Verify it
yourself.
You should not have to take any of our numbers on faith. Every claim Pedulli makes is something you can check with tools you already trust: a SHA-256 hash, a byte comparison, your own factors or keys. Here is exactly how.
Claim 1 - Lossless compression
Confirm a byte-exact restore
"Lossless" has a precise meaning: the restored file is identical to the original, byte for byte. You do not need our word for it - you can prove it on your own machine in under a minute.
The four steps
- Compress your original file with Pedulli, producing a .pdli artifact.
- Restore the .pdli artifact back to a new output file.
- Take the SHA-256 of the original and of the restored output.
- Compare the two hashes. If they are identical, the restore is byte-exact. If a single byte differs, the hashes differ and you will see it.
A SHA-256 collision has never been produced. If the two hashes match, the files are the same bytes - not "close enough", but identical.
Run it on your own machine
# macOS / Linux shasum -a 256 original.bin pedulli compress original.bin -o original.pdli pedulli restore original.pdli -o restored.bin shasum -a 256 restored.bin # the two SHA-256 lines must be identical
# Windows (PowerShell) Get-FileHash original.bin -Algorithm SHA256 Get-FileHash restored.bin -Algorithm SHA256
The +1 byte never-worse floor
Pedulli is a best-of-N racer: it runs your data through several established codecs (including xz, zstd and brotli) alongside its own path, and keeps the smallest result. Because of that, Pedulli's own output carries a +1 byte never-worse floor - in the worst case it stores the input plus a one-byte header, so it never expands your data by more than a single byte.
Because it races xz, zstd, brotli and your data's SRD math and keeps the smallest verified output, Pedulli is never larger than the best standard codec - worst case +1 byte. It wins outright on structured data, and ties the best codec on already-optimal or random data (where it simply selects that codec). You can confirm exactly which candidate won on your data by reading the size of each output. No "superior to all", no "past the Shannon bound".
Claim 2 - The exact verifier
How the witness-gated verifier works
The verifier is deterministic and narrow. You supply a witness - the factors, the key and ciphertext, the assignment. The verifier checks that witness exactly against the original input. It verifies; it does not search, solve, crack or break anything.
The public shape
input P (the original problem/data) witness W (supplied by you) verify V(P, W) (deterministic, exact) -> true : accept W, with a reproducible check -> false : reject, or return a diagnostic
P is the original input. W is a candidate witness you provide. V is the exact verifier tied back to P. It never accepts anything it cannot check against the original.
What it does not do
It does not search for witnesses on its own, does not "solve" hard problems, and does not break or crack anything. In particular it does not break RSA, AES or ECC - it only re-checks a witness you already hold.
If no witness is supplied, or the check does not close, the correct output is a diagnostic or "no witness". That restraint is the product.
Roundtrip check
You supply a .pdli artifact. V restores it and checks that the SHA-256 of the restored bytes equals the SHA-256 of the original - byte-exact, or it fails.
Re-encryption match
You supply the key, IV and plaintext. V re-encrypts and checks the ciphertext and auth tag match the supplied ones. It confirms a supplied pair - it does not recover keys or break AES.
Integer check
You supply factors p and q. V does a big-integer multiply and checks p*q == N. It verifies factors you supply - it does not factor N or break RSA.
Supplied-clause check
You supply a candidate assignment. V evaluates the supplied clauses and reports satisfied / unsatisfied. It checks an assignment you provide; it is not a general SAT solver.
Public boundary: protected runtime + exact verifier + measured closure gain + diagnostic if not closed. Self-sealed - proof without disclosure, engine internals remain closed, no absolute-security claims.
Claim 3 - MDP equivalence
Reproduce the exact equivalence test
For a SRD compared against full a reference baseline, the result is exact value and policy equivalence across(verified configurations) 2026 Pedulli - Francesco Pedulli - Made in Italy