feat(xfa): expose raw scripts via FormSchema.Scripts; drop heuristic Rule API

GAPS.md

@@ -105,25 +105,19 @@ and `/CA` (stroke alpha) at the requested opacity and applies it via the `gs`
 operator. `PageBuilder` gained an `extgstates` map and a `Build()` block that
 emits `/ExtGState << ... >>` in the page resources dictionary.
 
-### ~~XFA script parsing is approximate~~ ✅ Fixed in v1.2.0
-`convertXFAEventToRule` now dispatches to a structured `parseXFAScript`
-analyser that detects language (FormCalc vs JavaScript), then tries four
-pattern families in order:
-- **Visibility** — `$.presence = "visible"/"hidden"` → `RuleTypeVisibility` +
-  `ActionTypeShow`/`ActionTypeHide` with extracted target field
-- **SetValue** — `$.rawValue = expr` or `xfa.resolveNode("x").rawValue = expr`
-  → `RuleTypeSetValue` + `ActionTypeSetValue` with extracted expression
-- **Validation** — scripts containing `return true/false` or message-box calls
-  → `RuleTypeValidate` + `ActionTypeValidate`
-- **Calculate** — scripts using FormCalc built-ins (`Sum`, `Avg`, `Concat`, …)
-  or JavaScript `return expr` → `RuleTypeCalculate` + `ActionTypeCalculate`
-  with the extracted expression
-
-Conditional guards (`if (cond) then … endif` / `if (cond) { … }`) are parsed
-into a `*Condition` with operator, field reference, and literal value where the
-expression is a simple binary comparison. The full raw script is always
-preserved in `Action.Script` for evaluation. Complex scripts that match none of
-the above patterns fall back to `ActionTypeExecute` as before.
+### ~~XFA script parsing is approximate~~ ✅ Superseded in v2.0.0
+The structured `parseXFAScript` analyser shipped in v1.2.0 (regex-based
+visibility / set-value / validate / calculate classification, populating
+`Rule`/`Condition`/`Action`) was lossy and incorrect on non-trivial scripts —
+faithful interpretation needs a real ES5/FormCalc AST, which is out of scope
+for a PDF library. v2.0.0 removes the heuristic and the `Rule`/`Action` type
+surface entirely; `FormSchema.Scripts []FormScript` now exposes verbatim
+script bodies with their event activity, language (`javascript` | `formcalc`),
+SOM owner path, and owner ID, leaving interpretation to the caller. See the
+**Forms** section of the README for usage and the type comment on
+`types.FormScript` for known gaps (scripts attached to nodes pdfer does not
+surface — decorative `<draw>`, `bind="none"` non-AddAttachment buttons,
+`<pageArea>` events, per-option `<field>`s collapsed into an `<exclGroup>`).
 
 ---
 
@@ -340,7 +334,7 @@ CJK vertical text, some Arabic/Hebrew layout engines.
 
 ---
 
-### ~~XFA browser rendering gaps~~ ✅ Fixed in v1.4.0
+### ~~XFA browser rendering gaps~~ ✅ Fixed in v2.0.0
 
 Fourteen improvements to the XFA → `FormSchema` translation for browser rendering fidelity:
 

README.md

@@ -178,6 +178,21 @@ filled, err := form.Fill(pdfBytes, pdfer.FormData{"FirstName": "Alice"}, nil, fa
 out, err := pdfer.FlattenForm(filled, nil, false)
 ```
 
+For XFA forms, `schema.Scripts` exposes raw `<script>` blocks verbatim:
+
+```go
+for _, s := range schema.Scripts {
+    fmt.Printf("[%s] %s (%s) on %s\n%s\n", s.Event, s.Name, s.Language, s.OwnerPath, s.Body)
+}
+```
+
+`FormScript.Body` is the unmodified source. pdfer does not interpret
+FormCalc or JavaScript semantics — callers that need to evaluate scripts
+should plug in their own parser. `Question.Scripts` and `FormSection.Scripts`
+hold `FormScript.ID` references in declaration order. See
+[xfa-web](https://github.com/benedoc-inc/xfa-web) for one example of an
+interactive renderer built on this contract.
+
 ### Content extraction
 
 ```go
@@ -387,7 +402,7 @@ See [GAPS.md](GAPS.md) for the full history and detailed file pointers.
 **Forms**
 - `Form.Validate()` returns "not implemented" for XFA forms — structural extraction only.
 - Calculated form fields are not re-evaluated on `Fill()`; dependent fields remain stale until opened in a viewer.
-- XFA script parsing handles common patterns (visibility, set-value, validate, calculate) and falls back to `ActionTypeExecute` for scripts it cannot classify.
+- XFA scripts are exposed verbatim via `FormSchema.Scripts` — pdfer does not interpret FormCalc or JavaScript. Scripts attached to nodes pdfer doesn't surface in the schema (decorative `<draw>`, `bind="none"` non-AddAttachment buttons, `<pageArea>` events, per-option fields collapsed into an `<exclGroup>`) are not extracted.
 
 **Images / encoding**
 - JPEG2000 (`JPXDecode`) and JBIG2 image streams are detected but not decoded.

examples/extract_xfa/main.go

@@ -73,7 +73,7 @@ func main() {
 		} else {
 			fmt.Printf("\nParsed Form Structure:\n")
 			fmt.Printf("  Questions: %d\n", len(form.Questions))
-			fmt.Printf("  Rules: %d\n", len(form.Rules))
+			fmt.Printf("  Scripts: %d\n", len(form.Scripts))
 
 			if len(form.Questions) > 0 {
 				fmt.Println("\nFirst 10 questions:")

forms/acroform/parser.go

@@ -301,7 +301,6 @@ func (af *AcroForm) ToFormSchema() *types.FormSchema {
 			FormType: "AcroForm",
 		},
 		Questions: make([]types.Question, 0),
-		Rules:     make([]types.Rule, 0),
 	}
 
 	for _, field := range af.Fields {