https://tegridydev.com/research/Writing by tegridydev: researchenhttps://tegridydev.com/research/transformers/weight-similarity-and-svd-compression/Compare SVD factor storage with held out output error on a GPT Neo projection, and see why weight similarity alone cannot establish useful compression.urn:tegridy:research/weight-similarity-and-svd-compressionhttps://tegridydev.com/research/transformers/temporal-attention-that-changes-predictions/Compare learned time decay, timestamp features and a last observation baseline across synthetic switching, periodic and shifted arrival tasks.urn:tegridy:research/temporal-attention-that-changes-predictionshttps://tegridydev.com/research/transformers/tegridydev-adaptive-neural-architecture-concepts/Explore memory, routing, repair and compression experiments with a clear distinction between local implementations, measured results and proposals.urn:tegridy:research/tegridydev-adaptive-neural-architecture-conceptshttps://tegridydev.com/research/transformers/sparse-feature-recovery/A synthetic sparse autoencoder comparison separates reconstruction error from feature recovery using cosine matching, support scores and correlated features.urn:tegridy:research/sparse-feature-recoveryhttps://tegridydev.com/research/transformers/self-healing-tokens/Compare learned token repair with majority repair, including recovery of corrupt tokens and harm to clean inputs in a controlled synthetic task.urn:tegridy:research/self-healing-tokenshttps://tegridydev.com/research/transformers/reflective-transformer-memory-and-adaptation/Test delayed memory retrieval, episode reset and batch isolation, with explicit limits on comparisons between memory systems of different capacities.urn:tegridy:research/reflective-transformer-memory-and-adaptationhttps://tegridydev.com/research/transformers/neuron-mapping-and-token-trajectories/Track individual token occurrences, model revisions and shared projection bases so activation maps remain comparable and their limits stay visible.urn:tegridy:research/neuron-mapping-and-token-trajectorieshttps://tegridydev.com/research/transformers/judge-head-attention/Compare contextual attention head gating with standard attention and a token MLP control, separating initial failures from a longer recall comparison.urn:tegridy:research/judge-head-attentionhttps://tegridydev.com/research/transformers/hydraform/Examine attention head mutation, optimiser replacement and checkpoint recovery alongside fixed and random mutation controls on a synthetic task.urn:tegridy:research/hydraformhttps://tegridydev.com/research/transformers/face-based-attention-circuits/Compare gated feature mixing with dense and static controls, including parameter costs and results that do not establish an advantage for the gate.urn:tegridy:research/face-based-attention-circuitshttps://tegridydev.com/research/transformers/bitnet-quantization-and-conversion/Separate float conversion, ternary simulation, packed weight storage and measured execution time, including the limits of a scalar Python decoder.urn:tegridy:research/bitnet-quantization-and-conversionhttps://tegridydev.com/research/transformers/arithmetic-across-notations/A protocol for comparing digit and number word arithmetic circuits, including the failed answer generation gate that blocks causal interpretation.urn:tegridy:research/arithmetic-across-notationshttps://tegridydev.com/research/training-and-evaluation/self-rewarding-training-loops/Check preference loss direction, answer masks and judge bias, and distinguish objective tests from a toy training run that failed to generalise.urn:tegridy:research/self-rewarding-training-loopshttps://tegridydev.com/research/training-and-evaluation/recursive-self-correction-with-uncertainty/Evaluate helpful and harmful answer revisions, frozen stopping rules and baseline competence before drawing conclusions about LLM self correction.urn:tegridy:research/recursive-self-correction-with-uncertaintyhttps://tegridydev.com/research/training-and-evaluation/glow-worm-byte-model/A small byte level language model tested on WikiText 2, with bits per byte, a unigram baseline and an explicit account of training exposure.urn:tegridy:research/glow-worm-byte-modelhttps://tegridydev.com/research/security/tegridydev-security-research-catalogue/Browse defensive research ideas, local implementations and unresolved sources, with practical next checks for incident review and model evaluation.urn:tegridy:research/tegridydev-security-research-cataloguehttps://tegridydev.com/research/security/sigint-spectrum-analysis-and-covert-channel-concepts/Compare synthetic signal detection and false alarms under session shift, including a detector whose ranking score hides a failed operating threshold.urn:tegridy:research/sigint-spectrum-analysis-and-covert-channel-conceptshttps://tegridydev.com/research/security/internet-protocol-amplification-research/Inspect byte, work, connection and time budgets in local protocol simulations, without treating bounded fixtures as evidence of live vulnerabilities.urn:tegridy:research/internet-protocol-amplification-researchhttps://tegridydev.com/research/mathematics/hyper-matrix-lattice-and-numerical-search/Separate multidimensional range summaries from sorted interval search, with exact overlap checks, scan baselines and construction costs.urn:tegridy:research/hyper-matrix-lattice-and-numerical-searchhttps://tegridydev.com/research/mathematics/four-dimensional-lattice/Compare sparse summary trees, scanning and spatial search on synthetic 4D queries, including construction costs and distribution dependent results.urn:tegridy:research/four-dimensional-latticehttps://tegridydev.com/research/knowledge/sl5-and-independent-research-workflows/Preserve nested quote attribution, source timestamps and thread context, while separating exact attribution tests from forecast scoring and retrieval quality.urn:tegridy:research/sl5-and-independent-research-workflowshttps://tegridydev.com/research/knowledge/open-abstract-knowledge-governance/Follow a claim correction without erasing its history, separating replay correctness, support labels and the rules for the default reading view.urn:tegridy:research/open-abstract-knowledge-governancehttps://tegridydev.com/research/knowledge/graph-memory-with-a-paper-trail/Compare graph and metadata retrieval on versioned sources, keeping source history visible and accounting for cases where the simpler baseline matches it.urn:tegridy:research/graph-memory-with-a-paper-trailhttps://tegridydev.com/research/knowledge/cloudvec-paper-search/Follow updates, deletion and crash recovery through a SQLite outbox, with freshness and replay checks kept separate from semantic search quality.urn:tegridy:research/cloudvec-paper-searchhttps://tegridydev.com/research/agents/swarm-and-collective-agent-concepts/Test neighbour coupling in a small finite grammar, with baseline coverage and clear boundaries around untested communication and routing proposals.urn:tegridy:research/swarm-and-collective-agent-conceptshttps://tegridydev.com/research/agents/moa-framework/Trace streamed output, cancellation and injected failures through a replayable orchestration fixture before evaluating real model answer quality.urn:tegridy:research/moa-frameworkhttps://tegridydev.com/research/agents/mixture-of-perspectives/Separate hard constraints from competing preferences and retain useful disagreement in a synthetic decision fixture with explicit utility assumptions.urn:tegridy:research/mixture-of-perspectiveshttps://tegridydev.com/research/agents/local-assistants-and-memory/Compare current and historical assistant memory using ownership, validity and recording time, with a fixed word budget and bounded synthetic results.urn:tegridy:research/local-assistants-and-memoryhttps://tegridydev.com/research/agents/autoresearch-web-researcher/Track exact citations through revised sources without rewriting earlier answers, and distinguish citation preservation from evidence supporting a claim.urn:tegridy:research/autoresearch-web-researcher