test_that("freeze_prompt creates a frozen_prompt with correct hash", {
  fp <- freeze_prompt(
    system_prompt = "Classify sentiment.",
    user_template = "Text: {text}",
    model = "gpt-4o",
    temperature = 0,
    categories = c("positive", "negative", "neutral")
  )

  expect_s3_class(fp, "frozen_prompt")
  expect_type(fp$hash, "character")
  expect_equal(nchar(fp$hash), 64)  # SHA-256 = 64 hex chars
  expect_equal(fp$model, "gpt-4o")
  expect_equal(fp$temperature, 0)
  expect_equal(fp$categories, c("positive", "negative", "neutral"))
})

test_that("freeze_prompt hash is deterministic", {
  fp1 <- freeze_prompt("Classify.", "Text: {text}", "gpt-4o", 0)
  fp2 <- freeze_prompt("Classify.", "Text: {text}", "gpt-4o", 0)
  expect_equal(fp1$hash, fp2$hash)
})

test_that("freeze_prompt hash changes when spec changes", {
  fp1 <- freeze_prompt("Classify.", "Text: {text}", "gpt-4o", 0)
  fp2 <- freeze_prompt("Classify sentiment.", "Text: {text}", "gpt-4o", 0)
  expect_false(fp1$hash == fp2$hash)
})

test_that("freeze_prompt category order does not affect hash", {
  fp1 <- freeze_prompt("Classify.", "{text}", "gpt-4o", 0,
                        categories = c("a", "b", "c"))
  fp2 <- freeze_prompt("Classify.", "{text}", "gpt-4o", 0,
                        categories = c("c", "a", "b"))
  expect_equal(fp1$hash, fp2$hash)
})

test_that("verify_prompt returns TRUE for unmodified prompt", {
  fp <- freeze_prompt("Test.", "{text}", "gpt-4o", 0)
  expect_true(verify_prompt(fp))
})

test_that("verify_prompt returns FALSE for tampered prompt", {
  fp <- freeze_prompt("Test.", "{text}", "gpt-4o", 0)
  fp$system_prompt <- "Tampered!"
  expect_false(verify_prompt(fp))
})

test_that("print.frozen_prompt works", {
  fp <- freeze_prompt("Classify.", "{text}", "gpt-4o", 0,
                       categories = c("pos", "neg"))
  expect_output(print(fp), "Frozen Prompt")
  expect_output(print(fp), "gpt-4o")
})

test_that("export and reload round-trip preserves spec", {
  fp <- freeze_prompt(
    system_prompt = "Classify sentiment.",
    user_template = "Text: {text}",
    model = "gpt-4o",
    temperature = 0,
    categories = c("positive", "negative")
  )

  tmp <- withr::local_tempfile(fileext = ".json")
  export_preregistration(fp, tmp)
  expect_true(file.exists(tmp))

  reloaded <- load_frozen_prompt(tmp)
  expect_s3_class(reloaded, "frozen_prompt")
  expect_equal(reloaded$hash, fp$hash)
  expect_equal(reloaded$system_prompt, fp$system_prompt)
  expect_equal(reloaded$model, fp$model)
  expect_equal(reloaded$temperature, fp$temperature)

  expect_true(verify_prompt(reloaded))
})

test_that("export_preregistration refuses to overwrite by default", {
  fp <- freeze_prompt("Test.", "{text}", "gpt-4o", 0)
  tmp <- withr::local_tempfile(fileext = ".json")
  export_preregistration(fp, tmp)
  expect_error(export_preregistration(fp, tmp), "already exists")
})

test_that("export_preregistration overwrites with flag", {
  fp <- freeze_prompt("Test.", "{text}", "gpt-4o", 0)
  tmp <- withr::local_tempfile(fileext = ".json")
  export_preregistration(fp, tmp)
  expect_no_error(export_preregistration(fp, tmp, overwrite = TRUE))
})

test_that("freeze_prompt validates inputs", {
  expect_error(freeze_prompt(123, "{text}", "gpt-4o"))
  expect_error(freeze_prompt("Classify.", 123, "gpt-4o"))
})