https://hermanity.dev/tags/reliability/Recent content in Reliability on HermanHugoen-usThu, 16 Jul 2026 00:00:00 +0000https://hermanity.dev/projects/breakage-atlas/Thu, 16 Jul 2026 00:00:00 +0000https://hermanity.dev/projects/breakage-atlas/<h2 id="overview">Overview</h2> <p>The <strong>Skynet Breakage Atlas</strong> answers a practical routing question: when traffic goes through our proxy fleet instead of first-party subscription surfaces, which models go empty, hard-error, or slow-walk under adversarial families?</p> <p>Live at <a href="https://breakage.hermanity.dev/">breakage.hermanity.dev</a>.</p> <h2 id="whats-measured">What’s measured</h2> <ul> <li><strong>Controls:</strong> Claude Max, Codex, Grok (subscription / direct)</li> <li><strong>Fleet:</strong> Skynet-routed models (deepseek, glm, kimi, minimax, …)</li> <li><strong>Families:</strong> cascade seed, context boundary, cost threshold, empty probe, format JSON, multi-intent</li> <li><strong>Metrics:</strong> flag rate, empty rate, hard error rate, p50 latency</li> </ul> <h2 id="snapshot-public-table-on-the-site">Snapshot (public table on the site)</h2> <p>glm-5.2 is the known-broken reference (flag rate 1.0 across cells in the frozen N-batch). Controls stay low-flag; fleet variance is the point of the atlas — empty replies are a first-class failure mode, not a footnote.</p>