These are internal engineering directions Pedulli is exploring. They are listed here for transparency, not as achievements. None of them is a published result. Nothing on this page is claimed as a win, a record, or a solved problem.
The standard for moving anything off this page
A research track only graduates to our benchmarks or verified results when it is backed by one of:
Until then, a measured-but-unreproducible number stays here as an internal direction. We do not headline it, we do not call it a record, and we do not imply it generalises.
Open directions
A verifier-first direction in constraint satisfaction and formal methods. We do not claim independent SAT solving, and we make no comparison claim against established solvers. The published surface stays witness-checking only - we verify supplied solutions, we do not assert we solve.
Exploratory work on verifier-gated proof checking and formal-methods orchestration. The honest boundary is the same as everywhere else on the site: an exact checker that confirms supplied witnesses, not a system that decides open problems.
A retrieval/embedding acceleration direction combining a coarse pass with an exact re-score, alongside byte-exact index compression. Measured so far only on self-generated synthetic corpora; recall is reported as measured, with no "beats X" claim and no customer result behind it.
The exact value/policy equivalence under group symmetry is a verified result (see verified results). Any associated timing advantage remains a research direction: we describe it only as an asymptotic reduction under verification, and publish no wall-clock speedup figure until a reproducible timing harness on a non-trivial symmetry group ships.
Internal experiments in program transformation and structure-aware optimization. No public benchmark, no comparison claim, and nothing presented as a delivered capability.
Exploratory work on recovering structure from binary and structured inputs, framed strictly as research. It does not break, crack, or solve cryptography; the verifier only checks witnesses you supply.
On highly structured or periodic synthetic inputs the compressor can reach very large ratios. These are class-specific examples on data the method is designed to favour - they are not general-purpose and are never presented as a headline ratio. Because the Orchestrator is a best-of-N racer that includes xz, zstd and brotli and keeps the smallest verified output, it 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/random data. Real-world results are reported separately on the benchmarks page.
An honest, no-lookahead backtesting harness - costs included, baselines included, evaluated on data the system never saw. We ship this as an integrity / methodology artifact. It carries no performance claim: no returns, no Sharpe, no ROI.
Competitions
We publish a competition outcome only when it is actually won and backed by a public leaderboard link - never a projected or unverified claim. We currently make no competition-win claim of any kind. See our proof-gated competitions policy.
For what is already substantiated and reproducible, see the verified results and the public benchmarks.