Case Study Tab

Basic or Professional

The Case Study tab is the research ledger for Algo Lab. It binds a hypothesis to a Pine strategy tab, then keeps versions, trials, linked reports, Algo Agent evidence, and promotion decisions together.

Use this page as a reference for what each Case Study area stores and when to use it.

Algo Lab Case Study tab with the study list, variant tree, trial summary, evidence state, and candidate actions

Select a study on the left, a version or scope in the tree, and then inspect its summary, trials, and evidence before taking a candidate action.

Core Concepts

Concept	Meaning
Case Study	A named research investigation with a hypothesis, scope, linked Pine file, versions, trials, reports, evidence, and jobs.
Linked Pine tab	The editor tab the study is investigating. Loading a candidate into the linked tab keeps the research attached to the same file.
Version	A saved Pine source snapshot. Versions can come from manual saves, parameter work, or Algo Agent candidates.
Trial	A run result attached to a version and scope. Trials can be baseline runs, parameter rows, holdout checks, cross-market tests, or Algo Agent evaluations.
Evidence	The readiness state behind a candidate: baseline comparison, holdout result, trade count, stability, linked reports, and gate-policy status.
Promotion	A research marker for the best candidate. Promotion does not overwrite the editor tab and does not deploy or execute code.

When to Create One

Create or select a Case Study before serious optimization. Scratch runs are fine while checking syntax or data access, but use a Case Study when any of these are true:

You are comparing more than one version of the same strategy.
You want the Algo Agent to run a bounded research loop.
You want to save parameter sweep results.
You need a record of why one candidate was promoted over another.
You are linking Backtest vs Alerts or Alerts vs Fills reports to the research.

Panel Structure

The Case Study tab has a header, a left study list, and a detail panel.

Area	Purpose
Header	Shows the title, symbol, timeframe, date range, hypothesis, and rename controls.
Study list	Selects, creates, renames, archives, or deletes studies. The list helps you switch between investigations without losing the editor context.
Results	Shows the version/trial tree and detail tabs for `Summary` and `Trials`.
Evidence	Shows Algo Agent evidence maturity for the run or selected variant.
Gate policy	Edits research gates once the study has recorded Algo Agent evidence.
Jobs	Lists server-side research jobs associated with the Case Study.

Results View

The Results view is the main research table. It can group records two ways:

Grouping	Use when
By variant	You want to inspect each source version and see its related trials.
By scope	You want to compare how versions behaved across symbol/timeframe/date-range scopes.

Selecting a tree node changes the detail area:

Summary shows the selected version, best available trial, evidence state, linked reports, source actions, and promotion status.
Trials shows all runs associated with the selected version or scope, including origin, kind, status, metrics, settings, inputs, duration, warnings, and error details.

Version Actions

The Case Study page separates source review from source replacement.

Action	Result
View code	Opens the stored source for inspection without changing the editor.
Stage in editor	Creates a reviewable diff in the editor. Accepting or rejecting follows the same staged-candidate path used by the Algo Agent.
Load into tab name	Replaces the named, linked Pine tab with that version’s source after any required confirmation. Use this when the candidate belongs to the same investigation.
Load into editor	Creates a new editor tab when the study is not linked to an open tab.
Run on chart	Runs the selected version against the current chart scope without first making it the permanent editor source.
Delete version	Removes a saved source snapshot after confirmation. This can affect related trial visibility.

Trial Types

Trials are run records. Common origins include:

Trial source	What it records
Manual run	A direct backtest from the current editor tab and chart scope.
Parameter sweep	One row from Params Optimization, including the input values used for that row when available.
Algo Agent evaluation	A candidate evaluation created by the Algo Agent’s tool loop.
Holdout validation	A check against a held-out data slice.
Cross-market test	A test of the selected version against another symbol or market scope.

Use the Trials tab to confirm whether a candidate only won once, won across several scopes, or failed under a more realistic validation path.

Linked Reports

Linked reports connect the Case Study to evidence produced outside the Lab:

Backtest vs Alerts reports help explain whether TradingView alerts match the strategy logic.
Alerts vs Fills reports help explain fill quality, slippage, and live behavior after alerts fire.

Link reports when the research claim depends on alert or fill evidence. Unlink reports when they no longer describe the active hypothesis or version family.

Gate Policy and Evidence

Gate policy defines what counts as acceptable research evidence for Agent-driven work. A gate can be informational, mandatory, or blocking depending on the policy stored on the Case Study.

Typical gates include:

Baseline comparison
Holdout pass
Minimum trade count
Edge margin over baseline
Stability across relevant scopes
Linked operational evidence when live/fill behavior matters

The Evidence tab shows maturity. The Gate policy tab lets you update the policy after Algo Agent evidence exists. Existing studies keep their own policy; the Configuration tab only controls defaults for new Case Studies.

Recommended Workflow

Create or select a Case Study before optimization.
Write the hypothesis in plain language. Example: “Improve the AAPL 10-minute RSI baseline without increasing drawdown or reducing trade count below the minimum gate.”
Run a baseline from the Editor tab, then add the result to the Case Study.
Save meaningful versions before running parameter sweeps or asking the Algo Agent for structural changes.
Record trials from manual runs, parameter sweeps, holdout checks, and Algo Agent evaluations.
Link reports when alert behavior or fill behavior is part of the decision.
Promote one candidate only when its evidence is stronger than the baseline and its risks are understood.

Operational Rules

Promotion is a research decision, not a deployment.
The editor tab is only changed when you load, stage, or accept source.
The Algo Agent uses the active Case Study state; select the right study before starting a run.
Keep one hypothesis per Case Study. Start a new study when the objective or strategy family changes materially.
Use the linked tab for same-file candidates. Use a new tab for deliberate branches.
If a result is not reproducible under the same scope and settings, do not treat it as promoted evidence.

Troubleshooting

Symptom	What to check
A trial is not attached to the expected study	Confirm the correct Case Study was active before the run.
A version does not show the `in editor` chip	The active editor source hash does not match that saved version.
Algo Agent evidence is missing	The Algo Agent may not have committed evidence yet, or the run may have stopped before a ledger-changing tool call.
Gate policy is not editable	Gate policy appears after the study has recorded Algo Agent evidence.
A linked report looks stale	Open the report in its native page and confirm it matches the same symbol, timeframe, date range, and version family.
Candidate source would replace local edits	Use staged diff review or open a new editable tab instead of loading directly into the linked tab.

Algo Agent How Agent runs commit evidence into Case Studies

Params Optimization Save full sweeps into the active Case Study

Backtesting Report Add important run results to the ledger