Skip to content

workflow.md

Latest commit

 

History

History
184 lines (129 loc) · 5.96 KB

workflow.md

File metadata and controls

184 lines (129 loc) · 5.96 KB

Workflow

This document explains the real user flows that are implemented today, the expected result for each flow, and where the flow is still partial.

Workflow 1: Import and save a dataset

User steps

  1. Open /imports
  2. Select Excel or CSV
  3. Enter a local file path
  4. Set symbol, timeframe, and timezone
  5. Review or edit the inferred column mapping
  6. Click Preview source
  7. Click Save mapping

Backend flow

  1. Frontend posts DataSourceConfig to POST /data-sources/preview
  2. DatasetService.preview asks DataManager.preview_import for sample rows and inferred mapping
  3. Frontend shows preview information
  4. Frontend posts the same DataSourceConfig to POST /data-sources/save
  5. DatasetService.save loads the full file, normalizes timestamps and OHLCV columns, writes a dataset CSV, and stores dataset metadata

Expected result

  • preview shows row count, columns, active sheet, inferred mapping, and sample rows
  • save returns a dataset_id
  • saved dataset appears in /imports and becomes available for run creation

Current limitations

  • no native file picker yet; file path is typed or pasted manually
  • polygon exists in the UI contract but the backend loader currently supports only file-based Excel and CSV sources

Workflow 2: Replay a Python strategy against a dataset

User steps

  1. Save a dataset in /imports
  2. Open /workspace
  3. Paste or edit the Python strategy in the Python editor
  4. Optionally paste Pine code in the Pine editor
  5. Choose run mode and timeframe
  6. Click Run replay

Backend flow

  1. Frontend posts ReplayRunRequest to POST /runs/replay
  2. RunService.create_replay_run loads the saved dataset
  3. PythonStrategyEngine.execute runs the Python strategy against the normalized frame
  4. Candle data is normalized into CandlePoint[]
  5. If a bridge artifact is attached, Pine series and Pine trades are loaded from it
  6. ComparisonEngine compares Pine and Python outputs
  7. The run is persisted and returned

Expected result

  • Python screen shows candles
  • Python price-like indicator series overlay on the candle chart
  • mismatch analysis appears in the diff panel
  • run appears in /runs

Current limitations

  • only price-like indicator series are overlaid directly on the candle chart
  • oscillator-like series are not rendered in a second pane yet
  • Pine results only appear if a bridge artifact is selected

Workflow 3: Dataset-backed live playback

User steps

  1. Save a dataset
  2. Open /workspace
  3. Click Start live run

Backend flow

  1. Frontend posts the same run payload to POST /runs/live
  2. RunService.create_live_run uses warmup bars first
  3. A background thread increments the active dataset by one bar every second
  4. Python outputs are recalculated on the growing frame
  5. Comparison results are refreshed
  6. Frontend subscribes to /runs/{run_id}/stream
  7. On every event, the frontend refreshes the run from GET /runs/{run_id}

Expected result

  • the selected run enters live
  • candle count increases over time
  • comparison state changes as more bars arrive
  • /runs shows progress like current/total

Current limitations

  • the source is a saved historical dataset, not a provider-backed live feed
  • there is no pause/resume control yet
  • live threading is in-process, not a separate worker service

Workflow 4: Attach Pine truth through a bridge artifact

User steps

  1. Open /settings
  2. Paste a JSON artifact into Bridge JSON payload
  3. Click Save bridge artifact
  4. Return to /workspace
  5. Select the artifact from the Pine bridge artifact dropdown
  6. Run replay or live mode again

Backend flow

  1. Frontend posts to POST /pine-bridge/artifacts
  2. BridgeService.create persists the artifact as JSON
  3. RunService resolves the selected artifact during run creation
  4. Pine indicator series and Pine trade events are used as comparison truth

Expected result

  • Pine screen receives candle data from the run
  • Pine indicator series can be overlaid when the uploaded series is price-like
  • mismatch analysis compares Python outputs against the uploaded Pine values

Current limitations

  • TradingView automation is not implemented
  • the user must provide the exported JSON payload manually
  • non-price-like Pine series are listed as hidden rather than rendered in a sub-pane

Workflow 5: Ask the local LLM for analysis

User steps

  1. Open /workspace
  2. Choose a local chat model from the LLM panel
  3. Click a starter prompt or type a custom prompt
  4. Click Ask LLM

Backend flow

  1. Frontend calls GET /chat/models during startup and refreshes on demand
  2. The selected model is sent to POST /chat
  3. ChatService calls OllamaClient.chat
  4. OllamaClient sends the latest user message to Ollama with a plain-text-only system instruction
  5. Returned text is sanitized
  6. Frontend displays the cleaned response or a fallback status

Expected result

  • user sees which local Ollama models are available
  • embedding-only models are not auto-selected for chat
  • the assistant returns a plain-text answer without transcript noise

Current limitations

  • the current workspace sends intent=analysis only
  • there is no token streaming in the frontend yet
  • patch application is not wired from the UI into files

Workflow 6: Permissions and approvals

User steps

  1. Use the approval queue in /workspace
  2. Toggle a permission row to approve or revoke

Backend flow

  1. Frontend calls POST /permissions/grant
  2. PermissionManager appends the grant to permission history
  3. Chat requests can inspect active grants
  4. apply_fix intent is rejected if Python write access is not active

Expected result

  • permission history appears in the approval queue
  • write-sensitive chat actions can be guarded

Current limitations

  • approval entries are append-only history, not a full RBAC system
  • the current UI does not yet send apply_fix requests
  • permission toggles change app behavior, but no file patch is applied automatically