fix ltl semantics and unify action enumeration (#71)

* fix(ltl): give every thunk a construction identity

Two distinct unnamed predicates both described as "Thunk(...)", so obligation
collapse merged their residuals and could drop a live violation. Identity is
assigned at construction and the fields are unexported, so a thunk cannot be
built without one.

Claude-Session: https://claude.ai/code/session_01Fj4wJUikdABuMQEETwW55J

* fix(ltl): reduce a thrown-predicate residual instead of panicking

The verifier substitutes an ErrorFormula for the residual of a property whose
predicate threw, and that residual is fed back in on the next step. reduce had
no case for it, so the run crashed. It re-reports the same failure now.

Claude-Session: https://claude.ai/code/session_01Fj4wJUikdABuMQEETwW55J

* fix(ltl): make a bounded always the dual of a bounded eventually

G<=n(f) and not F<=n(not f) disagreed on traces where the inner was still
pending when the window closed, so nnf's negation normal form was not semantics
preserving. Both sides now range over the observations at which their inner can
definitely resolve: the eventually keeps a pending inner as a disjunct, and the
always discharges vacuously at window close.

Claude-Session: https://claude.ai/code/session_01Fj4wJUikdABuMQEETwW55J

* fix(ltl): arm a one-shot root once per run

A root that carries its own horizon is one obligation for the whole run, not one
per observation. Re-instantiating a top-level eventually monitored G F<=n(p)
instead of F<=n(p) and left one live obligation per step behind; a bounded
always restarted its window every step and never closed.

Claude-Session: https://claude.ai/code/session_01Fj4wJUikdABuMQEETwW55J

* fix(verifier): stop wrapping a top-level eventually in always

`eventually(p).within(300, "seconds")` as a property meant "within 300 seconds
of every step", which spawned an obligation per step with its own resolved
deadline. A 553-step run carried 553 of them and serialized a 75 KB residual.

Claude-Session: https://claude.ai/code/session_01Fj4wJUikdABuMQEETwW55J

* fix(ltl): serialize the resolved deadline of a bounded window

Two obligations spawned at different steps from one duration-bounded formula
differ only in the deadline the evaluator resolved for them, so they serialized
identically and the trace erased a distinction the evaluator makes. The authored
window stays in amount/unit; the resolved deadline rides alongside.

Claude-Session: https://claude.ai/code/session_01Fj4wJUikdABuMQEETwW55J

* fix(verifier): split a witness's origin step from its detection step

A deferred obligation spans two steps: the one that armed it and the one whose
reduction failed. They were conflated under one index, so the extractor snapshot
(which is the detecting step's state) was reported against the origin step.

Claude-Session: https://claude.ai/code/session_01Fj4wJUikdABuMQEETwW55J

* fix(runner): record a witness's detection step in the trace

Claude-Session: https://claude.ai/code/session_01Fj4wJUikdABuMQEETwW55J

* feat(replay-ui): show the step a violation was detected at

The witness evidence is the detecting step's state, so say which step that is
and let a reader jump to it.

Claude-Session: https://claude.ai/code/session_01Fj4wJUikdABuMQEETwW55J

* fix(verifier): record the extractor state the predicates actually read

On the web path extractor bodies are evaluated in V8 and injected here, but only
the goja value was replaced. The trace diff and the violation witness therefore
described a state no property ever saw.

Claude-Session: https://claude.ai/code/session_01Fj4wJUikdABuMQEETwW55J

* refactor(spec): one candidate producer over one target-eligibility rule

Both hosts routed verbs themselves and both policies enumerated their own
actions, and all four drifted. Web sent `swipes` to scrollable containers only,
so swipe-to-dismiss on a list row was reachable on native and unreachable on
web; the model policy folded gestures its own way and could not reach what the
seeded picker drew.

A host now reports facts about every element and never decides which verb may
act on it: targets.ts acceptsTarget owns that for both. pick.ts builtinCandidates
is the single enumeration, and the model policy reads it through
__sanderlingEnumerateBuiltin__ instead of reimplementing it in Go.

Gesture verbs change with it: scrolls stay vertical over scrollable containers,
swipes go free-form in all four directions from any element with real bounds.

Claude-Session: https://claude.ai/code/session_01Fj4wJUikdABuMQEETwW55J

* fix(runner): name a builtin scroll by its drag origin

A builtin gesture carries endpoints and no selector, so every scroll rendered as
"Scroll down " in the prompt's recent-action memory and two scrollable regions
were indistinguishable.

Claude-Session: https://claude.ai/code/session_01Fj4wJUikdABuMQEETwW55J

* fix(chrome): clear storage over cdp instead of scripting an opaque origin

Launch runs while the tab is still on about:blank, whose opaque origin denies
storage access, so localStorage.clear() threw SecurityError and every web run
died at launch. Storage.clearDataForOrigin needs no navigation. The exception
helper lands here because "Uncaught" is what hid this for so long.

Claude-Session: https://claude.ai/code/session_01Fj4wJUikdABuMQEETwW55J

* fix(chrome): enable the swiftshader webgl fallback

Headless Chrome runs with --disable-gpu, and without this flag it refuses the
software WebGL backend: getContext returns null, so a canvas-rendered app paints
nothing and every screenshot is identical black.

Claude-Session: https://claude.ai/code/session_01Fj4wJUikdABuMQEETwW55J

* fix(web): resolve testTag through data-testid or id

Compose Multiplatform emits its testTag into the element id, which the native
table already accepts via the resource-id alias. The two web selector tables
were the only place that rejected it.

Claude-Session: https://claude.ai/code/session_01Fj4wJUikdABuMQEETwW55J

* test(spec): type-check the spec api as part of make test

The fake runtime in api.test.ts did not return a chainable handle from extract,
so the file had not type-checked since named() was added. Wiring the check into
make test stops it drifting again.

Claude-Session: https://claude.ai/code/session_01Fj4wJUikdABuMQEETwW55J

* docs(manual): one-shot eventually and the gesture verbs

Claude-Session: https://claude.ai/code/session_01Fj4wJUikdABuMQEETwW55J

internal/verifier/policy_parity_test.go

@@ -0,0 +1,161 @@
+package verifier
+
+import (
+	"errors"
+	"fmt"
+	"maps"
+	"slices"
+	"testing"
+)
+
+// policyTreeJSON gives every builtin verb something to act on, with every label
+// distinct so no two candidates render the same way.
+const policyTreeJSON = `{
+  "attributes": {"bounds": "[0,0,400,800]"},
+  "children": [
+    {"attributes": {"resource-id": "Save", "text": "Save", "bounds": "[0,0,200,60]"}, "clickable": true, "enabled": true, "children": []},
+    {"attributes": {"resource-id": "Cancel", "text": "Cancel", "bounds": "[200,0,400,60]"}, "clickable": true, "enabled": true, "children": []},
+    {"attributes": {"resource-id": "Amount", "class": "EditText", "hintText": "Amount", "bounds": "[0,100,400,160]"}, "enabled": true, "children": []},
+    {"attributes": {"resource-id": "Note", "class": "EditText", "hintText": "Note", "bounds": "[0,200,400,260]"}, "enabled": true, "children": []},
+    {"attributes": {"resource-id": "List", "scrollable": "true", "bounds": "[0,300,400,700]"}, "children": []}
+  ]
+}`
+
+// policyVerbs is every builtin verb a spec can put in its action tree.
+var policyVerbs = []string{
+	"taps", "doubleTaps", "longPresses", "typing", "scrolls", "swipes", "pressKeys", "waitOnce",
+}
+
+// seededDrawBudget is how many times the seeded picker is driven per verb. The
+// candidate sets here are a handful of entries wide, so this exhausts them many
+// times over; a miss would mean the picker cannot reach one of its own
+// candidates, which is itself the bug worth failing on.
+const seededDrawBudget = 300
+
+// TestPoliciesEnumerateTheSameActions is the guard on the claim the paper makes
+// about the two action policies: they differ in the pick and in nothing else.
+// For every builtin verb, the actions the seeded picker can draw and the
+// candidates the model policy is offered must be the same set. Enumeration lives
+// in one place (pick.ts builtinCandidates) precisely so this cannot drift, and
+// this test is what notices if a second one ever grows back.
+func TestPoliciesEnumerateTheSameActions(t *testing.T) {
+	for _, verb := range policyVerbs {
+		t.Run(verb, func(t *testing.T) {
+			seeded := seededReachableActions(t, verb)
+			model := modelOfferedActions(t, verb)
+			if len(model) == 0 {
+				t.Fatalf("%s offered the model no candidates at all", verb)
+			}
+			if !slices.Equal(slices.Sorted(maps.Keys(seeded)), slices.Sorted(maps.Keys(model))) {
+				t.Errorf("action spaces differ for %s\n seeded=%v\n  model=%v",
+					verb, slices.Sorted(maps.Keys(seeded)), slices.Sorted(maps.Keys(model)))
+			}
+		})
+	}
+}
+
+// TestModelCandidateDescriptionsAreUniqueAndNamed checks the rendering half of
+// the contract: every enumerated action reaches the model as its own distinctly
+// named line, so the number it picks and the action that executes agree.
+func TestModelCandidateDescriptionsAreUniqueAndNamed(t *testing.T) {
+	for _, verb := range policyVerbs {
+		t.Run(verb, func(t *testing.T) {
+			verifier := loadVerbSpec(t, verb)
+			seen := map[string]bool{}
+			for _, candidate := range verifier.Candidates() {
+				if candidate.Description == "" {
+					t.Fatalf("%s produced a candidate with no description: %+v", verb, candidate.Action)
+				}
+				if seen[candidate.Description] {
+					t.Errorf("%s rendered %q twice, so the model cannot address both",
+						verb, candidate.Description)
+				}
+				seen[candidate.Description] = true
+			}
+		})
+	}
+}
+
+// TestModelIsOfferedTheUntargetedVerbs pins the two verbs the model arm used to
+// be blind to: with no element to enumerate over, a key press and a wait were
+// dropped, so the model could never navigate back or let the app settle.
+func TestModelIsOfferedTheUntargetedVerbs(t *testing.T) {
+	verifier := loadVerbSpec(t, "pressKeys")
+	if !hasCandidate(verifier.Candidates(), "Press back") {
+		t.Errorf("pressKeys missing from the model's candidates: %v",
+			descriptions(verifier.Candidates()))
+	}
+	verifier = loadVerbSpec(t, "waitOnce")
+	if !hasCandidate(verifier.Candidates(), "Wait") {
+		t.Errorf("waitOnce missing from the model's candidates: %v",
+			descriptions(verifier.Candidates()))
+	}
+}
+
+// loadVerbSpec builds a verifier whose whole action tree is one builtin verb,
+// with policyTreeJSON pushed as the current state.
+func loadVerbSpec(t *testing.T, verb string) *Verifier {
+	t.Helper()
+	verifier := newVerifier(t, WithSeed(0x5eed))
+	loadActionSpec(t, verifier, fmt.Sprintf(
+		"import { %s } from \"@sanderling/spec\";\nglobalThis.actions = %s;", verb, verb))
+	pushTree(t, verifier, policyTreeJSON)
+	return verifier
+}
+
+// seededReachableActions drives the seeded picker over its draw budget and
+// returns every distinct action it produced.
+func seededReachableActions(t *testing.T, verb string) map[string]Action {
+	t.Helper()
+	verifier := loadVerbSpec(t, verb)
+	reachable := map[string]Action{}
+	for range seededDrawBudget {
+		action, err := verifier.NextAction()
+		if errors.Is(err, ErrNoAction) {
+			continue
+		}
+		if err != nil {
+			t.Fatalf("%s next action: %v", verb, err)
+		}
+		reachable[actionIdentity(action)] = action
+	}
+	return reachable
+}
+
+// modelOfferedActions returns the actions behind the numbered list the model
+// policy picks from.
+func modelOfferedActions(t *testing.T, verb string) map[string]Action {
+	t.Helper()
+	verifier := loadVerbSpec(t, verb)
+	offered := map[string]Action{}
+	for _, candidate := range verifier.Candidates() {
+		offered[actionIdentity(candidate.Action)] = candidate.Action
+	}
+	return offered
+}
+
+// actionIdentity keys an action by everything except the values the policy owns
+// rather than the candidate set: the typed text, which the seeded arm draws from
+// the edge-case corpus and the model writes itself, and a swipe's drag distance,
+// which the seeded arm draws and the enumeration lists at a nominal length.
+// Comparing those would compare policies instead of action spaces. A swipe's
+// direction is NOT policy-owned, so it survives as the sign of the drag.
+func actionIdentity(action Action) string {
+	action.Text = ""
+	if action.Kind == ActionKindSwipe {
+		action.ToX = sign(action.ToX - action.FromX)
+		action.ToY = sign(action.ToY - action.FromY)
+	}
+	return fmt.Sprintf("%+v", action)
+}
+
+func sign(value int) int {
+	switch {
+	case value > 0:
+		return 1
+	case value < 0:
+		return -1
+	default:
+		return 0
+	}
+}