999 sunny·Sep 23Long-term memory is compression, not retrievalEpistemic status: A design note. The motivating failure is from a long-form narrative agent I ran. One ICLR 2026 workshop paper supplies the measurement. Scope is at the end. Related: Chen, Epistemic8 min read·sunny99.hashnode.dev3M
CConnectedResearchers·Aug 9Phoenix Treats Refresh Drift as a Recoverable Runtime StatePhoenix breaks DDR5 TRR by detecting missed refreshes and repairing phase. A technical reading of the evidence, exploit, limits, and design lesson.7 min read·connectedresearchers.hashnode.dev
DDan Mercede·Jul 9Schema Is a Second SurfaceShort Essay · 2026-07-09 · 01:20 AM PT Claim: On a prerendered SPA, a copy reframe has to land on two independent surfaces, not one. The visible React copy auto-bakes into the crawler HTML. The head J2 min read·danmercede.hashnode.dev
DDan Mercede·Jul 7Vet the AI PR FleetShort Essay · 2026-07-07 · 02:35 PM PT Claim: Merging a queue of AI-authored pull requests is not a rubber stamp. Three failure modes hide in the queue, and each needs verify-before-you-trust, not bli2 min read·danmercede.hashnode.dev
DDan Mercede·Jul 6Recon vs Prior DecisionsShort Essay · 2026-07-06 · 12:20 AM PT Claim: A fresh recon-and-synthesis agent sees only live system state, not the decisions you already made about it, so it will confidently recommend something you2 min read·danmercede.hashnode.dev
DDan Mercede·Jul 5Cut Agent Web-Read TokensExperiment Log · 2026-07-05 · 03:40 PM PT Raw HTML can burn 80 percent of an agent's context. Extract-first fetching keeps only the signal. Hypothesis: Converting a page to markdown before the model1 min read·danmercede.hashnode.dev
DDan Mercede·Jul 5PR-Merge Train GotchasShort Essay · 2026-07-05 · 12:15 AM PT Claim: Merging a stack of PRs into one branch has host-side mechanics that silently skip half your queue if you script it naively. Merged fourteen PRs into one r3 min read·danmercede.hashnode.dev
DDan Mercede·Jul 4Benchmarks Evict Hot ModelsExperiment Log · 2026-07-04 · 12:15 AM PT Hypothesis: Benchmarking a locally-served LLM is read-only and won't disturb whatever model is already loaded on the GPU. Constraint: One 24GB GPU, a single ~1 min read·danmercede.hashnode.dev
DDan Mercede·Jul 3Don't Index Your System of RecordThought Snippet · 2026-07-03 · 08:40 AM PT Went to drop a PDF into an agent's knowledge vault today and hit the right instinct by accident: the vault's ingest crawler explicitly excludes its own direc1 min read·danmercede.hashnode.dev
DDan Mercede·Jul 1Recheck What You RejectedShort Essay · 2026-07-01 · 12:45 AM PT Claim: A negative decision stored in an agent's memory ('we evaluated X and rejected it') is the most dangerous kind of stale fact, because it reads as settled a2 min read·danmercede.hashnode.dev
