The published 1.9.1 application saved 41,505 PDFs across the three main collections, 541 more than the previous 1.9.0 candidate record. No previously saved input became unsavable. Each normal collection received one warmup and five measured passes, with identical per-file outcomes throughout. All 80 damaged inputs, including files without a .pdf extension, were tested separately with a 30-second limit per file and produced no crashes or timeouts.

This measures batch opening and saving, not rendering accuracy or PDF conformance. There were 3,973 skips and 1,466 save failures in the normal collections. Timings describe this run on the main Windows computer; no cross-session speed or memory improvement is claimed.

Source: v1.9.1 report, measured runs, and test method.

The frozen 1.9.0 release candidate saved 40,964 PDFs across the three main collections, 401 more than 1.8.91, and no file got worse: 400 inputs that 1.8.91 failed and one it skipped now save. Both versions received one warmup and five measured passes of the regression and standards sets in alternating order, and every pass returned identical per-file outcomes. The 80 damaged inputs produced no crashes or timeouts on October 15.

Against 1.8.91, the standards and stress sets finished about 12% sooner and regression 3.4% sooner, with 4% to 7% less CPU time and 37% to 67% lower peak memory. Standards and regression are medians of five measured passes; stress is a single pass per version. Earlier 1.9.0 runs on October 13 and 15, compared with 1.8.90, are kept in the full report.

This measures batch opening and saving, not rendering accuracy or PDF conformance.

Source: v1.9.0 benchmark report and measured runs.

KillerPDF 1.8.90 saved 40,563 PDFs across the three main collections. Every file kept the same outcome and diagnostic detail as 1.8.80. All five measured passes matched, and the 80 damaged inputs produced no crashes or timeouts.

The corpus run took place on a busy host, so its median times are higher than the October 4 record. A same-session check on a quiet host, with 1.8.80 and 1.8.90 alternating, measured 1.8.90 2.6% faster for regression and 2.1% faster for standards and color, with identical results. 1.8.90 has no timing regression.

This measures batch opening and saving, not rendering accuracy or PDF conformance.

Source: v1.8.90 benchmark report and measured runs.

KillerPDF 1.8.80 saved 40,563 PDFs across the three main collections. Every file kept the same outcome and diagnostic detail as 1.8.71. Both versions received one warmup and five measured passes per collection in alternating order. The 80 damaged inputs produced no crashes or timeouts.

Compared directly with 1.8.71, median times were 0.90% longer for regression, 0.35% longer for standards and color, and 0.59% shorter for stress. All changes are below the 10% investigation threshold.

This measures batch opening and saving, not rendering accuracy or PDF conformance.

Source: v1.8.80 benchmark report and measured runs.

KillerPDF 1.8.71 saved 40,563 PDFs across the three main collections. This baseline was measured alongside 1.8.80 on October 4, 2026. Both versions received one warmup and five measured passes per collection in alternating order, with identical per-file outcomes and diagnostic details. The 80 damaged inputs produced no crashes or timeouts.

The paired 1.8.80 run was 0.90% slower for regression, 0.35% slower for standards and color, and 0.59% faster for stress. The full report preserves the measurements for both versions.

This measures batch opening and saving, not rendering accuracy or PDF conformance.

Source: v1.8.71 benchmark report and measured runs.

KillerPDF 1.8.70 saved 40,563 PDFs across the three main collections. Every file kept the same outcome and diagnostic detail as 1.8.6. The warmup and all five measured passes matched on every path, status, diagnostic, and exit code. The 80 damaged inputs produced no crashes or timeouts.

Compared with 1.8.6, median times were 2.0% longer for regression, 1.0% longer for standards and color, and 3.9% shorter for stress. All changes are below the 10% investigation threshold.

This measures batch opening and saving, not rendering accuracy or PDF conformance.

Source: v1.8.70 benchmark report and measured runs.

KillerPDF 1.8.6 saved 40,563 PDFs across the three main collections. Every file kept the same outcome and diagnostic detail as 1.8.4. All five measured passes matched, and the 80 damaged inputs produced no crashes or timeouts.

Compared with 1.8.4, median times were 18.2% shorter for regression, 3.4% shorter for stress, and 1.9% shorter for standards and color.

This measures batch opening and saving, not rendering accuracy or PDF conformance.

Source: v1.8.6 benchmark report and measured runs.

The official KillerPDF 1.8.4 release payload saved 40,563 PDFs across the three main collections. Every file kept the same outcome and diagnostic detail as 1.8.3. All five measured passes matched, and the 80 damaged inputs produced no crashes or timeouts.

Compared with the separately recorded 1.8.3 session, median times were 13.7% longer for regression, 6.4% longer for stress, and 0.5% shorter for standards and color. The 1.8.3 payload was framework-dependent, while the official 1.8.4 portable release contains a self-contained .NET 10.0.11 payload, so the difference is not isolated to application code.

This measures batch opening and saving, not rendering accuracy or PDF conformance.

Source: v1.8.4 benchmark report and measured runs.

KillerPDF 1.8.3 saved 40,563 PDFs across the three main collections, 1,485 more than 1.8.2. No previously successful input became a skip or save failure. All five measured passes matched on every file's outcome and diagnostic detail. The 80 damaged inputs produced no crashes or timeouts.

Compared with the recorded 1.8.2 baseline, median times were 8.5% longer for regression, 2.0% longer for stress, and essentially unchanged for standards and color. KillerPDF 1.8.3 completed more successful saves; both versions were not rerun in alternating passes during the same session.

This measures batch opening and saving, not rendering accuracy or PDF conformance. Nineteen previously skipped inputs now open but fail save-time checks; the cards include those failures.

Source: v1.8.3 benchmark report and measured runs.

All five measured runs returned identical outcome totals. The test completed with zero application crashes or timeouts.

The first full 47,024-file baseline. All measured runs returned identical outcome totals, with zero application crashes or timeouts.

Source: machine-readable result summaries and measured runs.