feat(test-runner): add deterministic gameplay scenarios

crates/desktop/src/cartridge/suite.rs

@@ -13,12 +13,14 @@ use std::{
 use gba_core::{
     bios,
     cpu::InstructionSet,
+    input::Buttons,
     machine::{Machine, StepKind},
+    video::{Framebuffer, HEIGHT, WIDTH},
 };
 use serde::{Deserialize, Deserializer, Serialize};
 use serde_json::{json, Value};
 
-use super::{load, Stats, MAX_STEPS};
+use super::{load, Stats, MAX_FRAMES, MAX_STEPS};
 
 const MAX_MANIFEST_BYTES: u64 = 1024 * 1024;
 const MAX_CASES: usize = 256;
@@ -76,15 +78,30 @@ struct Case {
     rom: PathBuf,
     step_limit: u64,
     completion: Completion,
+    #[serde(default)]
+    inputs: Option<Vec<Input>>,
     checks: Vec<Check>,
 }
 
 #[derive(Debug, Deserialize, Serialize)]
-#[serde(deny_unknown_fields)]
-struct Completion {
+#[serde(untagged, deny_unknown_fields)]
+enum Completion {
+    Pc {
         #[serde(deserialize_with = "word")]
         pc: u32,
         instruction_set: State,
+    },
+    VBlanks {
+        vblanks: u64,
+    },
+}
+
+#[derive(Debug, Deserialize, Serialize)]
+#[serde(deny_unknown_fields)]
+struct Input {
+    vblank: u64,
+    #[serde(deserialize_with = "word")]
+    buttons: u32,
 }
 
 #[derive(Debug, Clone, Copy, Deserialize, Serialize)]
@@ -106,6 +123,12 @@ impl State {
 #[derive(Debug, Deserialize, Serialize)]
 #[serde(tag = "kind", rename_all = "snake_case", deny_unknown_fields)]
 enum Check {
+    Pixel {
+        x: usize,
+        y: usize,
+        #[serde(deserialize_with = "word")]
+        equals: u32,
+    },
     Register {
         index: usize,
         #[serde(deserialize_with = "word")]
@@ -126,8 +149,8 @@ enum Check {
 impl Suite {
     fn validate(&self) -> io::Result<()> {
         let invalid = |message: &str| io::Error::new(io::ErrorKind::InvalidInput, message);
-        if self.version != 1 {
-            return Err(invalid("suite version must be 1"));
+        if !matches!(self.version, 1 | 2) {
+            return Err(invalid("suite version must be 1 or 2"));
         }
         if self.cases.is_empty() || self.cases.len() > MAX_CASES {
             return Err(invalid("suite must contain 1..=256 cases"));
@@ -150,18 +173,62 @@ impl Suite {
             if total > MAX_STEPS {
                 return Err(invalid("sum of case step limits must not exceed 100000000"));
             }
-            let alignment = match case.completion.instruction_set {
+            match case.completion {
+                Completion::Pc {
+                    pc,
+                    instruction_set,
+                } => {
+                    let alignment = match instruction_set {
                         State::Arm => 4,
                         State::Thumb => 2,
                     };
-            if case.completion.pc % alignment != 0 {
+                    if pc % alignment != 0 {
                         return Err(invalid("completion PC must be aligned for its instruction set; do not set the Thumb address bit"));
                     }
+                    if case.inputs.is_some() {
+                        return Err(invalid("input schedules require VBlank completion"));
+                    }
+                }
+                Completion::VBlanks { vblanks } => {
+                    if self.version != 2 || !(1..=MAX_FRAMES).contains(&vblanks) {
+                        return Err(invalid(
+                            "VBlank completion requires version 2 and 1..=100000 vblanks",
+                        ));
+                    }
+                    if let Some(inputs) = &case.inputs {
+                        if inputs.len() > 256 {
+                            return Err(invalid(
+                                "input schedule must contain at most 256 snapshots",
+                            ));
+                        }
+                        let mut previous = None;
+                        for input in inputs {
+                            if input.buttons > 0x3ff
+                                || input.vblank >= vblanks
+                                || previous.is_some_and(|frame| input.vblank <= frame)
+                            {
+                                return Err(invalid("inputs need 10-bit button masks and strictly increasing VBlanks before completion"));
+                            }
+                            previous = Some(input.vblank);
+                        }
+                    }
+                }
+            }
             if case.checks.is_empty() || case.checks.len() > MAX_CHECKS {
                 return Err(invalid("each case must contain 1..=256 checks"));
             }
             for check in &case.checks {
                 match check {
+                    Check::Pixel { x, y, equals } => {
+                        if self.version != 2
+                            || !matches!(case.completion, Completion::VBlanks { .. })
+                            || *x >= WIDTH
+                            || *y >= HEIGHT
+                            || *equals > 0xff_ffff
+                        {
+                            return Err(invalid("pixel checks require version 2, VBlank completion, x < 240, y < 160, and RGB888 values"));
+                        }
+                    }
                     Check::Register { index, .. } if *index > 15 => {
                         return Err(invalid("register index must be 0..=15"))
                     }
@@ -205,20 +272,53 @@ enum Outcome {
     StepLimit,
     Stopped,
     EmulationError(String),
+    RenderError(String),
 }
 
 fn run_to_checkpoint(machine: &mut Machine, case: &Case) -> (Stats, Outcome) {
     let mut stats = Stats::default();
+    let capture = matches!(case.completion, Completion::VBlanks { .. });
+    machine.memory_mut().set_scanline_rendering(capture);
+    let mut frame = Framebuffer::default();
+    let mut last_vblank = 0;
+    let mut inputs = case.inputs.iter().flatten().peekable();
     loop {
         if machine.stopped() {
             return (stats, Outcome::Stopped);
         }
-        // Check after successful progress, before trying to execute the checkpoint.
-        // The final budgeted step can reach it. No initial/reset-state pass is allowed.
-        if stats.steps != 0
-            && machine.cpu().pc() == case.completion.pc
-            && machine.cpu().instruction_set() == case.completion.instruction_set.instruction_set()
-        {
+        let vblanks = machine.memory().display_position().vblanks;
+        if capture && vblanks != last_vblank {
+            match machine.memory().present_frame(&mut frame) {
+                Ok(true) => {}
+                Ok(false) => {
+                    return (
+                        stats,
+                        Outcome::RenderError("no complete captured frame at VBlank".into()),
+                    )
+                }
+                Err(error) => return (stats, Outcome::RenderError(error.to_string())),
+            }
+            last_vblank = vblanks;
+        }
+        // Snapshots apply after this VBlank's capture, before the next machine step.
+        while inputs.peek().is_some_and(|input| input.vblank <= vblanks) {
+            let input = inputs.next().unwrap();
+            machine
+                .memory_mut()
+                .set_buttons(Buttons::from_bits(input.buttons as u16));
+        }
+        // The final budgeted step can complete a case; reset alone cannot.
+        let complete = match case.completion {
+            Completion::Pc {
+                pc,
+                instruction_set,
+            } => {
+                machine.cpu().pc() == pc
+                    && machine.cpu().instruction_set() == instruction_set.instruction_set()
+            }
+            Completion::VBlanks { vblanks: target } => vblanks >= target,
+        };
+        if stats.steps != 0 && complete {
             return (stats, Outcome::Checkpoint);
         }
         if stats.steps == case.step_limit {
@@ -234,8 +334,23 @@ fn run_to_checkpoint(machine: &mut Machine, case: &Case) -> (Stats, Outcome) {
 
 fn check(machine: &Machine, check: &Check) -> Value {
     let (expected, actual) = match check {
+        Check::Pixel { x, y, equals } => {
+            let mut frame = Framebuffer::default();
+            let actual = match machine.memory().present_frame(&mut frame) {
+                Ok(true) => Ok(frame.pixels()[y * WIDTH + x]),
+                Ok(false) => Err("no complete captured frame".to_owned()),
+                Err(error) => Err(error.to_string()),
+            };
+            (*equals, actual)
+        }
         Check::Register { index, equals } => (*equals, Ok(machine.cpu().registers()[*index])),
-        Check::Memory32 { address, equals } => (*equals, machine.memory().read32(*address)),
+        Check::Memory32 { address, equals } => (
+            *equals,
+            machine
+                .memory()
+                .read32(*address)
+                .map_err(|error| error.to_string()),
+        ),
         Check::Cpsr { equals } => (*equals, Ok(machine.cpu().cpsr())),
     };
     match actual {
@@ -285,6 +400,11 @@ fn run_case(case: &Case, base: &Path) -> Value {
     };
     let (stats, outcome) = run_to_checkpoint(&mut machine, case);
     report["state"] = state(&machine, &stats);
+    if matches!(case.completion, Completion::VBlanks { .. }) {
+        report["inputs"] = json!(case.inputs.as_deref().unwrap_or_default());
+        report["state"]["vblanks"] = json!(machine.memory().display_position().vblanks);
+        report["state"]["captured_vblank"] = json!(machine.memory().captured_vblank());
+    }
     let reason = match outcome {
         Outcome::Checkpoint => {
             let checks: Vec<_> = case
@@ -303,6 +423,10 @@ fn run_case(case: &Case, base: &Path) -> Value {
         }
         Outcome::StepLimit => "step_limit",
         Outcome::Stopped => "stopped",
+        Outcome::RenderError(error) => {
+            report["error"] = json!(error);
+            "render_error"
+        }
         Outcome::EmulationError(error) => {
             report["error"] = json!(error);
             "emulation_error"
@@ -330,7 +454,7 @@ pub fn execute(path: &Path, writer: &mut impl Write) -> Result<(), Box<dyn Error
         .collect();
     let failed = cases.iter().filter(|case| case["passed"] != true).count();
     let report = json!({
-        "format_version": 1, "bios": "original", "passed": failed == 0,
+        "format_version": suite.version, "bios": "original", "passed": failed == 0,
         "case_count": cases.len(), "failed_count": failed, "cases": cases,
     });
     write_json_report(writer, &report)?;
@@ -340,5 +464,7 @@ pub fn execute(path: &Path, writer: &mut impl Write) -> Result<(), Box<dyn Error
     Ok(())
 }
 
+#[cfg(test)]
+mod gameplay_tests;
 #[cfg(test)]
 mod tests;

crates/desktop/src/cartridge/suite/gameplay_tests.rs

@@ -0,0 +1,142 @@
+use super::tests::{boot, suite};
+use super::*;
+use gba_core::io::{DISPCNT, KEYINPUT};
+
+fn value() -> Value {
+    json!({"name":"input", "rom":"original.gba", "step_limit":1000000,
+        "completion":{"vblanks":4},
+        "inputs":[{"vblank":0,"buttons":1}, {"vblank":1,"buttons":17}, {"vblank":3,"buttons":0}],
+        "checks":[{"kind":"pixel","x":0,"y":0,"equals":255}]})
+}
+fn config(value: Value) -> Case {
+    suite(json!({"version":2,"cases":[value]}))
+        .unwrap()
+        .cases
+        .remove(0)
+}
+
+#[test]
+fn strict_gameplay_schema_and_bounds() {
+    assert!(suite(json!({"version":1,"cases":[value()]})).is_err());
+    assert!(suite(json!({"version":2,"cases":[value()]})).is_ok());
+    for (pointer, replacement) in [
+        ("/completion", json!({"vblanks":0})),
+        ("/completion", json!({"vblanks":100001})),
+        (
+            "/completion",
+            json!({"vblanks":1,"pc":0,"instruction_set":"arm"}),
+        ),
+        ("/completion", json!({"vblanks":1,"typo":0})),
+        ("/completion", json!({"pc":0})),
+        ("/completion", json!({"pc":0,"instruction_set":"arm"})),
+        ("/inputs", json!([{"vblank":0,"buttons":1024}])),
+        ("/inputs", json!([{"vblank":4,"buttons":0}])),
+        (
+            "/inputs",
+            json!([{"vblank":1,"buttons":0},{"vblank":1,"buttons":1}]),
+        ),
+        (
+            "/inputs",
+            json!([{"vblank":2,"buttons":0},{"vblank":1,"buttons":1}]),
+        ),
+        ("/inputs", json!([{"vblank":0,"buttons":0,"typo":0}])),
+        ("/inputs", json!(vec![json!({"vblank":0,"buttons":0}); 257])),
+        ("/inputs/0/buttons", json!(-1)),
+        ("/inputs/0/vblank", json!(0.5)),
+        ("/checks/0/x", json!(240)),
+        ("/checks/0/y", json!(160)),
+        ("/checks/0/equals", json!(0x1000000)),
+    ] {
+        let mut input = value();
+        *input.pointer_mut(pointer).unwrap() = replacement;
+        assert!(
+            suite(json!({"version":2,"cases":[input]})).is_err(),
+            "{pointer}"
+        );
+    }
+    let mut input = value();
+    input["completion"]["vblanks"] = json!(100000);
+    input["inputs"] = json!([{"vblank":99999,"buttons":"0x3ff"}]);
+    input["checks"][0] = json!({"kind":"pixel","x":239,"y":159,"equals":"0xffffff"});
+    config(input);
+}
+
+#[test]
+fn cpu_observes_initial_held_replaced_and_released_snapshots() {
+    for (target, expected, keys) in [
+        (1, 0xff0000, 0x3fe),
+        (2, 0x00ff00, 0x3ee),
+        (3, 0x00ff00, 0x3ee),
+        (4, 0x0000ff, 0x3ff),
+    ] {
+        let mut input = value();
+        input["completion"]["vblanks"] = json!(target);
+        input["inputs"]
+            .as_array_mut()
+            .unwrap()
+            .retain(|event| event["vblank"].as_u64().unwrap() < target);
+        input["checks"][0]["equals"] = json!(expected);
+        let case = config(input);
+        let mut machine = bios::boot(gba_demos::input_rom()).unwrap();
+        let (_, outcome) = run_to_checkpoint(&mut machine, &case);
+        assert_eq!(outcome, Outcome::Checkpoint);
+        assert_eq!(machine.memory().captured_vblank(), Some(target));
+        assert_eq!(machine.memory().read16(KEYINPUT).unwrap(), keys);
+        assert_eq!(check(&machine, &case.checks[0])["passed"], true);
+        // A post-capture palette write must not change the asserted image.
+        machine.memory_mut().write16(0x05000000, 0x7fff).unwrap();
+        assert_eq!(check(&machine, &case.checks[0])["actual"], expected);
+        let mut frame = Framebuffer::default();
+        machine.memory().present_frame(&mut frame).unwrap();
+        assert!(frame.pixels().iter().all(|pixel| *pixel == expected));
+    }
+}
+
+#[test]
+fn frame_completion_accepts_final_step_but_not_a_short_budget() {
+    let mut case = config(value());
+    let run =
+        |case: &Case| run_to_checkpoint(&mut bios::boot(gba_demos::input_rom()).unwrap(), case);
+    let (stats, outcome) = run(&case);
+    assert_eq!(outcome, Outcome::Checkpoint);
+    case.step_limit = stats.steps;
+    assert_eq!(run(&case).1, Outcome::Checkpoint);
+    case.step_limit -= 1;
+    assert_eq!(run(&case).1, Outcome::StepLimit);
+}
+
+#[test]
+fn no_frame_is_not_a_black_pixel_pass_and_mismatch_is_reported() {
+    let mut machine = bios::boot(gba_demos::input_rom()).unwrap();
+    let black = Check::Pixel {
+        x: 0,
+        y: 0,
+        equals: 0,
+    };
+    let result = check(&machine, &black);
+    assert_eq!(result["passed"], false);
+    assert!(result["actual"].is_null());
+    assert!(result["error"].is_string());
+    run_to_checkpoint(&mut machine, &config(value()));
+    let result = check(&machine, &black);
+    assert_eq!(result["passed"], false);
+    assert_eq!(result["actual"], 255);
+}
+
+#[test]
+fn first_render_diagnostic_stops_before_later_checkpoint() {
+    let mut machine = boot(&[0xeaff_fffe]);
+    machine.memory_mut().write16(DISPCNT, 7).unwrap();
+    let (stats, outcome) = run_to_checkpoint(&mut machine, &config(value()));
+    assert!(matches!(outcome, Outcome::RenderError(_)));
+    assert_eq!(machine.memory().display_position().vblanks, 1);
+    assert!(stats.steps > 0);
+}
+
+#[test]
+fn future_input_does_not_advance_stopped_hardware() {
+    let mut machine = boot(&[0xef03_0000, 0xeaff_fffe]);
+    let (_, outcome) = run_to_checkpoint(&mut machine, &config(value()));
+    assert_eq!(outcome, Outcome::Stopped);
+    assert_eq!(machine.memory().display_position().vblanks, 0);
+}

crates/desktop/src/cartridge/suite/tests.rs

@@ -13,12 +13,12 @@ fn case_value() -> Value {
 fn case() -> Case {
     serde_json::from_value(case_value()).unwrap()
 }
-fn suite(value: Value) -> Result<Suite, Box<dyn Error>> {
+pub(super) fn suite(value: Value) -> Result<Suite, Box<dyn Error>> {
     let suite: Suite = serde_json::from_value(value)?;
     suite.validate()?;
     Ok(suite)
 }
-fn boot(words: &[u32]) -> Machine {
+pub(super) fn boot(words: &[u32]) -> Machine {
     bios::boot(words.iter().flat_map(|word| word.to_le_bytes()).collect()).unwrap()
 }
 
@@ -90,7 +90,7 @@ fn unknown_fields_and_unknown_check_kinds_are_rejected() {
 
 #[test]
 fn manifest_rejects_empty_duplicate_excessive_or_invalid_cases() {
-    assert!(suite(json!({"version": 2, "cases": [case_value()]})).is_err());
+    assert!(suite(json!({"version": 3, "cases": [case_value()]})).is_err());
     assert!(suite(json!({"version": 1, "cases": []})).is_err());
     assert!(suite(json!({"version": 1, "cases": [case_value(), case_value()]})).is_err());
     for (field, invalid) in [
@@ -184,7 +184,10 @@ fn json_report_propagates_write_and_flush_failures() {
 #[test]
 fn reset_pc_is_not_an_immediate_pass_without_execution() {
     let mut config = case();
-    config.completion.pc = 0;
+    config.completion = Completion::Pc {
+        pc: 0,
+        instruction_set: State::Arm,
+    };
     config.step_limit = 1;
     let (stats, outcome) = run_to_checkpoint(&mut boot(&[0xeaff_fffe]), &config);
     assert_eq!(stats.steps, 1);
@@ -215,14 +218,19 @@ fn checkpoint_is_checked_before_execution_and_on_the_last_budgeted_step() {
 #[test]
 fn matching_assertions_do_not_pass_without_the_checkpoint() {
     let mut config = case();
-    config.completion.pc = ROM_START + 8;
+    config.completion = Completion::Pc {
+        pc: ROM_START + 8,
+        instruction_set: State::Arm,
+    };
     let mut machine = boot(&[0xe3a0_002a, 0xeaff_fffe]);
     let (stats, outcome) = run_to_checkpoint(&mut machine, &config);
     assert_eq!(outcome, Outcome::StepLimit);
     assert_eq!(stats.steps, config.step_limit);
     assert_eq!(check(&machine, &config.checks[0])["passed"], true);
-    config.completion.pc = ROM_START + 4;
-    config.completion.instruction_set = State::Thumb;
+    config.completion = Completion::Pc {
+        pc: ROM_START + 4,
+        instruction_set: State::Thumb,
+    };
     assert_eq!(
         run_to_checkpoint(&mut boot(&[0xe3a0_002a, 0xeaff_fffe]), &config).1,
         Outcome::StepLimit
@@ -238,8 +246,10 @@ fn thumb_checkpoint_uses_aligned_architectural_pc() {
     bytes.extend([0x2a, 0x20, 0xfe, 0xe7]);
     let mut machine = bios::boot(bytes).unwrap();
     let mut config = case();
-    config.completion.pc = ROM_START + 14;
-    config.completion.instruction_set = State::Thumb;
+    config.completion = Completion::Pc {
+        pc: ROM_START + 14,
+        instruction_set: State::Thumb,
+    };
     assert_eq!(
         run_to_checkpoint(&mut machine, &config).1,
         Outcome::Checkpoint
@@ -250,7 +260,10 @@ fn thumb_checkpoint_uses_aligned_architectural_pc() {
 #[test]
 fn memory_and_cpsr_checks_report_actual_values_and_read_errors() {
     let mut config = case();
-    config.completion.pc = ROM_START + 12;
+    config.completion = Completion::Pc {
+        pc: ROM_START + 12,
+        instruction_set: State::Arm,
+    };
     let mut machine = boot(&[0xe3a0_002a, 0xe3a0_1402, 0xe581_0000, 0xeaff_fffe]);
     assert_eq!(
         run_to_checkpoint(&mut machine, &config).1,
@@ -291,7 +304,10 @@ fn memory_and_cpsr_checks_report_actual_values_and_read_errors() {
 #[test]
 fn stop_is_failure_while_halt_consumes_a_bounded_budget() {
     let mut config = case();
-    config.completion.pc = ROM_START + 8;
+    config.completion = Completion::Pc {
+        pc: ROM_START + 8,
+        instruction_set: State::Arm,
+    };
     let (stats, outcome) = run_to_checkpoint(&mut boot(&[0xef03_0000, 0xeaff_fffe]), &config);
     assert_eq!(outcome, Outcome::Stopped);
     assert!(stats.steps < config.step_limit);
@@ -322,7 +338,10 @@ fn dma_and_irq_each_consume_a_budget_step_and_dma_errors_fail() {
     let mut machine = boot(&[0xe3a0_002a, 0xeaff_fffe]);
     run_to_checkpoint(&mut machine, &case());
     let mut config = case();
-    config.completion.pc = ROM_START + 8;
+    config.completion = Completion::Pc {
+        pc: ROM_START + 8,
+        instruction_set: State::Arm,
+    };
     config.step_limit = 3;
     let bus = machine.memory_mut();
     bus.write32(DMA_BASE, 0x0200_0000).unwrap();

crates/desktop/tests/rom_suite_cli.rs

@@ -204,6 +204,49 @@ fn malformed_or_oversized_suites_fail_before_any_case_execution() {
     assert!(output.stdout.is_empty());
 }
 
+#[test]
+fn gameplay_reports_inputs_captured_pixels_and_failures_repeatably() {
+    let fixture = Fixture::new();
+    fs::write(fixture.0.join("input.gba"), gba_demos::input_rom()).unwrap();
+    fixture.rom(
+        "bad-video.gba",
+        &[0xe3a0_1301, 0xe3a0_0007, 0xe1c1_00b0, 0xeaff_fffe],
+    );
+    let good = json!({"name":"gameplay", "rom":"input.gba", "step_limit":1000000,
+        "completion":{"vblanks":3},
+        "inputs":[{"vblank":0,"buttons":1},{"vblank":1,"buttons":17},{"vblank":2,"buttons":0}],
+        "checks":[{"kind":"pixel","x":239,"y":159,"equals":255}]});
+    let mut mismatch = good.clone();
+    mismatch["name"] = json!("mismatch");
+    mismatch["checks"][0]["equals"] = json!(0);
+    let mut render_error = good.clone();
+    render_error["name"] = json!("bad-video");
+    render_error["rom"] = json!("bad-video.gba");
+    let path = fixture.0.join("gameplay.json");
+    fs::write(
+        &path,
+        serde_json::to_vec(&json!({"version":2,"cases":[good,mismatch,render_error]})).unwrap(),
+    )
+    .unwrap();
+    let first = run(&path);
+    assert_eq!(first.status.code(), Some(1));
+    assert_eq!(first.stdout, run(&path).stdout);
+    let report = report(&first);
+    assert_eq!(report["format_version"], 2);
+    assert_eq!(report["failed_count"], 2);
+    let good = &report["cases"][0];
+    assert_eq!(good["passed"], true);
+    assert_eq!(good["state"]["vblanks"], 3);
+    assert_eq!(good["state"]["captured_vblank"], 3);
+    assert_eq!(good["inputs"][1]["buttons"], 17);
+    assert_eq!(good["checks"][0]["actual"], 255);
+    assert_eq!(report["cases"][1]["reason"], "assertion_failed");
+    assert_eq!(report["cases"][2]["reason"], "render_error");
+    assert_eq!(report["cases"][2]["state"]["vblanks"], 1);
+    assert_eq!(report["cases"][2]["checks"], json!([]));
+    assert!(report["cases"][2]["error"].is_string());
+}
+
 #[test]
 fn suite_mode_rejects_other_modes_and_help_advertises_it() {
     for args in [

docs/architecture.md

@@ -117,6 +117,7 @@ The repository is a Cargo workspace with three crates:
 | `crates/desktop/src/cartridge.rs` | Read-only ROM files, mode/limit checks, bounded terminal execution, and shared state reports |
 | `crates/desktop/src/cartridge/suite.rs` | Strict JSON manifests, bounded checkpoint execution, assertions, and structured reports |
 | `crates/desktop/src/cartridge/suite/tests.rs` | Manifest limits, checkpoint boundaries, memory/register checks, and diagnostic accounting |
+| `crates/desktop/src/cartridge/suite/gameplay_tests.rs` | VBlank completion, input snapshots, captured pixels, and rendering failures |
 | `crates/desktop/tests/rom_suite_cli.rs` | File-backed suite execution, case isolation, output schema, determinism, and exit status |
 | `crates/desktop/examples/write_test_suite.rs` | Original ARM/Thumb/BIOS fixture generator with a local JSON suite |
 | `crates/desktop/src/cartridge/window.rs` | ROM-window session, bounded slices, capture, STOP/input handling, and progress limits |

docs/rom-tests.md

@@ -1,7 +1,7 @@
 # Headless ROM test suites
 
 Use `--test-suite PATH.json` to run repeatable ROM checks without opening a window.
-Each case has a machine-step budget, a completion address, and explicit assertions.
+Each case has a machine-step budget, a completion condition, and explicit assertions.
 A timeout never counts as a pass. Existing `--rom ... --steps ...` remains a diagnostic mode, not a test verdict.
 
 Original programs cover the runner itself. Pinned public [ARM](public-arm-tests.md), [Thumb](public-thumb-tests.md), [memory](public-memory-tests.md), and [BIOS](public-bios-tests.md) ROMs provide independent compatibility results.
@@ -71,12 +71,55 @@ The runner validates the whole manifest before opening or executing any ROM.
 It uses the same read-only [ROM loader](hardware/cartridge.md#loading-and-boot), with its 4-byte minimum and 32 MiB maximum.
 The CLI accepts no other options with `--test-suite`.
 
+## Manifest version 2: gameplay checks
+
+Version 2 accepts version 1 PC checkpoints and adds VBlank completion, button snapshots, and captured-pixel checks.
+Version 1 behavior and report fields remain unchanged.
+
+```json
+{
+  "version": 2,
+  "cases": [{
+    "name": "original-input-release",
+    "rom": "input.gba",
+    "step_limit": 1000000,
+    "completion": {"vblanks": 3},
+    "inputs": [
+      {"vblank": 0, "buttons": "0x001"},
+      {"vblank": 1, "buttons": "0x011"},
+      {"vblank": 2, "buttons": 0}
+    ],
+    "checks": [{"kind": "pixel", "x": 239, "y": 159, "equals": "0x0000ff"}]
+  }]
+}
+```
+
+This example uses the original input ROM from `cargo run --example write_rom_demo -- /tmp/input.gba`.
+Place the manifest alongside that ROM. The CPU writes red for A, green for Right, and blue after release.
+
+- Use exactly one completion form: PC/state or `vblanks`. Mixed forms are invalid.
+- VBlank completion requires an integer count from 1 through 100,000, measured from machine reset.
+- The existing step budget still applies. A VBlank target does not permit unbounded execution.
+- `inputs` is optional and only valid with VBlank completion. Supply at most 256 snapshots.
+- Snapshot VBlank numbers must increase strictly and remain below the completion count. Zero applies before the first step.
+- Each snapshot replaces all held buttons. Its state remains active until the next snapshot.
+- Button masks use bits 0–9: A, B, Select, Start, Right, Left, Up, Down, R, L. One means pressed.
+- Snapshots apply at the first machine-step boundary at or after their VBlank, after capture and before the next step.
+- Pixel checks require VBlank completion, `x` from 0–239, `y` from 0–159, and packed `0xRRGGBB` expected values.
+- Pixel checks read the latest completed scanline capture, not a snapshot of current video RAM.
+- All VBlank cases enable capture. The first published render diagnostic fails the case, even before its target.
+- A missing captured frame is an error, never an implicit black image.
+
+Input timing is deterministic but not sub-instruction accurate. A snapshot cannot affect the frame already captured at that VBlank.
+STOP still fails: emulated VBlanks cannot advance to a future scheduled wake-up event.
+This is a regression runner, not a replacement for manual gameplay and host-keyboard checks.
+
 ## Completion and failure semantics
 
 Each case starts with fresh RAM, devices, CPU state, and the original BIOS replacement.
 The budget includes BIOS boot, instructions, DMA units, IRQ entries, and HALT idle batches.
 A failed CPU or DMA step does not consume a successful step or advance the reported state.
-The runner supplies no input and does not enable frame capture. Device clocks still advance through normal machine stepping.
+PC-checkpoint cases supply no input and do not enable frame capture. Device clocks still advance through normal machine stepping.
 
 After successful progress, the runner checks PC and instruction set at machine-step boundaries.
 It stops before executing the instruction at the completion address.
@@ -96,13 +139,13 @@ A failure does not skip later cases. No machine state passes from one case to an
 The budget bounds machine steps, not elapsed host time. Use an external process timeout in continuous integration if needed.
 This is not a filesystem security sandbox. Run trusted suite configurations and use files you may lawfully use.
 
-## Report version 1 and exit status
+## Reports and exit status
 
 Top-level fields:
 
 | Field | Meaning |
 | --- | --- |
-| `format_version` | Report schema version, currently 1 |
+| `format_version` | Manifest version: 1 or 2 |
 | `bios` | `original`; no external BIOS option |
 | `passed` | True only when every case passed |
 | `case_count`, `failed_count` | Number of executed cases and failures |
@@ -115,6 +158,9 @@ A load failure has null `state` and `rom_bytes`.
 `checks` contains the configured check, actual value, pass flag, and optional read error when the checkpoint was reached.
 Otherwise `checks` is empty. A failed read has a null actual value.
 
+Version 2 VBlank cases also report their normalized `inputs` and state counters `vblanks` and `captured_vblank`.
+These state fields are absent when loading fails. A PC case retains the version 1 case layout.
+
 Case reasons are:
 
 - `checkpoint`: completion reached and all assertions passed.
@@ -123,6 +169,7 @@ Case reasons are:
 - `stopped`: STOP prevented further execution.
 - `emulation_error`: an instruction or DMA diagnostic stopped execution; `error` contains its message.
 - `load_error`: loading or machine creation failed; `error` contains its message.
+- `render_error`: frame capture failed or no complete frame was available at VBlank; `error` contains its message.
 
 Exit status 0 means all cases passed and the report was written successfully.
 Any failed case, invalid configuration, or output failure returns a nonzero status.
@@ -143,5 +190,6 @@ The ROM, assertions, checkpoint, and budget remain unchanged. This is not a comp
 For further public ARM7TDMI/GBA tests, inspect their source, license, entry assumptions, and result protocol.
 Pin source revisions and build instructions before comparing emulator changes.
 Adapt suites to verified completion addresses and result locations; do not guess them or treat timeouts as passes.
-Some tests need debug-port logging, input, different firmware behavior, or rendered-image checks that this runner does not support yet.
+Version 2 supports scheduled input and exact captured-pixel checks.
+Debug-port logging, external firmware selection, audio assertions, and whole-image comparisons remain unsupported.
 Keep external binaries local. Record consulted sources in [references.md](references.md).

docs/status.md

@@ -78,7 +78,7 @@ Update this file when a feature lands or a limit is removed.
 - Raw ROM-file loading with original BIOS boot, terminal diagnostics, and final CPU/step reports.
 - ROM windows with scanline capture, keyboard input, focus-loss release, bounded execution slices, and interactive keypad wake from STOP.
 - Optional ROM-window frame limits and an original input-test ROM generator.
-- Headless JSON ROM suites with per-case checkpoints, register/CPSR/memory assertions, bounded budgets, and structured failure reports.
+- Headless JSON ROM suites with PC or VBlank completion, register/CPSR/memory/pixel assertions, scheduled buttons, bounded budgets, and structured failure reports.
 - Hash-pinned preparation of public `jsmolka/gba-tests` ARM, Thumb, memory, and BIOS ROMs, with verified result-register checkpoints.
 - These pinned ROMs pass their checkpoints on Darwin arm64 in debug and release builds; coverage limits remain documented.
 - The original BIOS exposes documented boot, SoftReset, SWI, and IRQ protected-read words through its actual exit layout.

docs/testing.md

@@ -48,6 +48,8 @@ Do not set a Linux cross-compilation target for this validation.
 
 The [headless ROM suite runner](rom-tests.md) provides explicit checkpoint and assertion results through `--test-suite PATH.json`.
 Its tests verify manifest limits, whole-suite validation, exact step boundaries, ARM/Thumb checkpoints, and register/CPSR/memory checks.
+Version 2 tests cover VBlank targets, initial/held/released input, captured RGB888 pixels, and render diagnostics.
+The original input ROM verifies CPU-observed button snapshots and every captured pixel without external game assets.
 Additional tests cover DMA/IRQ accounting, HALT/STOP, diagnostic state, case isolation, repeatable JSON output, and failure exit status.
 These original tests validate the runner. The [public ARM adapter](public-arm-tests.md) provides a separate passing checkpoint result for its pinned ROM.
 The [public Thumb adapter](public-thumb-tests.md) uses a separate lock and r7 assertion after its return to ARM state.