Skip to content

docs(evidence): N1 competitor gate — GO-CANDIDATE-N1 (3/4 missed by CA2000 AllPaths) - #417

Merged
PhysShell merged 2 commits into
mainfrom
research/n1-competitor-gate
Oct 20, 2026
Merged

PhysShell merged 2 commits into
mainfrom
research/n1-competitor-gate

Conversation

@PhysShell

Copy link
Copy Markdown
Owner

N1 competitor gate under GO-N1-COMPETITOR-GATE (0.25 agent-day; no third-party build, no package download).

Result: GO-CANDIDATE-N1, for the owner's decision; not a GO on a PoC. 3 of 4 confirmed N1 defects are missed by every configuration of CA2000 (default, AllPaths, both ownership-transfer switches, AllPathsOnlyNotDisposed), IDisposableAnalyzers 4.0.8 and Sonar 10.35.

Pair Verdict Acquisition form
whizbang#1204 CAUGHT (CA2000 AllPaths) new ServiceBusClient(…)
rockbot#638 MISSED await McpClient.CreateAsync(…): external async factory
NexusKit#34 MISSED services.BuildServiceProvider(…)
BareWire#73 MISSED await _connection.CreateChannelAsync(…): interface async factory
  • Pre-registration. d06caf75, committed before any analyzer run. The harness control is valid: AllPaths reports and the default does not.
  • Why they are missed. Probes with opaque throwing steps show CA2000 AllPaths tracks new, sync factories, stores, wrapping and async iterators. It does not track awaited external factories, interface-method results or BuildServiceProvider. Ablation finds no other feature that silences it, so the misses come from the real code's own acquisition API.
  • Disclosed. Run 0 hit a SARIF 1.0 reader bug and is not a result. Run 1 compiled the stubs inside the analysed compilation; the scored run 2 builds them as a separate assembly, and the verdicts are the same.
  • Limits. These are faithful reproductions with stub third-party types, not builds of the four repositories. interprocedural_analysis_kind was not varied. Own.NET today misses all three as well (Phase 0).

Research only, under docs/evidence/n1-competitor-gate-2026-10/.

🤖 Generated with Claude Code

https://claude.ai/code/session_01TFkNAFFQz9KoTzpPws8vYG

PhysShell and others added 2 commits October 20, 2026 21:13
Four confirmed N1 fixes (whizbang#1204, rockbot#638, NexusKit#34,
BareWire#73) as buggy/fixed reproductions with stubbed third-party
types; CA2000 (default, AllPaths, ownership-transfer variants,
AllPathsOnlyNotDisposed), IDisposableAnalyzers 4.0.8 and Sonar
10.35 with every rule on. GO-CANDIDATE only if >=2 pairs are missed
by every configuration; harness control must pass first.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TFkNAFFQz9KoTzpPws8vYG
…A2000 AllPaths)

Scored run (stubs as a separate assembly): CA2000 with AllPaths catches
whizbang#1204 (`new ServiceBusClient`) and misses rockbot#638 (awaited
external async factory), NexusKit#34 (BuildServiceProvider) and
BareWire#73 (awaited interface factory) in K1-K5; IDisposableAnalyzers
and Sonar report nothing on any acquisition line. Probes with opaque
throwing steps pin the limit to the acquisition form; ablation finds no
other silencing feature. Disclosed: SARIF 1.0 reader bug (run 0) and
in-compilation stubs (run 1, same verdicts).

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TFkNAFFQz9KoTzpPws8vYG
@PhysShell
PhysShell merged commit 38fb952 into main Oct 20, 2026
76 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant