fix(compiler+benchmarks): emit compute expressions, fix fact indices, add QEMU support

Compiler fixes:
- canonical.py: extract compute expressions from rule 'then.compute' blocks
  into a flat CanonicalModel.expressions list with resolved fact IDs
  (UINT16_MAX sentinel for literal operands). Annotate each rule with
  _expr_start/_expr_count for the emitter.
- emit_c.py: add _EXPR_OP_MAP, emit model_expressions[] table, wire
  .expr_count and .expressions into the ARBITER_model struct, add
  .expr_start/.expr_count to each rule entry.

Benchmark fixes:
- Remove hand-coded fact enums that used the wrong alphabetical ordering.
  Replace with #defines aliasing generated ARBITER_FACT_* constants from
  arbiter_model.h. Fixes type mismatch warnings and wrong Kalman output.
- Increase BENCH_ITERATIONS 1k->100k for measurable ns counts on native_sim.

QEMU support:
- Add qemu_x86 to platform_allow in both benchmark testcase.yaml files.
  Local: west build -p -b qemu_x86 app/tests/benchmarks/kalman_benchmark

Tests (26 passing):
- tests/python/test_emit_c.py: 14 new tests for expression emission.

Regenerated arbiter_model.{c,h} for all 19 models with fixed compiler.

Co-Authored-By: Oz <oz-agent@warp.dev>

Author: tbitcs

.gitignore

@@ -46,3 +46,6 @@ coverage.xml
 # specsmith machine-local state
 .chronomemory/
 .specsmith/*.bak
+
+twister-results-*/
+

python/arbiter/canonical.py

@@ -20,6 +20,7 @@ class CanonicalModel:
     conditions: list[dict[str, Any]]
     actions: list[dict[str, Any]]
     modes: list[dict[str, Any]]
+    expressions: list[dict[str, Any]] = field(default_factory=list)
     hazards: list[dict[str, Any]] = field(default_factory=list)
     safety_goals: list[dict[str, Any]] = field(default_factory=list)
     model_hash: str = ""
@@ -46,6 +47,10 @@ def max_conditions(self) -> int:
     def max_actions(self) -> int:
         return len(self.actions)
 
+    @property
+    def max_expressions(self) -> int:
+        return len(self.expressions)
+
 
 def canonicalize(data: dict[str, Any]) -> CanonicalModel:
     """Canonicalize a parsed ARB model.
@@ -70,17 +75,33 @@ def canonicalize(data: dict[str, Any]) -> CanonicalModel:
     actions = sorted(data.get("actions", []), key=lambda a: a.get("id", ""))
     action_id_map = {a["id"]: i for i, a in enumerate(actions)}
 
-    # Sort rules by id and flatten conditions
-    rules = sorted(data.get("rules", []), key=lambda r: r.get("id", ""))
-    rule_id_map = {r["id"]: i for i, r in enumerate(rules)}
+    # Sort rules by id and flatten conditions + expressions
+    rules_raw = sorted(data.get("rules", []), key=lambda r: r.get("id", ""))
+    rule_id_map = {r["id"]: i for i, r in enumerate(rules_raw)}
 
-    # Flatten condition trees from all rules
+    # Flatten condition trees and compute expressions from all rules.
+    # Annotate each rule dict with expr_start / expr_count so the emitter
+    # can write them into the rules table without a second pass.
     conditions: list[dict[str, Any]] = []
-    for rule in rules:
+    expressions: list[dict[str, Any]] = []
+    rules: list[dict[str, Any]] = []
+    for rule in rules_raw:
         when = rule.get("when", {})
         if isinstance(when, dict):
             _flatten_conditions(when, conditions, fact_id_map)
 
+        expr_start = len(expressions)
+        rule_exprs = _flatten_expressions(
+            rule.get("then", {}), fact_id_map
+        )
+        expressions.extend(rule_exprs)
+
+        # Attach start/count to a copy of the rule so we don't mutate input.
+        annotated = dict(rule)
+        annotated["_expr_start"] = expr_start
+        annotated["_expr_count"] = len(rule_exprs)
+        rules.append(annotated)
+
     model = CanonicalModel(
         name=data.get("model", "unnamed"),
         arb_version=data.get("arb_version", 0.1),
@@ -89,6 +110,7 @@ def canonicalize(data: dict[str, Any]) -> CanonicalModel:
         conditions=conditions,
         actions=actions,
         modes=modes,
+        expressions=expressions,
         hazards=data.get("hazards", []),
         safety_goals=data.get("safety_goals", []),
         fact_id_map=fact_id_map,
@@ -106,6 +128,73 @@ def canonicalize(data: dict[str, Any]) -> CanonicalModel:
     return model
 
 
+_UINT16_MAX = 65535  # UINT16_MAX sentinel for "use literal"
+
+_EXPR_OP_ALIASES: dict[str, str] = {
+    "assign": "assign", "add": "add", "sub": "sub", "mul": "mul",
+    "div": "div", "mod": "mod", "abs": "abs", "negate": "negate",
+    "min": "min", "max": "max", "clamp": "clamp",
+    "shift_r": "shift_r", "shift_l": "shift_l",
+    "scale": "scale", "accumulate": "accumulate",
+}
+
+
+def _flatten_expressions(
+    then: Any,
+    fact_id_map: dict[str, int],
+) -> list[dict[str, Any]]:
+    """Extract compute expressions from a rule's then block.
+
+    Each entry maps to an ``ARBITER_expr_def`` with resolved fact indices.
+    ``UINT16_MAX`` (65535) signals "use the literal instead of a fact".
+    """
+    if not isinstance(then, dict):
+        return []
+    compute = then.get("compute", [])
+    if not isinstance(compute, list):
+        return []
+
+    out: list[dict[str, Any]] = []
+    for expr in compute:
+        if not isinstance(expr, dict):
+            continue
+
+        target = expr.get("target", "")
+        target_id = fact_id_map.get(target, 0)
+
+        # Left operand
+        left_name = expr.get("left")
+        if left_name is not None and left_name in fact_id_map:
+            left_fact_id = fact_id_map[left_name]
+            left_literal: int = 0
+        else:
+            left_fact_id = _UINT16_MAX
+            left_literal = int(expr.get("left_literal", 0))
+
+        # Right operand
+        right_name = expr.get("right")
+        if right_name is not None and right_name in fact_id_map:
+            right_fact_id = fact_id_map[right_name]
+            right_literal: int = 0
+        else:
+            right_fact_id = _UINT16_MAX
+            right_literal = int(expr.get("right_literal", 0))
+
+        op = _EXPR_OP_ALIASES.get(expr.get("op", "assign"), "assign")
+        scale = int(expr.get("scale", 1))
+
+        out.append({
+            "target_fact_id": target_id,
+            "op": op,
+            "left_fact_id": left_fact_id,
+            "left_literal": left_literal,
+            "right_fact_id": right_fact_id,
+            "right_literal": right_literal,
+            "scale": scale,
+        })
+    return out
+
+
 def _flatten_conditions(
     when: dict[str, Any],
     conditions: list[dict[str, Any]],

python/arbiter/emit_c.py

@@ -33,6 +33,17 @@
     "raise_fault": "ARBITER_ACTION_RAISE_FAULT", "clear_fault": "ARBITER_ACTION_CLEAR_FAULT",
 }
 
+_EXPR_OP_MAP = {
+    "add": "ARBITER_EXPR_ADD", "sub": "ARBITER_EXPR_SUB",
+    "mul": "ARBITER_EXPR_MUL", "div": "ARBITER_EXPR_DIV",
+    "mod": "ARBITER_EXPR_MOD", "abs": "ARBITER_EXPR_ABS",
+    "negate": "ARBITER_EXPR_NEGATE", "min": "ARBITER_EXPR_MIN",
+    "max": "ARBITER_EXPR_MAX", "clamp": "ARBITER_EXPR_CLAMP",
+    "shift_r": "ARBITER_EXPR_SHIFT_R", "shift_l": "ARBITER_EXPR_SHIFT_L",
+    "scale": "ARBITER_EXPR_SCALE", "accumulate": "ARBITER_EXPR_ACCUMULATE",
+    "assign": "ARBITER_EXPR_ASSIGN",
+}
+
 
 def _c_str(s: str | None) -> str:
     return f'"{s}"' if s else "NULL"
@@ -190,10 +201,13 @@ def emit_c_source(model: CanonicalModel, header_name: str = "arbiter_model.h",
         raw_expl = then.get("explanation") if isinstance(then, dict) else None
         explanation = _c_str(raw_expl) if emit_trace_strings else "NULL"
 
+        expr_start = r.get("_expr_start", 0)
+        expr_count = r.get("_expr_count", 0)
         lines.append(
             f"\t{{ .id = {i}, .rule_class = {rclass}, "
             f".condition_start = {cond_offset}, .condition_count = {cond_count}, "
             f".action_start = {action_start}, .action_count = {action_count}, "
+            f".expr_start = {expr_start}, .expr_count = {expr_count}, "
             f".safety_goal_id = UINT16_MAX, .set_mode = {set_mode}, "
             f".safety_critical = {safety_critical}, "
             f".name = {name}, .explanation = {explanation} }},"
@@ -205,6 +219,26 @@ def emit_c_source(model: CanonicalModel, header_name: str = "arbiter_model.h",
     lines.append("};")
     lines.append("")
 
+    # Expressions table
+    expressions = getattr(model, "expressions", [])
+    if expressions:
+        lines.append("static const struct ARBITER_expr_def model_expressions[] = {")
+        for e in expressions:
+            op_enum = _EXPR_OP_MAP.get(e.get("op", "assign"), "ARBITER_EXPR_ASSIGN")
+            lines.append(
+                f"\t{{ .target_fact_id = {e['target_fact_id']}, "
+                f".op = {op_enum}, "
+                f".left_fact_id = {e['left_fact_id']}, "
+                f".left_literal = {e['left_literal']}, "
+                f".right_fact_id = {e['right_fact_id']}, "
+                f".right_literal = {e['right_literal']}, "
+                f".scale = {e['scale']} }},"
+            )
+        lines.append("};")
+    else:
+        lines.append("static const struct ARBITER_expr_def *model_expressions = NULL;")
+    lines.append("")
+
     # Mode names
     if model.modes and emit_trace_strings:
         lines.append("static const char *model_mode_names[] = {")
@@ -222,6 +256,7 @@ def emit_c_source(model: CanonicalModel, header_name: str = "arbiter_model.h",
     schema_hex = model.schema_hash[:64].ljust(64, "0")
     schema_bytes = ", ".join(f"0x{schema_hex[j:j+2]}" for j in range(0, 64, 2))
 
+    expr_count_total = len(getattr(model, "expressions", []))
     lines.extend([
         "const struct ARBITER_model ARBITER_generated_model = {",
         f'\t.name = "{model.name}",',
@@ -231,11 +266,13 @@ def emit_c_source(model: CanonicalModel, header_name: str = "arbiter_model.h",
         f"\t.rule_count = {model.max_rules},",
         f"\t.condition_count = {model.max_conditions},",
         f"\t.action_count = {model.max_actions},",
+        f"\t.expr_count = {expr_count_total},",
         f"\t.mode_count = {len([m for m in model.modes if isinstance(m, dict)])},",
         "\t.facts = model_facts,",
         "\t.rules = model_rules,",
         "\t.conditions = model_conditions,",
         "\t.actions = model_actions,",
+        "\t.expressions = model_expressions,",
         "\t.mode_names = model_mode_names,",
         "};",
         "",

samples/access_control/src/arbiter_model.c

@@ -46,18 +46,35 @@ static const struct ARBITER_action_def model_actions[] = {
 };
 
 static const struct ARBITER_rule_def model_rules[] = {
-	{ .id = 0, .rule_class = ARBITER_RULE_INFERENCE, .condition_start = 0, .condition_count = 1, .action_start = 0, .action_count = 0, .safety_goal_id = UINT16_MAX, .set_mode = 1, .safety_critical = false, .name = "01_cors.preflight", .explanation = "OPTIONS request — CORS preflight." },
-	{ .id = 1, .rule_class = ARBITER_RULE_INFERENCE, .condition_start = 1, .condition_count = 2, .action_start = 0, .action_count = 0, .safety_goal_id = UINT16_MAX, .set_mode = UINT16_MAX, .safety_critical = false, .name = "02_cors.allow_origin", .explanation = "Same-origin or allowed origin — CORS pass." },
-	{ .id = 2, .rule_class = ARBITER_RULE_CONSTRAINT, .condition_start = 3, .condition_count = 1, .action_start = 0, .action_count = 0, .safety_goal_id = UINT16_MAX, .set_mode = 2, .safety_critical = false, .name = "03_cors.block_origin", .explanation = "Blocked origin — 403." },
-	{ .id = 3, .rule_class = ARBITER_RULE_CONSTRAINT, .condition_start = 4, .condition_count = 1, .action_start = 0, .action_count = 1, .safety_goal_id = UINT16_MAX, .set_mode = 3, .safety_critical = false, .name = "10_rate.check", .explanation = "Client > 60 req/min — 429 Too Many Requests." },
-	{ .id = 4, .rule_class = ARBITER_RULE_INFERENCE, .condition_start = 5, .condition_count = 1, .action_start = 0, .action_count = 0, .safety_goal_id = UINT16_MAX, .set_mode = 0, .safety_critical = false, .name = "20_auth.public_path", .explanation = "Public path — no auth required." },
-	{ .id = 5, .rule_class = ARBITER_RULE_INFERENCE, .condition_start = 6, .condition_count = 2, .action_start = 0, .action_count = 0, .safety_goal_id = UINT16_MAX, .set_mode = 0, .safety_critical = false, .name = "21_auth.health_check", .explanation = "Health check GET — always allowed." },
-	{ .id = 6, .rule_class = ARBITER_RULE_CONSTRAINT, .condition_start = 8, .condition_count = 2, .action_start = 0, .action_count = 0, .safety_goal_id = UINT16_MAX, .set_mode = 2, .safety_critical = false, .name = "22_auth.no_token", .explanation = "Auth-required path but no token — 401." },
-	{ .id = 7, .rule_class = ARBITER_RULE_CONSTRAINT, .condition_start = 10, .condition_count = 2, .action_start = 0, .action_count = 0, .safety_goal_id = UINT16_MAX, .set_mode = 2, .safety_critical = false, .name = "23_auth.invalid_token", .explanation = "Invalid/expired token — 401." },
-	{ .id = 8, .rule_class = ARBITER_RULE_CONSTRAINT, .condition_start = 12, .condition_count = 2, .action_start = 0, .action_count = 0, .safety_goal_id = UINT16_MAX, .set_mode = 2, .safety_critical = false, .name = "24_auth.admin_check", .explanation = "Admin path requires admin role — 403." },
-	{ .id = 9, .rule_class = ARBITER_RULE_INFERENCE, .condition_start = 14, .condition_count = 3, .action_start = 0, .action_count = 0, .safety_goal_id = UINT16_MAX, .set_mode = 0, .safety_critical = false, .name = "25_auth.valid_token", .explanation = "Valid token, authorized — allow." },
-	{ .id = 10, .rule_class = ARBITER_RULE_INFERENCE, .condition_start = 17, .condition_count = 1, .action_start = 0, .action_count = 0, .safety_goal_id = UINT16_MAX, .set_mode = UINT16_MAX, .safety_critical = false, .name = "30_risk.compute", .explanation = NULL },
-	{ .id = 11, .rule_class = ARBITER_RULE_CONSTRAINT, .condition_start = 18, .condition_count = 1, .action_start = 0, .action_count = 0, .safety_goal_id = UINT16_MAX, .set_mode = 2, .safety_critical = false, .name = "40_payload.too_large", .explanation = "Payload > 64KB — rejected." },
+	{ .id = 0, .rule_class = ARBITER_RULE_INFERENCE, .condition_start = 0, .condition_count = 1, .action_start = 0, .action_count = 0, .expr_start = 0, .expr_count = 2, .safety_goal_id = UINT16_MAX, .set_mode = 1, .safety_critical = false, .name = "01_cors.preflight", .explanation = "OPTIONS request — CORS preflight." },
+	{ .id = 1, .rule_class = ARBITER_RULE_INFERENCE, .condition_start = 1, .condition_count = 2, .action_start = 0, .action_count = 0, .expr_start = 2, .expr_count = 1, .safety_goal_id = UINT16_MAX, .set_mode = UINT16_MAX, .safety_critical = false, .name = "02_cors.allow_origin", .explanation = "Same-origin or allowed origin — CORS pass." },
+	{ .id = 2, .rule_class = ARBITER_RULE_CONSTRAINT, .condition_start = 3, .condition_count = 1, .action_start = 0, .action_count = 0, .expr_start = 3, .expr_count = 1, .safety_goal_id = UINT16_MAX, .set_mode = 2, .safety_critical = false, .name = "03_cors.block_origin", .explanation = "Blocked origin — 403." },
+	{ .id = 3, .rule_class = ARBITER_RULE_CONSTRAINT, .condition_start = 4, .condition_count = 1, .action_start = 0, .action_count = 1, .expr_start = 4, .expr_count = 1, .safety_goal_id = UINT16_MAX, .set_mode = 3, .safety_critical = false, .name = "10_rate.check", .explanation = "Client > 60 req/min — 429 Too Many Requests." },
+	{ .id = 4, .rule_class = ARBITER_RULE_INFERENCE, .condition_start = 5, .condition_count = 1, .action_start = 0, .action_count = 0, .expr_start = 5, .expr_count = 1, .safety_goal_id = UINT16_MAX, .set_mode = 0, .safety_critical = false, .name = "20_auth.public_path", .explanation = "Public path — no auth required." },
+	{ .id = 5, .rule_class = ARBITER_RULE_INFERENCE, .condition_start = 6, .condition_count = 2, .action_start = 0, .action_count = 0, .expr_start = 6, .expr_count = 1, .safety_goal_id = UINT16_MAX, .set_mode = 0, .safety_critical = false, .name = "21_auth.health_check", .explanation = "Health check GET — always allowed." },
+	{ .id = 6, .rule_class = ARBITER_RULE_CONSTRAINT, .condition_start = 8, .condition_count = 2, .action_start = 0, .action_count = 0, .expr_start = 7, .expr_count = 1, .safety_goal_id = UINT16_MAX, .set_mode = 2, .safety_critical = false, .name = "22_auth.no_token", .explanation = "Auth-required path but no token — 401." },
+	{ .id = 7, .rule_class = ARBITER_RULE_CONSTRAINT, .condition_start = 10, .condition_count = 2, .action_start = 0, .action_count = 0, .expr_start = 8, .expr_count = 1, .safety_goal_id = UINT16_MAX, .set_mode = 2, .safety_critical = false, .name = "23_auth.invalid_token", .explanation = "Invalid/expired token — 401." },
+	{ .id = 8, .rule_class = ARBITER_RULE_CONSTRAINT, .condition_start = 12, .condition_count = 2, .action_start = 0, .action_count = 0, .expr_start = 9, .expr_count = 1, .safety_goal_id = UINT16_MAX, .set_mode = 2, .safety_critical = false, .name = "24_auth.admin_check", .explanation = "Admin path requires admin role — 403." },
+	{ .id = 9, .rule_class = ARBITER_RULE_INFERENCE, .condition_start = 14, .condition_count = 3, .action_start = 0, .action_count = 0, .expr_start = 10, .expr_count = 1, .safety_goal_id = UINT16_MAX, .set_mode = 0, .safety_critical = false, .name = "25_auth.valid_token", .explanation = "Valid token, authorized — allow." },
+	{ .id = 10, .rule_class = ARBITER_RULE_INFERENCE, .condition_start = 17, .condition_count = 1, .action_start = 0, .action_count = 0, .expr_start = 11, .expr_count = 2, .safety_goal_id = UINT16_MAX, .set_mode = UINT16_MAX, .safety_critical = false, .name = "30_risk.compute", .explanation = NULL },
+	{ .id = 11, .rule_class = ARBITER_RULE_CONSTRAINT, .condition_start = 18, .condition_count = 1, .action_start = 0, .action_count = 0, .expr_start = 13, .expr_count = 1, .safety_goal_id = UINT16_MAX, .set_mode = 2, .safety_critical = false, .name = "40_payload.too_large", .explanation = "Payload > 64KB — rejected." },
+};
+
+static const struct ARBITER_expr_def model_expressions[] = {
+	{ .target_fact_id = 3, .op = ARBITER_EXPR_ASSIGN, .left_fact_id = 65535, .left_literal = 1, .right_fact_id = 65535, .right_literal = 0, .scale = 1 },
+	{ .target_fact_id = 2, .op = ARBITER_EXPR_ASSIGN, .left_fact_id = 65535, .left_literal = 0, .right_fact_id = 65535, .right_literal = 0, .scale = 1 },
+	{ .target_fact_id = 2, .op = ARBITER_EXPR_ASSIGN, .left_fact_id = 65535, .left_literal = 1, .right_fact_id = 65535, .right_literal = 0, .scale = 1 },
+	{ .target_fact_id = 0, .op = ARBITER_EXPR_ASSIGN, .left_fact_id = 65535, .left_literal = 3, .right_fact_id = 65535, .right_literal = 0, .scale = 1 },
+	{ .target_fact_id = 0, .op = ARBITER_EXPR_ASSIGN, .left_fact_id = 65535, .left_literal = 4, .right_fact_id = 65535, .right_literal = 0, .scale = 1 },
+	{ .target_fact_id = 0, .op = ARBITER_EXPR_ASSIGN, .left_fact_id = 65535, .left_literal = 1, .right_fact_id = 65535, .right_literal = 0, .scale = 1 },
+	{ .target_fact_id = 0, .op = ARBITER_EXPR_ASSIGN, .left_fact_id = 65535, .left_literal = 1, .right_fact_id = 65535, .right_literal = 0, .scale = 1 },
+	{ .target_fact_id = 0, .op = ARBITER_EXPR_ASSIGN, .left_fact_id = 65535, .left_literal = 2, .right_fact_id = 65535, .right_literal = 0, .scale = 1 },
+	{ .target_fact_id = 0, .op = ARBITER_EXPR_ASSIGN, .left_fact_id = 65535, .left_literal = 2, .right_fact_id = 65535, .right_literal = 0, .scale = 1 },
+	{ .target_fact_id = 0, .op = ARBITER_EXPR_ASSIGN, .left_fact_id = 65535, .left_literal = 3, .right_fact_id = 65535, .right_literal = 0, .scale = 1 },
+	{ .target_fact_id = 0, .op = ARBITER_EXPR_ASSIGN, .left_fact_id = 65535, .left_literal = 1, .right_fact_id = 65535, .right_literal = 0, .scale = 1 },
+	{ .target_fact_id = 1, .op = ARBITER_EXPR_SCALE, .left_fact_id = 4, .left_literal = 0, .right_fact_id = 65535, .right_literal = 100, .scale = 120 },
+	{ .target_fact_id = 1, .op = ARBITER_EXPR_CLAMP, .left_fact_id = 1, .left_literal = 0, .right_fact_id = 65535, .right_literal = 0, .scale = 100 },
+	{ .target_fact_id = 0, .op = ARBITER_EXPR_ASSIGN, .left_fact_id = 65535, .left_literal = 3, .right_fact_id = 65535, .right_literal = 0, .scale = 1 },
 };
 
 static const char *model_mode_names[] = {
@@ -75,10 +92,12 @@ const struct ARBITER_model ARBITER_generated_model = {
 	.rule_count = 12,
 	.condition_count = 19,
 	.action_count = 1,
+	.expr_count = 14,
 	.mode_count = 4,
 	.facts = model_facts,
 	.rules = model_rules,
 	.conditions = model_conditions,
 	.actions = model_actions,
+	.expressions = model_expressions,
 	.mode_names = model_mode_names,
 };