm configmap.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
CI executor pods (label foxhunt-ci) were selected by allow-monitoring-scrape
(which uses matchExpressions In [foxhunt, foxhunt-ci]) making them
policy-controlled, but allow-dns only selected foxhunt pods via matchLabels.
This blocked DNS resolution → git clone failure in deploy jobs.
- Update allow-dns podSelector to matchExpressions In [foxhunt, foxhunt-ci]
- Add ci-egress policy granting broad egress for ephemeral CI pods
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
DQN and PPO trainers now resolve hidden_dim_base from GPU VRAM when no
explicit override is given, so L4/L40S/H100 GPUs use proportionally
larger networks instead of being stuck at RTX 3050 Ti defaults (256).
- Add resolve_hidden_dim_base() tiered lookup (256/512/768/1024 by VRAM)
- Add network_param_count() for accurate model size estimation
- DQN: pre-compute hidden dims, use real param count for batch sizing
- PPO: add hidden_dim_base field, VRAM resolution in PpoTrainer::new()
- PPO hyperopt: raise hidden_dim_base ceiling from 2048 to 4096
- TFT hyperopt: expand hidden_sizes from [128,256,512] to 5 tiers
- Update stale estimates (DQN 50K→200K, PPO 100K→400K params)
- Fix pre-existing clippy lints in prefetch.rs and ppo.rs
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
