ci: dispatch workflows for folio and the replay ui (#73)

* feat(runner): stop the step loop at the first violation on request

* feat(testrun): report violations as a typed error under exit-on-violation

* feat(cli): add --exit-on-violation and exit 2 when it fires

* docs(cli): document --exit-on-violation, --max-steps, and exit codes

* fix(web): enumerate and query across shadow roots in both producers

* test(chrome): compare both producers on a shadow-dom parity page

* test(browser): drive a canvas-under-shadow-root fixture end to end

* fix(web): select the focused field inside a shadow root before typing

* fix(web): report the pathname as the screen when there is no hash route

* fix(web): settle on dom quiescence instead of returning at body ready

* feat(replay-ui): add data-testid hooks the dogfood spec drives

* feat(replay-ui): add the dogfood spec sanderling runs against the replay ui

* fix(replay-ui): scope the screenshot property to the named state panel

* chore(make): add per-platform sanderling build targets

* ci: add dispatch workflows for folio and the replay ui

* docs: describe the dispatch workflows and how to read a failure

* ci(folio): give the ios leg its jdk, android sdk, just, and a clean app start

* refactor(web): use max for the settle budget

* ci: pin calibrated seeds, skip the flaky ios reinstall, bound every job

* ci: authenticate and pin the buf setup step

the anonymous release download hit the shared runner ip rate limit and
failed the job with 'socket hang up' after three retries.

* docs: record that canvas apps need a dom proxy to be text-fuzzable

* fix(ios): bound lifecycle rpcs and claim the target device

a launch the simulator rejects sent the xctest session into a recovery
chain that answered minutes late or never, and the rpc had no deadline,
so the run hung with no trace and no error. also take a per-udid flock:
a second run's reinstall lands under the first's live automation session
and wedges it.

* docs(ci): correct the ios hang wording and note the device lock

* refactor(verifier): derive the lastAction shape from one field list

both hosts must show a spec the same lastAction. one ordered list now
feeds the goja object and the json the web host installs, so they
cannot drift.

* fix(web): install lastAction in the page before extractors read it

state.lastAction was hardcoded null on web, so every property reading it
was silently vacuous: a correct property passed without ever firing.

* fix(web): carry element identity on actions and fix findAll on paths

an action's target was coordinates only, so a property matching on which
element was acted upon could never fire. ax.findAll([a,b]) also returned
nothing on web.

* fix(chrome): wait out a route transition before sampling facts

the tree stays byte-identical and quiet across a cross-fade, so both the
quiet timer and the unchanged-tree escape called it settled mid-flight
and extractors read two screens at once.

* fix: bound the pre-run app launch

launch happens before the runner starts, so --duration never covered it
and a wedged driver hung with no trace and no error.

* fix(folio): read balances from merged cards and treat unreadable as unknown

compose for web merges the whole accountcard subtree, so the balance
child never exists there and every card parsed as 0. the property then
compared 0 to 0 and fired on any submit, which is a false positive
generator. unknown is now null and null is vacuously true.

* test(folio): cover merged-card parsing and unknown balances

* ci(folio): make web an expect-the-bug leg

the web runtime can observe the double submit now, so the health gate
understates it. seed 1 finds it at step 109, 3 runs out of 3.

* docs(ci): explain why a submit tap landing on home is the bug

* fix(ios): read a StaticText's label as its text

AXValue was the only source for text, but a StaticText carries its
string in AXLabel, so nothing on screen had .text on ios: a spec reading
it saw everything on android and nothing here.

* docs(ci): correct the calibrated step ranges

* fix(folio): stop convicting on arithmetic float64 cannot hold

past 2^53 cents the gap between representable values is 128, so a real
1600-cent move reads back as something else and the equality is false
for a healthy submit as readily as a double one. also match parseCents:
a sign or an oversized amount is rejected, not read as an amount.

* test(folio): pin the safe-integer guard and its boundary

* docs: stop teaching the zero-default that caused a false alarm

* docs: write down the silent-vacuity failure modes

* feat(folio): tag the home total and the card transaction count

the total was the only untagged node on the screen, so the spec had to
sum cards and a clipped card broke the sum.

* fix(folio): read the app's own total and refuse contaminated windows

summing cards went null when one was clipped, and the null poisoned the
carrier for the rest of the run. the balance window also spanned every
transaction since the last home visit, so the property convicted on
deltas it could not attribute: the old web witness was 3.16x the typed
amount, not 2x.

* test(folio): pin the window rules and the count invariant

* fix(folio): never read a frame that shows two screens

android dumps a cross-fade with both screens in the tree. the route said
add-transaction while an unscoped find said home, so the oracle took a
half-rendered total as fresh and convicted on a tap that committed
nothing. one function now decides the route and returns null when the
frame is ambiguous.

* test(folio): cover transition frames, card readings and creation

* fix(folio): only disambiguate counts that came from merged text

the equal-length digit rule exists because web merges the card and an
account named -1 makes '12' ambiguous. a dedicated count node has
nothing to disambiguate, so applying it there threw away real evidence.

* ci(folio): pin the recalibrated seeds and drop android to a health gate

web 3 and ios 7 convict 3 runs out of 3 with an exactly 2x witness.
android convicts 2 in 5 because the same seed does not walk the same
trajectory there, so it proves the app runs instead.

* docs(ci): describe the two properties and why android cannot convict

* fix(android): wait out a route cross-fade before snapshotting

the dump could hold two screens at once, and the runner refuses to act
on such a tree, so a quarter of android steps applied no action and the
count varied per run: the same seed never walked the same trajectory.
the ios companion and the chrome driver already do this.

* ci(folio): let the android leg run far enough to see its conviction

* docs: only the repo owner merges

* ci(folio): a thrown predicate is not a conviction

exit 2 means the run recorded a violation, and a predicate that throws
is recorded as one too. so was newAccountBalanceIsZero, an unrelated
property in the same spec. the gate read the exit code and went green
with detection dead.

* ci: install idb-companion from its tap and stop interpolating inputs

idb-companion is not in homebrew-core, so the ios leg died before it
built anything. replay-ui expanded dispatch inputs into the shell.

* docs: correct the snippets and numbers that drifted from the code

* test(sidecar): pin that a slow read counts toward the stability streak

* fix(web): read the page's extractors only on steps that count

the page advances the spec's carriers when it evaluates, but the runner
applied the result only on non-transitional steps. a discarded step
moved the window forward anyway, so the next accepted pair bracketed two
transactions while counting one submit, and convicted a healthy app.
extractor errors now fail the run instead of leaving goja's values in
current against v8's in previous.

* fix(chrome): anchor the transition deadline when the dom goes quiet

it was anchored at script start, so a page that churned past the window
reached the check already expired and returned mid cross-fade. the
driver now publishes the idle timeout it needs, since the caller's 1s
could never spend the 800ms window.

* fix(web): fail on a partial extractor override

same mixed-producer hazard as the install error: some extractors hold
the page's value and the rest hold goja's, and a property comparing
across that split fires on a healthy app.

* docs: six of seven, the seventh is the stock property

* fix(folio): drop a name two cards answer to

homeTxnCountsOf keyed on the account name and let the last card win, so
two accounts the fuzzer named the same collapsed into one entry. a
reading that saw one Travel card and a later one that saw both then
subtracted two different accounts' counts, and
submitCommitsOneTransactionPerAction convicted a healthy app of
double-submitting. it is a gated property in folio-run.sh, so that reads
as "found the submit bug" over a card scrolling into view.

same rule createdAccountHasNonZeroBalance already applies: a name
nothing can attribute is no evidence. counted over every card, since an
unreadable twin spoils the identity too.

* perf(folio): read each frame once

every extractor asked routeOf, and routeOf does five ax.find calls. on
web each find walks the document and every shadow root beneath it, so
the spec cost 110 tree walks a step; homeCards was parsed four times
over. now 5 and once.

keyed on the identity of the state object because both hosts build a new
one per step and hand that one object to every getter, so it cannot
outlive its frame. holding the reference is what keeps that true rather
than likely.

* fix(web): keep an undefined reading's index through JSON

json has no undefined, so an extractor whose getter returned one had its
whole index dropped by JSON.stringify. that index then kept goja's
dump-derived value while its neighbours held the page's, and a property
comparing previous to current across the split fires on a healthy app.
folio has nine on(route, tag) extractors, so this was most extractors on
most steps.

each reading is wrapped in a {value} envelope: the drop now happens
inside the entry, and an absent value means the getter returned
undefined, which is what the goja host records for the same getter. a
json null would instead claim it returned null and x.current ===
undefined would answer differently on the two hosts.

* feat(verifier): report the registered extractor count

the web path needs it to check the page sent one reading per extractor.

* fix(runner): fail when the page reports fewer readings than extractors

the comment here already claimed a partial override was fatal. it was
not: the skipped check only catches indices outside the extractor list,
so a page reporting values for some extractors and not others left the
rest holding goja's reading of the dump with nothing said.

* test(browser): drive an undefined reading through the whole web path

four layers carry it: the page's envelope, the driver's unwrap, the
runner's count check and the verifier's decode. each has a unit test and
only a run proves they compose. goes red both ways, decoding an absent
value as null and dropping the envelope.

* fix(web): offer the aria roles a user activates

only role=button was in the tappable set, so link, checkbox, radio,
switch, tab, option, the menuitems and treeitem were invisible to the
enumeration however plain the control looked. the replay ui builds its
step rows as <li role="option">, and the spec dogfooding it had to
hand-write an action to reach them because no default verb could see a
single row.

both producers build the set from the same role list, since the parity
test compares them element by element.

* test(browser): tap a role-based control end to end

every control on the page is an <li role="option">, the shape the
replay ui gives its step rows, and the spec carries no action of its
own: the property firing is the evidence the default enumeration offered
a tap on one.

* fix(web): read aria-disabled as disabled

the enabled fact came off the disabled property, which only real form
controls have. it reads undefined on the role-based controls the
tappable set now covers, so every one of them looked enabled however
plainly it was marked otherwise, and the fuzzer would spend actions on
inert ones.

both producers answer the same two ways, and the parity fixture carries
a disabled row so the comparison covers it: reverting one side alone
names the element and the fact.

* docs(replay-ui): the enumeration reaches step rows now

the comment said role="option" is not in the tappable selector set,
which stopped being true a few commits ago. selectAStep stays, for the
reason the tab weight below it stays: one row among the page's clickable
elements is a thin chance, and both step-facing properties go vacuous on
a run that never selects one.

* test(runner): bound the last-action test by steps, not wall clock

100ms of wall clock against an assertion that two steps ran fatals under
load with "the web path never installed it", which reads as a
regression. every sibling test in the package uses a long duration and
MaxSteps.

* ci: run the kotlin tests in make test

RouteTransitionTest and the stability poll cover the android settle and
nothing in ci ran them. :sidecar:test needs no android sdk, checked by
running it with ANDROID_HOME pointed at nothing.

* fix(sidecar): measure the stability streak as observed quiet

parameterising pollUntilStable also moved the clock to the start of the
read that opened a run of identical snapshots, so a read's own duration
counted as quiet. the pre-existing caller polls a real uiautomator dump:
at 400ms a read, 750ms of required quiet became 250ms of observed quiet
and the poll settled in two reads instead of four.

the parameters stay, the semantics go back.

* test(sidecar): pin the transition cap by driving it

it asserted 1500 >= 700 + 300, two constants, which can only fail if
someone edits a constant. it now drives awaitSettledTree against a fade
that lands after 700ms and asserts it hands back the settled tree before
the cap. cut the cap to 1000 and it goes red.

* ci: pin buf-setup-action to a commit

it takes a token now, so a floating tag is a token handed to whatever
that tag moves to. note v1 there is a branch, not a tag, so the ref
lookup that resolves it is matching-refs/heads/v1.

* ci: declare least-privilege permissions

none of the three declared any, so each got the repository default.
release.yml and docs.yml already do this. all three only check out,
build, test and upload artifacts.

* ci: fail fast when a server never comes up

the readiness loops fell through silently after 30 tries, so a server
that never started surfaced as an opaque driver failure minutes later.
each now says what did not answer and on which port.

* ci(folio): a missing trace is not a verdict

with no trace the android gate ran its grep against ./trace.jsonl and
reported "never reached AddTransactionScreen, so it never got past
login", which is not what happened. the web and ios branches had the
same misdiagnosis on exit 0.

same class, one line up: the classifier's own failure was swallowed, so
with the evidence reader dead the gate printed a healthy run and exited
0.

* ci(replay-ui): skip a run directory with no trace

the summarise step is if: always(), and under github's bash -eo pipefail
an unmatched glob stays literal, the redirect fails, pipefail carries it
into the assignment and -e kills the step. so a failed fuzz run went red
twice, once for the real reason.

internal/runner/runner.go

@@ -31,6 +31,11 @@ type Options struct {
 	// positive value stops the loop once that many steps have run.
 	MaxSteps int
 
+	// StopOnViolation ends the step loop as soon as a step records a
+	// violation, so a run that exists to find one bug stops at the evidence
+	// instead of spending the rest of its budget past it.
+	StopOnViolation bool
+
 	BundleID    string
 	Driver      driver.DeviceDriver
 	Verifier    *verifier.Verifier
@@ -69,6 +74,7 @@ func Run(ctx context.Context, options Options) (Summary, error) {
 	if logger == nil {
 		logger = slog.Default()
 	}
+	options.IdleTimeout = resolveIdleTimeout(options)
 
 	// Gate on the app actually being on top before acting, so the first
 	// action never fires against a leftover screen or a system dialog. Done
@@ -82,6 +88,7 @@ func Run(ctx context.Context, options Options) (Summary, error) {
 	if err != nil {
 		return Summary{}, err
 	}
+	_, pageExtractors := extractorSource.(webSource)
 
 	summary := Summary{StartTime: time.Now()}
 	deadline := summary.StartTime.Add(options.Duration)
@@ -117,9 +124,8 @@ func Run(ctx context.Context, options Options) (Summary, error) {
 		var logs []verifier.LogEntry
 
 		// gctx is bound to the errgroup so a returned error (or outer
-		// cancellation) propagates to siblings - notably the V8 extractor
-		// goroutine, whose CDP round-trip can otherwise outrun the step
-		// budget on a hung tab.
+		// cancellation) propagates to every sibling read rather than leaving
+		// one blocked on a hung device.
 		g, gctx := errgroup.WithContext(ctx)
 		si := stepIndex
 		// fetchSyncedState issues a single Snapshot RPC so hierarchy and
@@ -138,16 +144,6 @@ func Run(ctx context.Context, options Options) (Summary, error) {
 			logs = collectLogs(gctx, options.Driver, logSince)
 			return nil
 		})
-		var v8Overrides map[int]json.RawMessage
-		g.Go(func() error {
-			overrides, err := extractorSource.ExtractorOverrides(gctx)
-			if err != nil {
-				logger.Warn("v8 extractor evaluation failed", "step", si, "err", err)
-				return nil
-			}
-			v8Overrides = overrides
-			return nil
-		})
 		// All goroutines write to local variables and return nil, so the Wait
 		// error is always nil; ignored intentionally.
 		_ = g.Wait()
@@ -190,6 +186,29 @@ func Run(ctx context.Context, options Options) (Summary, error) {
 		var witnesses map[string]trace.Witness
 		skippedVerification := false
 		if !transitional {
+			// The page-side extractors evaluate only on steps the verifier will
+			// accept, which is why this read waits for the tree instead of
+			// racing it. A spec's extractor getters carry state across steps
+			// (folio's last-seen Home total, its submit counters) and that state
+			// advances every time they run: evaluating them on a step whose
+			// values are then thrown away leaves the page one window ahead of
+			// the verifier, so the next accepted pair brackets two committed
+			// transactions while having counted one submit, and the property
+			// convicts a healthy app. It costs the latency the read used to hide
+			// behind the hierarchy fetch; the fetch is what decides whether this
+			// step counts at all, so it has to go first.
+			//
+			// lastAction is the same value PushSnapshot hands the goja state
+			// below: the two engines evaluate this step against one action.
+			v8Overrides, overridesErr := extractorSource.ExtractorOverrides(ctx, lastAction)
+			if overridesErr != nil {
+				// Not a warning. Without the page's values this step's
+				// extractors keep goja's dump-derived readings while the
+				// previous step holds the page's, and a delta property then
+				// compares two producers and fires on an app that did nothing
+				// wrong.
+				return summary, fmt.Errorf("step %d extractor overrides: %w", stepIndex, overridesErr)
+			}
 			if err := options.Verifier.PushSnapshot(verifier.SnapshotInput{
 				Tree:          tree,
 				ScreenshotPNG: screenshotPNG,
@@ -201,13 +220,26 @@ func Run(ctx context.Context, options Options) (Summary, error) {
 			}); err != nil {
 				return summary, fmt.Errorf("step %d push: %w", stepIndex, err)
 			}
+			// Every failure below leaves some extractors holding the page's
+			// value and the rest holding goja's reading of the dump, and a
+			// property comparing previous to current across that split fires
+			// on a healthy app. Each also means the two engines loaded
+			// different bundles, which nothing downstream can reconcile.
+			if pageExtractors && len(v8Overrides) != options.Verifier.ExtractorCount() {
+				return summary, fmt.Errorf(
+					"step %d: the page reported values for %d of the spec's %d extractors; "+
+						"the page and the host are running different bundles",
+					stepIndex, len(v8Overrides), options.Verifier.ExtractorCount())
+			}
 			skipped, overrideErr := options.Verifier.OverrideExtractorValues(v8Overrides)
 			if overrideErr != nil {
-				logger.Warn("v8 override apply failed", "step", stepIndex, "err", overrideErr)
+				return summary, fmt.Errorf("step %d apply extractor overrides: %w", stepIndex, overrideErr)
 			}
 			if skipped > 0 {
-				logger.Warn("v8 override skipped out-of-range entries",
-					"step", stepIndex, "skipped", skipped, "have", len(v8Overrides))
+				return summary, fmt.Errorf(
+					"step %d: %d of %d extractor overrides fell outside the spec's extractor list; "+
+						"the page and the host are running different bundles",
+					stepIndex, skipped, len(v8Overrides))
 			}
 			options.Verifier.EvaluateProperties()
 			violations = options.Verifier.NewlyViolatedProperties()
@@ -296,6 +328,12 @@ func Run(ctx context.Context, options Options) (Summary, error) {
 		summary.Steps = stepIndex
 		if len(violations) > 0 {
 			summary.Violations = append(summary.Violations, violationRecords(violations, witnesses, stepIndex)...)
+			// The step is already written, so the trace ends on the state that
+			// produced the violation. Finalize below still runs, so pending
+			// liveness obligations are reported alongside it.
+			if options.StopOnViolation {
+				break
+			}
 		}
 		// Wait actions are themselves a settling: skip the idle poll. Actions
 		// that mutate the UI fall through to WaitForIdle so the next step's
@@ -370,12 +408,36 @@ func validate(options Options) error {
 	if options.Duration <= 0 {
 		return errors.New("runner: Duration must be positive")
 	}
-	if options.IdleTimeout <= 0 {
-		options.IdleTimeout = 2 * time.Second
-	}
 	return nil
 }
 
+// defaultIdleTimeout is the settle budget a caller that names none gets.
+const defaultIdleTimeout = 2 * time.Second
+
+// idleTimeoutFloor is a driver that knows how long its own settle can take.
+// Declared here rather than in the driver package (like lastActionInstaller in
+// source.go) so the mobile drivers stay untouched.
+type idleTimeoutFloor interface {
+	MinIdleTimeout() time.Duration
+}
+
+// resolveIdleTimeout settles the per-step settle budget: the caller's value,
+// defaulted when unset, and raised to whatever the driver says its own settle
+// needs. The chrome driver's settle waits for the DOM to go quiet and only then
+// opens its route-transition window; handed less than their sum it is cut off
+// mid-transition, and the step samples the screen the app is leaving. A driver
+// that reports no floor keeps the caller's value exactly.
+func resolveIdleTimeout(options Options) time.Duration {
+	timeout := options.IdleTimeout
+	if timeout <= 0 {
+		timeout = defaultIdleTimeout
+	}
+	if floor, ok := options.Driver.(idleTimeoutFloor); ok {
+		timeout = max(timeout, floor.MinIdleTimeout())
+	}
+	return timeout
+}
+
 // ensureForeground keeps the app under test in the foreground. When the driver
 // can report the foreground app and it no longer matches the bundle under test,
 // the app is relaunched. Returns true when a relaunch happened so the caller

internal/runner/runner_test.go

@@ -1976,3 +1976,99 @@ func TestRunner_SkipsActionWhenOverlayStealsFocusAtApplyTime(t *testing.T) {
 		t.Error("apply-time guard failed: a tap fired while a system overlay held focus")
 	}
 }
+
+// TestRunner_StopOnViolationEndsAtTheFirstViolation pins the gate CI runs on:
+// a step budget of 8 against a spec that only violates on the third step must
+// end on step 3 and write nothing after it, so the trace's last state is the
+// one that produced the violation.
+func TestRunner_StopOnViolationEndsAtTheFirstViolation(t *testing.T) {
+	const thirdStepViolationSpec = `
+import { actions, always, extract } from "@sanderling/spec";
+let observed = 0;
+const tick = extract(() => ++observed);
+globalThis.properties = {
+  staysUnderThree: always(() => tick.current < 3),
+};
+globalThis.actions = actions(() => []);
+`
+	state := newHarnessWithSpec(t, thirdStepViolationSpec)
+
+	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
+	defer cancel()
+	summary, err := Run(ctx, Options{
+		Duration:        time.Hour,
+		IdleTimeout:     20 * time.Millisecond,
+		MaxSteps:        8,
+		StopOnViolation: true,
+		Driver:          state.mock,
+		Verifier:        state.verifier,
+		TraceWriter:     state.writer,
+	})
+	if err != nil {
+		t.Fatalf("Run: %v", err)
+	}
+	if !containsProperty(summary.Violations, "staysUnderThree") {
+		t.Fatalf("expected staysUnderThree to fire, got %v", summary.Violations)
+	}
+	if summary.Steps != 3 {
+		t.Errorf("steps: got %d, want 3 (the run must stop at the violating step, not run the 8-step budget)",
+			summary.Steps)
+	}
+	for _, step := range traceStepIndices(t, state.writer.Directory()) {
+		if step > summary.Steps {
+			t.Errorf("trace kept stepping after the violation: found step %d past step %d",
+				step, summary.Steps)
+		}
+	}
+}
+
+// TestRunner_WithoutStopOnViolationRunsTheWholeBudget is the other half: the
+// default must stay a full-budget fuzz run, so turning the flag on is the only
+// thing that shortens a run.
+func TestRunner_WithoutStopOnViolationRunsTheWholeBudget(t *testing.T) {
+	state := newHarnessWithSpec(t, violationSpec)
+
+	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
+	defer cancel()
+	summary, err := Run(ctx, Options{
+		Duration:    time.Hour,
+		IdleTimeout: 20 * time.Millisecond,
+		MaxSteps:    4,
+		Driver:      state.mock,
+		Verifier:    state.verifier,
+		TraceWriter: state.writer,
+	})
+	if err != nil {
+		t.Fatalf("Run: %v", err)
+	}
+	if summary.Steps != 4 {
+		t.Errorf("steps: got %d, want 4; a violation must not shorten a default run", summary.Steps)
+	}
+}
+
+// traceStepIndices reads every step index the trace recorded, so a test can
+// assert on what the run actually wrote rather than on the summary alone.
+func traceStepIndices(t *testing.T, directory string) []int {
+	t.Helper()
+	file, err := os.Open(filepath.Join(directory, "trace.jsonl"))
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer file.Close()
+	var steps []int
+	scanner := bufio.NewScanner(file)
+	scanner.Buffer(make([]byte, 0, 64*1024), 8*1024*1024)
+	for scanner.Scan() {
+		var line struct {
+			Step int `json:"step"`
+		}
+		if err := json.Unmarshal(scanner.Bytes(), &line); err != nil {
+			t.Fatalf("trace line decode: %v", err)
+		}
+		steps = append(steps, line.Step)
+	}
+	if err := scanner.Err(); err != nil {
+		t.Fatalf("scan trace: %v", err)
+	}
+	return steps
+}

internal/runner/source.go

@@ -23,8 +23,23 @@ type ActionSource interface {
 // ExtractorSource yields per-step extractor overrides the runner applies after
 // PushSnapshot. The mobile path has none (returns nil); the web path returns the
 // values its extractors computed in V8 against the real DOM.
+//
+// lastAction is the action the previous step actually applied, the same value
+// PushSnapshot hands the goja state. The web path has to install it in the page
+// before its extractors run: a spec extractor reading state.lastAction runs in
+// V8 there, and V8 has no way to know what the runner dispatched.
 type ExtractorSource interface {
-	ExtractorOverrides(ctx context.Context) (map[int]json.RawMessage, error)
+	ExtractorOverrides(
+		ctx context.Context,
+		lastAction *verifier.Action,
+	) (map[int]json.RawMessage, error)
+}
+
+// lastActionInstaller is the web driver's channel for the previous step's
+// action. It is declared here rather than folded into driver.WebDriver so the
+// mobile drivers stay untouched; every web driver must implement it.
+type lastActionInstaller interface {
+	SetLastAction(ctx context.Context, encoded json.RawMessage) error
 }
 
 // gojaSource drives both action selection and (trivially) extractor overrides
@@ -38,7 +53,10 @@ func (s gojaSource) NextAction(context.Context) (verifier.Action, error) {
 	return s.verifier.NextAction()
 }
 
-func (gojaSource) ExtractorOverrides(context.Context) (map[int]json.RawMessage, error) {
+func (gojaSource) ExtractorOverrides(
+	context.Context,
+	*verifier.Action,
+) (map[int]json.RawMessage, error) {
 	return nil, nil
 }
 
@@ -59,7 +77,24 @@ func (s webSource) NextAction(ctx context.Context) (verifier.Action, error) {
 	return verifier.DecodeAction(raw)
 }
 
-func (s webSource) ExtractorOverrides(ctx context.Context) (map[int]json.RawMessage, error) {
+// ExtractorOverrides installs the previous step's action in the page, then
+// reads back what the spec's extractors computed against the live DOM. The
+// install is not best-effort: a web driver that cannot take it leaves
+// state.lastAction null in V8, which silently turns every action-gated
+// property vacuously true, so it is reported as an error instead.
+func (s webSource) ExtractorOverrides(
+	ctx context.Context,
+	lastAction *verifier.Action,
+) (map[int]json.RawMessage, error) {
+	installer, ok := s.web.(lastActionInstaller)
+	if !ok {
+		return nil, fmt.Errorf(
+			"web driver %T cannot install state.lastAction; every property gated "+
+				"on the last action would be vacuously true", s.web)
+	}
+	if err := installer.SetLastAction(ctx, verifier.EncodeLastAction(lastAction)); err != nil {
+		return nil, fmt.Errorf("install last action: %w", err)
+	}
 	return s.web.EvaluateExtractors(ctx)
 }
 

internal/runner/web_carrier_test.go

@@ -0,0 +1,196 @@
+package runner
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"os"
+	"path/filepath"
+	"strconv"
+	"testing"
+	"time"
+
+	"github.com/priyanshujain/sanderling/internal/driver"
+	mockdriver "github.com/priyanshujain/sanderling/internal/driver/mock"
+	"github.com/priyanshujain/sanderling/internal/trace"
+)
+
+// carrierSpec registers one extractor whose value the page supplies. It stands
+// in for every spec whose getters carry state across steps (folio's last-seen
+// Home total, its submit counters): what matters is that ASKING the page for
+// the value is what advances it.
+const carrierSpec = `
+import { actions, extract } from "@sanderling/spec";
+const carrier = extract("carrier", () => 0);
+globalThis.properties = {};
+globalThis.actions = actions(() => []);
+`
+
+// carrierWebDriver is a web target that alternates between a cross-fading
+// hierarchy (which the runner discards as transitional) and a settled one, and
+// whose page-side extractor advances a counter on every evaluation - exactly
+// what a spec-authored carrier does in V8.
+type carrierWebDriver struct {
+	*mockdriver.Driver
+	transitional bool
+	snapshots    int
+	reads        int
+}
+
+func (d *carrierWebDriver) Snapshot(ctx context.Context) (string, driver.Image, error) {
+	_, image, err := d.Driver.Snapshot(ctx)
+	d.snapshots++
+	if !d.transitional {
+		return `{"attributes":{"resource-id":"HomeScreen"},"children":[]}`, image, err
+	}
+	// A genuine cross-fade: two live routes, and a tree that keeps changing
+	// between retries so the runner spends its whole retry budget on it.
+	return fmt.Sprintf(`{"attributes":{"resource-id":"root"},"children":[
+	  {"attributes":{"resource-id":"HomeScreen","text":"frame-%d"},"children":[]},
+	  {"attributes":{"resource-id":"LedgerScreen"},"children":[]}
+	]}`, d.snapshots), image, err
+}
+
+func (d *carrierWebDriver) InstallBundle(context.Context, []byte) error { return nil }
+
+func (d *carrierWebDriver) EvaluateExtractors(context.Context) (map[int]json.RawMessage, error) {
+	d.reads++
+	return map[int]json.RawMessage{0: json.RawMessage(strconv.Itoa(d.reads))}, nil
+}
+
+// NextActionFromV8 runs once per step, after the hierarchy fetch, so flipping
+// here makes every other step a cross-fade.
+func (d *carrierWebDriver) NextActionFromV8(context.Context) (json.RawMessage, error) {
+	d.transitional = !d.transitional
+	return json.RawMessage(`{"kind":"Tap","x":5,"y":5}`), nil
+}
+
+func (d *carrierWebDriver) SetLastAction(context.Context, json.RawMessage) error { return nil }
+
+// TestRunner_TransitionalStepNeverAdvancesThePageCarrier pins the ordering the
+// web path depends on. The page-side extractors must run only on steps the
+// verifier accepts: their getters advance spec state every time they evaluate,
+// so evaluating them on a step whose values are then discarded leaves the page
+// one window ahead of the verifier. The next accepted pair then brackets two
+// committed transactions while having counted one submit, and the property
+// convicts an app that did nothing wrong.
+func TestRunner_TransitionalStepNeverAdvancesThePageCarrier(t *testing.T) {
+	state := newHarnessWithSpec(t, carrierSpec)
+	web := &carrierWebDriver{Driver: state.mock}
+
+	ctx, cancel := context.WithTimeout(context.Background(), 60*time.Second)
+	defer cancel()
+	summary, err := Run(ctx, Options{
+		Duration:    30 * time.Second,
+		IdleTimeout: 20 * time.Millisecond,
+		MaxSteps:    5,
+		Driver:      web,
+		Verifier:    state.verifier,
+		TraceWriter: state.writer,
+	})
+	if err != nil {
+		t.Fatalf("Run: %v", err)
+	}
+	if summary.Steps != 5 {
+		t.Fatalf("steps = %d, want 5", summary.Steps)
+	}
+
+	type traceLine struct {
+		Step             int                              `json:"step"`
+		Transitional     bool                             `json:"transitional"`
+		ExtractorChanges map[string]trace.ExtractorChange `json:"extractor_changes"`
+	}
+	body, err := os.ReadFile(filepath.Join(state.writer.Directory(), "trace.jsonl"))
+	if err != nil {
+		t.Fatal(err)
+	}
+	verified, transitional := 0, 0
+	previous := 0
+	for _, raw := range bytes.Split(bytes.TrimSpace(body), []byte("\n")) {
+		var line traceLine
+		if err := json.Unmarshal(raw, &line); err != nil {
+			t.Fatalf("decode trace line: %v", err)
+		}
+		if line.Transitional {
+			transitional++
+			continue
+		}
+		verified++
+		change, ok := line.ExtractorChanges["carrier"]
+		if !ok {
+			t.Fatalf("step %d: no carrier value reached the verifier", line.Step)
+		}
+		current, convErr := strconv.Atoi(string(change.Curr))
+		if convErr != nil {
+			t.Fatalf("step %d: carrier value %s: %v", line.Step, change.Curr, convErr)
+		}
+		if current != previous+1 {
+			t.Errorf("step %d: carrier went %d -> %d; the page advanced it on a "+
+				"step the verifier discarded, so the verifier's window is wider "+
+				"than the one the spec counted actions over",
+				line.Step, previous, current)
+		}
+		previous = current
+	}
+	if verified == 0 || transitional == 0 {
+		t.Fatalf("need both kinds of step to prove anything: %d verified, %d transitional",
+			verified, transitional)
+	}
+	if web.reads != verified {
+		t.Errorf("the page evaluated its extractors %d time(s) across %d verified step(s); "+
+			"every evaluation the verifier does not use still advances spec state",
+			web.reads, verified)
+	}
+}
+
+// installFailsWebDriver is a web target whose page cannot take the runner's
+// lastAction: an older published @sanderling/spec runtime, a bundle that never
+// installed, a tab that navigated away from it.
+type installFailsWebDriver struct {
+	*mockdriver.Driver
+}
+
+func (d *installFailsWebDriver) InstallBundle(context.Context, []byte) error { return nil }
+
+func (d *installFailsWebDriver) EvaluateExtractors(context.Context) (map[int]json.RawMessage, error) {
+	return map[int]json.RawMessage{0: json.RawMessage(`1`)}, nil
+}
+
+func (d *installFailsWebDriver) NextActionFromV8(context.Context) (json.RawMessage, error) {
+	return json.RawMessage(`{"kind":"Tap","x":5,"y":5}`), nil
+}
+
+func (d *installFailsWebDriver) SetLastAction(context.Context, json.RawMessage) error {
+	return errors.New("__sanderlingSetLastAction__ is not a function")
+}
+
+// TestRunner_LastActionInstallFailureFailsTheRun covers the other half of the
+// same trust boundary. A run that cannot install lastAction in the page cannot
+// apply the page's extractor values either, so the step keeps goja's
+// dump-derived readings while the step before it holds the page's, and a delta
+// property compares two producers and fires. Downgraded to a warning that is a
+// green run reporting a violation nobody can reproduce.
+func TestRunner_LastActionInstallFailureFailsTheRun(t *testing.T) {
+	state := newHarnessWithSpec(t, carrierSpec)
+	web := &installFailsWebDriver{Driver: state.mock}
+
+	ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
+	defer cancel()
+	_, err := Run(ctx, Options{
+		Duration:    2 * time.Second,
+		IdleTimeout: 20 * time.Millisecond,
+		MaxSteps:    3,
+		Driver:      web,
+		Verifier:    state.verifier,
+		TraceWriter: state.writer,
+	})
+	if err == nil {
+		t.Fatal("Run succeeded with a page that cannot take lastAction; the run " +
+			"reported green while its extractor values came from two engines")
+	}
+	if !bytes.Contains([]byte(err.Error()), []byte("install last action")) {
+		t.Errorf("Run error = %v, want it to name the failed lastAction install", err)
+	}
+}

internal/runner/web_extractor_trace_test.go

@@ -6,6 +6,7 @@ import (
 	"encoding/json"
 	"os"
 	"path/filepath"
+	"strings"
 	"testing"
 	"time"
 
@@ -44,6 +45,8 @@ func (d *webMockDriver) NextActionFromV8(context.Context) (json.RawMessage, erro
 	return nil, nil
 }
 
+func (d *webMockDriver) SetLastAction(context.Context, json.RawMessage) error { return nil }
+
 // TestRunner_TraceRecordsTheValueTheVerdictUsed fails if the trace and the
 // verdict disagree about an extractor. A witness is only an explanation of a
 // violation if it holds the state the violated property was evaluated against.
@@ -116,3 +119,48 @@ func TestRunner_TraceRecordsTheValueTheVerdictUsed(t *testing.T) {
 		t.Error("no witness reached the trace; nothing was compared")
 	}
 }
+
+// splitTableSpec registers two extractors whose goja bodies both answer "goja".
+// The page below reports only the first, so index 1 keeps goja's dump-derived
+// reading while index 0 holds the page's.
+const splitTableSpec = `
+import { actions, extract } from "@sanderling/spec";
+extract("first", () => "goja");
+extract("second", () => "goja");
+globalThis.properties = {};
+globalThis.actions = actions(() => []);
+`
+
+// TestRunner_PartialExtractorTableIsFatal pins the failure the runner used to
+// let through. JSON.stringify drops an undefined-valued key, so a page whose
+// extractors are mostly undefined off their own screen reported a table with
+// holes in it, and the run completed with half the extractors reading from V8
+// and half from goja. A delta property spanning that split convicts an app that
+// did nothing wrong, which is worse than a crash: it is a green report of a bug
+// that is not there, or a red one for a bug nobody can reproduce.
+func TestRunner_PartialExtractorTableIsFatal(t *testing.T) {
+	state := newHarnessWithSpec(t, splitTableSpec)
+	web := &webMockDriver{
+		Driver:    state.mock,
+		overrides: map[int]json.RawMessage{0: json.RawMessage(`"v8"`)},
+	}
+
+	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
+	defer cancel()
+	_, err := Run(ctx, Options{
+		Duration:    time.Hour,
+		IdleTimeout: 20 * time.Millisecond,
+		MaxSteps:    2,
+		Driver:      web,
+		Verifier:    state.verifier,
+		TraceWriter: state.writer,
+	})
+	if err == nil {
+		t.Fatal("the run completed on a page that reported 1 of 2 extractors; " +
+			"the second extractor silently kept goja's value")
+	}
+	const want = "the page reported values for 1 of the spec's 2 extractors"
+	if !strings.Contains(err.Error(), want) {
+		t.Errorf("Run failed with %q, want it to name the split: %q", err, want)
+	}
+}

internal/runner/web_last_action_test.go

@@ -0,0 +1,78 @@
+package runner
+
+import (
+	"context"
+	"encoding/json"
+	"testing"
+	"time"
+
+	mockdriver "github.com/priyanshujain/sanderling/internal/driver/mock"
+)
+
+// On web the spec's extractors run in the page, so state.lastAction has to be
+// installed there by the runner. It used to be hardcoded null in
+// pkg/spec/src/web-runtime.ts, which made every property gated on the last
+// action (folio's submitMovesBalanceByTypedAmount, for one) vacuously true on
+// web: no failure, no warning, just a green run that proved nothing.
+
+const lastActionSpec = `
+import { actions } from "@sanderling/spec";
+globalThis.actions = actions(() => []);
+globalThis.properties = {};
+`
+
+// tappingWebDriver is a web target whose V8 picker always taps one named
+// control, so the runner has a real applied action to report on the next step.
+type tappingWebDriver struct {
+	*mockdriver.Driver
+	installed []string
+}
+
+func (d *tappingWebDriver) InstallBundle(context.Context, []byte) error { return nil }
+
+func (d *tappingWebDriver) EvaluateExtractors(context.Context) (map[int]json.RawMessage, error) {
+	return nil, nil
+}
+
+func (d *tappingWebDriver) NextActionFromV8(context.Context) (json.RawMessage, error) {
+	return json.RawMessage(`{"kind":"Tap","x":12,"y":34,"selector":"id:TxnSubmit"}`), nil
+}
+
+func (d *tappingWebDriver) SetLastAction(_ context.Context, encoded json.RawMessage) error {
+	d.installed = append(d.installed, string(encoded))
+	return nil
+}
+
+func TestRunner_WebInstallsLastActionInThePage(t *testing.T) {
+	state := newHarnessWithSpec(t, lastActionSpec)
+	web := &tappingWebDriver{Driver: state.mock}
+
+	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
+	defer cancel()
+	if _, err := Run(ctx, Options{
+		Duration:    time.Hour,
+		IdleTimeout: 20 * time.Millisecond,
+		MaxSteps:    3,
+		Driver:      web,
+		Verifier:    state.verifier,
+		TraceWriter: state.writer,
+	}); err != nil {
+		t.Fatalf("Run: %v", err)
+	}
+
+	if len(web.installed) < 2 {
+		t.Fatalf("the page was handed lastAction %d time(s); the web path never installed it",
+			len(web.installed))
+	}
+	// Step 1 has no previous action, exactly as the goja host reports it.
+	if web.installed[0] != "null" {
+		t.Errorf("step 1 installed %s, want null", web.installed[0])
+	}
+	// Every later step carries what the runner actually applied. The shape is
+	// the goja host's (internal/verifier/marshal.go lastActionFields), pinned
+	// against it by TestLastAction_WebJSONMatchesTheGojaObject.
+	const want = `{"kind":"Tap","on":"id:TxnSubmit"}`
+	if web.installed[1] != want {
+		t.Errorf("step 2 installed %s, want %s", web.installed[1], want)
+	}
+}