Policy Evaluation and Approval

Select an evaluation, compare it to the active production policy, and approve when ready.

Policy Evaluations

Select a row to open the compare and approve panel below.

Loading policy evaluations…

Category	Policy	Status	Score Δ	Return	Horizon
Loading policy evaluations…

Evaluation detail

Close

Select View on an evaluation row to inspect the policy evaluation run.

Compare & approve

Selected evaluation vs active production policy on the same ranking periods.

Reviewing

Approve

Candidate ticker diff

Waiting for lane data...

Waiting for lane data...

Inspect Policy

Review the persisted weights, filters, and execution settings for the selected model.

Close

Select a policy to inspect.

Edit factors and triggers in Section 1 — Policy creation or change.

Policy Creation or Change

Define the goal, tune the policy contract, then save before running or approving.

Policy context

Select an existing policy to revise its goal and contract, or create a new policy evaluation in the category.

Policy action

Policy action

Create new policy Update existing policy

Policy Category Select policy category Growth Yield Trade Policy Choose policy…

Goal

Write a brief, then interpret the goal before Save policy.

Goal name Goal brief

No interpreted goal yet.

Interpret goal

Save goal

Policy contract

Factor and trigger search ranges, scoring, run design, and evaluation window for the next run.

Trigger planner

Tune only persisted lane triggers and existing numeric thresholds. Trigger fields and operators remain fixed in v1.

Select a lane to load trigger controls.

Factor weight planner

Include lane factors in the search and set min / max / step. Unchecked = locked at baseline. Defaults may auto-lock factors to stay within the run budget.

Select a lane to load factor controls.

Scoring & planner

Optimisation objective Composite Excess return Average return Sharpe ratio Sortino ratio Success rate

Planner controls

Preset, group tuning mode, and granularity write back to persisted search_space and planner_state; they are not a second source of truth.

Planner preset Balanced Coarse search Focused search Factor search mode Grid search Fixed baseline Factor granularity Coarse Fine Trigger granularity Coarse Fine

Enable factor tuning Enable trigger tuning

Draft state: saved

Composite weights

Used only when the objective is Composite; persisted with the lane configuration.

Average return Sharpe ratio Sortino ratio Success rate

Run design

Controlled mutation proposes bounded variants per generation. Grid optimisation walks the contract cross-product up to the run budget cap.

Experiment mode Controlled mutation Grid optimisation Run budget (max combinations) Caps grid search size and how many factors you can unlock in the factor planner (e.g. 729 for six factors).

Mutation budget Variants per generation. Survivors Generations Random seed

Evaluation

Evaluation window planner

Each evaluation run samples ranking dates inside this window. Lists define search grids; the first value mirrors lane scalars when you save or validate.

Lookback days Depth of history each snapshot draws from. Max ranking dates How many ranked dates are sampled per path.

Sampling cadence Spacing between sampled ranking dates (fewer steps = denser history).

All Weekly Fortnightly Monthly

Final list planner

Variants for shortlist size, outcome horizon, and protective exit — each list expands the combination grid the search evaluates.

Selection counts Names kept after ranking gate. Forward days Holding period for measuring outcome. Stop loss % Exit if price falls this far below entry.

Advanced planner

Walk-forward split (TVT)

Train/validate/test ratios partition the ranking-date window for role-aware evaluation. Test is the residual.

Train ratio Validate ratio

Test ratio is computed as 1 − train − validate. Leave train/validate blank to use legacy single-split evaluation.

Search restrictions

Enumeration-time filters: required factors are always present, excluded ones are always absent. Same for triggers.

Required factors Excluded factors Required triggers Excluded triggers

Survivor selection + checkpoint

Survivor mode chooses which top results seed the next mutation generation; checkpoint persistence saves run state every generation so timed-out runs can be resumed.

Survivor mode Default (best) best top_n diversity Top N Persist each generation Snapshots run state every generation so a timed-out run can be resumed from partial.

{} {}

Validate contract Save policy

Select a policy, interpret the goal, tune the contract, then save.

Policy Optimisation Schedule

Configure scheduled policy evaluations and evaluation runs.

Scheduled Policy Evaluations

Select a row to edit its schedule or run the saved research objective setup now.

Category	Policy	Schedule Status	Last Evaluation	Actions
Loading scheduled policy evaluations…

Last Evaluation

-

Schedule Status

-

Last Status

Idle

Evaluation Budget

-

Schedule Status

Disabled Enabled

Schedule Mode

Manual only Every N hours Daily time

Interval Hours

Start Time

Save Schedule Run Evaluation Now Grid & Mutate

Select a policy

Select a policy to load scheduling status.

Approved Live Policies

The policies and versions used for live candidate selection.

Loading...

Evaluation detail

Close

Select a run to inspect its detail.

---