Editing & Reuse

For the complete documentation index, see llms.txt. Markdown versions of documentation pages are available by appending .md to page URLs; this page is available as Markdown.

AI Chart Analysis — Editing & Reuse

Refinement Commands

Once an analysis is on the chart, you refine it the same way you started it — in plain English:

Use the previous swing low

Remove Fibonacci

Show the daily resistance

Use closing prices

Only keep confirmed levels

Make this analysis simpler

Show the bearish interpretation

Analyze only today's session

Keep the range and remove everything else

Recalculate using the broader impulse

The Refinement Loop

Every refinement is planned against your active analysis context — the previous engine result's object IDs, types, states, and any edits you've made (see User Flow). That context is what makes "the previous swing low" resolvable at all: the planner can see swing_low_1, swing_low_2, and so on, and binds your instruction to the right one by ID rather than guessing.

Refinement plans carry mode: "refinement", a parent_request_id, and explicit retained_objects / removed_objects diffs (see Data Contracts). The engine recalculates only what the diff touches — a refinement updates your current analysis, it never spins up a disconnected new one.

Local Refinements (No AI Round-Trip)

Some commands are simple and unambiguous enough that the UI and engine execute them directly, with no AI call — which is also why they're instant:

• Remove / hide / show a named layer ("remove Fibonacci," or tapping a layer toggle)

• Lock / unlock a drawing

• Simplify (drops every secondary-priority object)

• Undo / redo

• Clear analysis

These complete in 150 milliseconds or less (see Reliability & Limits). Anything that actually requires interpretation — "use the broader impulse," "show the bearish interpretation" — goes through the AI planner instead.

User Edit Persistence

Your manual edits are treated as durable, not as a suggestion the system can overwrite on the next pass:

1. When you edit an object's anchors, it's flagged user_edited: true and behaves like a locked object for automatic changes.

2. On any later refinement or rerun, your edited objects are preserved by default — the engine recalculates everything else around them.

3. If a refinement explicitly targets an object you've edited ("recalculate the fib"), the system asks one inline question — "Recalculate and discard your manual anchors?" — with a keep/discard choice. This is one of the few places the feature will interrupt you with a prompt.

4. "Restore AI-selected anchors" is always available per object, reverting it back to the engine's own anchors.

5. Contradictory refinements in the same session (say, "remove RSI" then later "add RSI") simply apply in order — the latest instruction wins, and no preference is inferred from a pair of instructions that contradict each other.

Saving and Reuse

There are two distinct ways to keep an analysis for later, and they behave very differently:

Save Snapshot captures the current chart analysis exactly as it stands — symbol, timeframe, scope, prompt and refinement history, the plan(s) with their schema version, engine results, drawings, indicators, states, and any edits you made. Reopening a snapshot renders those exact stored coordinates; it is never recalculated.

Save Recipe captures a reusable method instead of a one-time result:

Breakout Analysis (recipe, schema 1.0)

Always include:
- Range boundaries
- Breakout level
- Relative volume
- Retest zone

Include when relevant:
- Fibonacci retracement
- Higher-timeframe resistance
- Failed breakout warning

Maximum drawings: 6

A saved recipe can be run later on the current chart, a different symbol, a different timeframe, a watchlist item, or a selected chart region.

Rerun always uses current market data while keeping the saved recipe and its preferences intact. It produces a fresh execution (mode: "recipe_run") — not a replay of old coordinates. Schema versioning ensures recipes keep working across releases; see Data Contracts for how migration is handled.

User Controls

The full control surface available at any point in an analysis:

• Analyze

• Stop analysis

• Undo / Redo

• Clear analysis

• Simplify

• Explain

• Save snapshot / Save recipe / Rerun

• Lock drawings

• Hide individual layers

• Edit drawing anchors

• Restore AI-selected anchors

• Freeze preference learning

AI-created drawings are always visually distinguishable from your own drawings, and any AI drawing you've edited is marked as such.

Trust and Transparency

Every AI-generated drawing is explainable. Selecting one reveals:

• What the object represents

• Why it was added

• Which data points were used

• Which parameters were applied

• Its state (developing / provisional / confirmed / invalidated) and its full confirmation criteria table

• Exactly what would invalidate it — the published rule

• Whether you've edited it

The system is always explicit about what kind of thing you're looking at, distinguishing between:

• Measured fact

• Rule-based evaluation, with its criteria

• AI-selected interpretation

• Your own preference — whether explicit or inferred

• An alternative interpretation

Every plan and result is auditable: the pass-through selection source, the plan diff, and the engine version are all stored alongside each saved snapshot.