Evidence

12 merged changes. 5 open pull requests. 2 open reports.

Of the merged changes, 4 came from pull requests opened from Warren Smith’s account. Other developers made the other 8 after his reports, one as an opt-in mitigation. A merged fix is a maintainer’s review of one change, not an endorsement.

cvc5Released in 1.4.2
Details: cvc5, Released in 1.4.2
What was wrong
SMT solver returned a wrong “sat” on a separation-logic problem (reported by another user).
My part
Traced the root cause; fix submitted
Date
1 October 2026
Report
Original report (another user)
PyBaMMReleased in 26.9.0.0
Details: PyBaMM, Released in 26.9.0.0
What was wrong
BasicDFN with variable transference gained 2.44% of its lithium over one cycle.
My part
Measured and diagnosed; fix submitted (first raised by Nicola Courtier in #5700)
Date
18 September 2026
Report
My report
PyBaMMReleased in 26.9.0.0
Details: PyBaMM, Released in 26.9.0.0
What was wrong
Half-cell lithium diagnostic used the wrong length scale (factor 11.45).
My part
Reported; fixed by a maintainer
Date
22 September 2026
Report
My report
PULP common_cellsMerged
Details: PULP common_cells, Merged
What was wrong
ECC decoder output declared and never driven; first formal proof of the module.
My part
Found; fix submitted
Date
3 September 2026
pandapowerOpen
Details: pandapower, Open
What was wrong
Four unsupported transformer types returned converged=True with impossible results.
My part
Reported
Date
6 September 2026
PyTorchOpen
Details: PyTorch, Open
What was wrong
With deterministic algorithms enabled, a CUDA sparse (CSR) matrix-vector product still varied between runs, with no error or warning.
My part
Reported, in a comment on an issue opened by another user
Date
27 October 2026
Report
Issue #202030 (another user)
ChempropSubmitted, approved by a maintainer, not merged
Details: Chemprop, Submitted, approved by a maintainer, not merged
What was wrong
Three loss settings were silently ignored unless the loss function was also named.
My part
Investigated; fix submitted (pull request #1417)
Date
18 October 2026
Report
Original report (another user)
TorchIOSubmitted, not reviewed, not merged
Details: TorchIO, Submitted, not reviewed, not merged
What was wrong
In the v2 pre-release, the option setting the order of two image transforms was inverted. Stable 1.2.1 is not affected.
My part
Investigated; fix submitted (pull request #1536)
Date
18 October 2026
Report
Original report (another user)
QCoDeSSubmitted, approved by a maintainer, not merged
Details: QCoDeS, Submitted, approved by a maintainer, not merged
What was wrong
Fixed-width numpy integers made stepped instrument settings jump to the target or fail.
My part
Investigated; fix submitted (pull request #8627)
Date
18 October 2026
Report
Original report (another user)
CedarOpen pull request, not merged
Details: Cedar, Open pull request, not merged
What was wrong
A proposed test compares the typed expressions from the Rust and Lean typecheckers. No defect in either is established.
My part
Test proposed; a maintainer requested changes, and a response is posted
Date
24 August 2026
StrataOpen pull request, not merged
Details: Strata, Open pull request, not merged
What was wrong
In bug-finding mode, a correct program was reported as seven failed goals.
My part
Fix proposed; a maintainer asked for a different classification
Date
4 August 2026
Inspect AI (UK AI Security Institute)Merged
Details: Inspect AI (UK AI Security Institute), Merged
What was wrong
The same failing run reported 6 of 6 or 4 of 6 completed samples depending on scorer order.
My part
Reported; fixed by a contributor
Date
29 July 2026
Report
My report
inspect-robotsMerged
Details: inspect-robots, Merged
What was wrong
Run records did not include the grader configuration, so differently graded runs looked identical.
My part
Found; fix submitted
Date
21 September 2026
inspect-robotsMerged
Details: inspect-robots, Merged
What was wrong
Logs did not record which path produced each operator judgement.
My part
Reported; fixed by a maintainer
Date
31 August 2026
Report
My report
LemmaScriptMitigated (opt-in)
Details: LemmaScript, Mitigated (opt-in)
What was wrong
“1 verified, 0 errors” on code that called a random function, modelled as pure.
My part
Reported; the maintainer added an opt-in annotation (default unchanged)
Date
25 August 2026
Report
My report
con-leche (Lean)Merged
Details: con-leche (Lean), Merged
What was wrong
The checker command could exit successfully without running the checker.
My part
Reported; fixed by a maintainer
Date
10 September 2026
Report
My report
NVIDIA labs-OO-AgentsMerged
Details: NVIDIA labs-OO-Agents, Merged
What was wrong
Generator methods were wrapped as coroutines, so tracing spans ended before any work ran.
My part
Reported; fixed by a maintainer
Date
31 August 2026
Report
My report
NVIDIA SkillSpectorMerged
Details: NVIDIA SkillSpector, Merged
What was wrong
Every scan through one model provider failed on a schema the endpoint rejected.
My part
Reported; resolved in a maintainer’s release sync
Date
16 June 2026
Report
My report
OrbitMerged
Details: Orbit, Merged
What was wrong
A documented clean install failed at import: a dependency was undeclared.
My part
Reported; fixed by a maintainer
Date
17 September 2026
Report
My report

More evidence

About this record

Problems found in other people’s software, and what happened next. Solver, simulation and hardware items come first; the rest are in AI-evaluation and developer tooling.

The open pull requests are proposals, not established upstream fixes: two have a maintainer’s approval, and none is merged. Two reports remain open with no fix claimed. Authorship here means the public GitHub record, which does not itemise tool or AI assistance. All of it was self-initiated. Every row links to the public fix, pull request or issue; status was checked on 2 November 2026.

The longer OMEGA research programme, including formal-methods and AI-evaluation studies built for research rather than found in other people’s code, is in the research archive and the evidence ledger.