https://ardana.ai/blog/ Notes from building Ardana and its System 1 models. en https://ardana.ai/blog/compare-supervised-local-text-classifiers/ https://ardana.ai/blog/compare-supervised-local-text-classifiers/ Thu, 22 Oct 2026 00:00:00 GMT Compare local decision models with TF IDF and frozen embedding classifiers using explicit label budgets, matched inputs and an untouched final test. https://ardana.ai/blog/detect-duplicate-tickets-local-models/ https://ardana.ai/blog/detect-duplicate-tickets-local-models/ Thu, 22 Oct 2026 00:00:00 GMT Separate candidate retrieval, same incident classification and cluster merging. Use a local decision model to score pairs while auditing false links. https://ardana.ai/blog/preserve-human-annotation-disagreement/ https://ardana.ai/blog/preserve-human-annotation-disagreement/ Thu, 22 Oct 2026 00:00:00 GMT Keep human judgments and evaluate local decision probabilities against a declared reviewer population. Separate majority accuracy from distribution distance. https://ardana.ai/blog/recalibrate-local-decision-probabilities/ https://ardana.ai/blog/recalibrate-local-decision-probabilities/ Thu, 22 Oct 2026 00:00:00 GMT Fit and freeze a task specific isotonic map for local decision probabilities. Preserve raw scores, save its contract and evaluate on independent labels. https://ardana.ai/blog/enforce-local-decision-constraints/ https://ardana.ai/blog/enforce-local-decision-constraints/ Wed, 21 Oct 2026 00:00:00 GMT Validate implications, exclusions and label counts after local classification. Preserve raw probabilities and send contradictory assignments for review. https://ardana.ai/blog/evaluate-agent-decision-workflows/ https://ardana.ai/blog/evaluate-agent-decision-workflows/ Wed, 21 Oct 2026 00:00:00 GMT Score complete agent runs by final state and required replies. Track recovery and stopping reasons, then distinguish repeated success from a lucky attempt. https://ardana.ai/blog/pairwise-ranking-local-decision-models/ https://ardana.ai/blog/pairwise-ranking-local-decision-models/ Wed, 21 Oct 2026 00:00:00 GMT Compare candidates in both orders with local decision questions. Preserve ties and disagreements, detect preference cycles and audit the final display order. https://ardana.ai/blog/cache-local-decision-results/ https://ardana.ai/blog/cache-local-decision-results/ Tue, 20 Oct 2026 00:00:00 GMT Build an exact cache around local decision probabilities. Preserve request semantics, version model inputs, expire stale entries and recompute actions. https://ardana.ai/blog/hierarchical-local-decision-classification/ https://ardana.ai/blog/hierarchical-local-decision-classification/ Tue, 20 Oct 2026 00:00:00 GMT Build a caller owned label hierarchy with local decision questions. Trace parent and child choices, validate paths and interpret branch probabilities. https://ardana.ai/blog/sample-local-decision-review-queues/ https://ardana.ai/blog/sample-local-decision-review-queues/ Tue, 20 Oct 2026 00:00:00 GMT Build a review queue from local decision probabilities while keeping a separate random audit sample. Record selection reasons and protect evaluation. https://ardana.ai/blog/weak-supervision-local-decision-labels/ https://ardana.ai/blog/weak-supervision-local-decision-labels/ Tue, 20 Oct 2026 00:00:00 GMT Create candidate labels from fallible rules with explicit abstention and conflicts. Inspect vote coverage and keep weak labels separate from evaluation. https://ardana.ai/blog/write-decision-label-descriptions/ https://ardana.ai/blog/write-decision-label-descriptions/ Tue, 20 Oct 2026 00:00:00 GMT Treat classifier label meanings as a versioned contract. Write option descriptions, preserve dispatch IDs and test label splits with fresh model requests. https://ardana.ai/blog/conformal-sets-local-decision-models/ https://ardana.ai/blog/conformal-sets-local-decision-models/ Mon, 19 Oct 2026 00:00:00 GMT Turn local choice probabilities into conformal label sets. Implement the finite sample rank and understand coverage, empty sets and exchangeability. https://ardana.ai/blog/select-few-shot-decision-examples/ https://ardana.ai/blog/select-few-shot-decision-examples/ Mon, 19 Oct 2026 00:00:00 GMT Compare a baseline, selected examples and reordered demonstrations with fixed routing labels. Keep test labels outside prompts and report failures. https://ardana.ai/blog/shadow-test-local-decision-upgrades/ https://ardana.ai/blog/shadow-test-local-decision-upgrades/ Mon, 19 Oct 2026 00:00:00 GMT Compare a candidate with the current decision model on matching requests. Keep its actions separate, count missing pairs and set promotion criteria. https://ardana.ai/blog/test-bundled-decision-question-interference/ https://ardana.ai/blog/test-bundled-decision-question-interference/ Mon, 19 Oct 2026 00:00:00 GMT Trace how Ardana isolates question rows, then compare bundled and single question probabilities with stable IDs, fixed options and explicit tolerances. https://ardana.ai/blog/test-decision-evaluator-contracts/ https://ardana.ai/blog/test-decision-evaluator-contracts/ Mon, 19 Oct 2026 00:00:00 GMT Check question and option mapping, probability losses and failed-answer denominators with authored controls before comparing local decision models. https://ardana.ai/blog/benchmark-local-decision-api-latency/ https://ardana.ai/blog/benchmark-local-decision-api-latency/ Sun, 18 Oct 2026 00:00:00 GMT Separate cold loads from warm requests, measure complete decision responses, and report queueing, tail latency and successful throughput under load. https://ardana.ai/blog/cost-sensitive-local-decision-policies/ https://ardana.ai/blog/cost-sensitive-local-decision-policies/ Sun, 18 Oct 2026 00:00:00 GMT Turn local model probabilities into actions with explicit error costs, derive a binary threshold, and compare multiclass policies on held out data. https://ardana.ai/blog/evaluate-rare-event-decision-models/ https://ardana.ai/blog/evaluate-rare-event-decision-models/ Sun, 18 Oct 2026 00:00:00 GMT Read precision and recall at realistic event prevalence, calculate review volume, and evaluate average precision without hiding rare event failures. https://ardana.ai/blog/local-decision-context-token-budgets/ https://ardana.ai/blog/local-decision-context-token-budgets/ Sat, 17 Oct 2026 00:00:00 GMT Measure rendered decision requests, distinguish reported usage from row limits, and test input length before sending long states to a local model. https://ardana.ai/blog/minimize-private-local-decision-state/ https://ardana.ai/blog/minimize-private-local-decision-state/ Sat, 17 Oct 2026 00:00:00 GMT Choose the fields a local model needs, trace downloads and request handling, and test whether redaction preserves the decision before sharing inputs. https://ardana.ai/blog/monitor-decision-probability-calibration-drift/ https://ardana.ai/blog/monitor-decision-probability-calibration-drift/ Sat, 17 Oct 2026 00:00:00 GMT Compare mature labels and stable score groups to monitor calibration after rollout, separate traffic shifts, and decide when thresholds need review. https://ardana.ai/blog/agent-tool-authorization-boundaries/ https://ardana.ai/blog/agent-tool-authorization-boundaries/ Fri, 16 Oct 2026 00:00:00 GMT Let a local model choose an action, then check identity, resource access and exact arguments in application code before executing the tool. https://ardana.ai/blog/handle-local-inference-failures/ https://ardana.ai/blog/handle-local-inference-failures/ Fri, 16 Oct 2026 00:00:00 GMT Separate uncertain answers from failed inference. Handle validation errors, queue overload and runtime failures before applying any action threshold. https://ardana.ai/blog/multi-label-classification-local-models/ https://ardana.ai/blog/multi-label-classification-local-models/ Fri, 16 Oct 2026 00:00:00 GMT Use separate yes or no questions for overlapping text labels. Read each probability, set per label policies and avoid inventing a joint score. https://ardana.ai/blog/ordinal-scores-decision-models/ https://ardana.ai/blog/ordinal-scores-decision-models/ Fri, 16 Oct 2026 00:00:00 GMT Read ordinal decision scores as expected level indices. Inspect the distribution, level fits and rubric before mapping a score to an action. https://ardana.ai/blog/reproduce-local-model-decisions/ https://ardana.ai/blog/reproduce-local-model-decisions/ Fri, 16 Oct 2026 00:00:00 GMT Preserve requests, model revisions, quantization and runtime settings before an upgrade. Separate a useful audit record from identical inference. https://ardana.ai/blog/rerank-retrieval-local-decision-model/ https://ardana.ai/blog/rerank-retrieval-local-decision-model/ Fri, 16 Oct 2026 00:00:00 GMT Score query and document pairs through a local decision API, preserve candidate IDs and compare the resulting order with retrieval and a reranker. https://ardana.ai/blog/classifier-out-of-scope-requests/ https://ardana.ai/blog/classifier-out-of-scope-requests/ Wed, 14 Oct 2026 00:00:00 GMT Define unsupported and ambiguous requests separately, give each an explicit outcome, and test whether your classifier sends unknown requests to a handler. https://ardana.ai/blog/evaluate-prompt-injection-detectors/ https://ardana.ai/blog/evaluate-prompt-injection-detectors/ Wed, 14 Oct 2026 00:00:00 GMT Measure prompt injection detectors against harmless inputs, freeze thresholds before the final test, and check what happens to the protected agent. https://ardana.ai/blog/missing-information-ai-decisions/ https://ardana.ai/blog/missing-information-ai-decisions/ Wed, 14 Oct 2026 00:00:00 GMT Separate model mistakes, missing evidence and uncertain labels. Plan a controlled input audit before assuming a larger model will fix a decision. https://ardana.ai/blog/test-changing-decision-rules/ https://ardana.ai/blog/test-changing-decision-rules/ Wed, 14 Oct 2026 00:00:00 GMT Build paired tests that change a rule or a relevant fact, check both expected answers, and distinguish correct adaptation from a model that merely flips. https://ardana.ai/blog/test-classifiers-across-languages/ https://ardana.ai/blog/test-classifiers-across-languages/ Wed, 14 Oct 2026 00:00:00 GMT Evaluate classifiers with native, translated and mixed language inputs. Keep label meanings stable and compare results within each task and language. https://ardana.ai/blog/trustworthy-classification-test-sets/ https://ardana.ai/blog/trustworthy-classification-test-sets/ Wed, 14 Oct 2026 00:00:00 GMT Build classification evaluations with traceable labels, grouped splits and leakage checks. Keep model inputs separate from the answers used to grade them. https://ardana.ai/blog/gemma-kv-cache-memory/ https://ardana.ai/blog/gemma-kv-cache-memory/ Tue, 13 Oct 2026 00:00:00 GMT Ardana v0.3.2 fixes an oversized Gemma 4 cache allocation. Model weights alone cannot tell us how much memory a run needs. https://ardana.ai/blog/run-safetensors-models-locally/ https://ardana.ai/blog/run-safetensors-models-locally/ Tue, 13 Oct 2026 00:00:00 GMT Pull supported Hugging Face safetensors checkpoints into Ardana, run typed decisions locally, and understand where GGUF still fits. https://ardana.ai/blog/thx-01-local-decision-model/ https://ardana.ai/blog/thx-01-local-decision-model/ Tue, 13 Oct 2026 00:00:00 GMT Run THX-01 through Ardana's CLI and HTTP API. See how its encoder answers typed questions and where its token and option limits apply. https://ardana.ai/blog/decider-4b-quantization-jevbench/ https://ardana.ai/blog/decider-4b-quantization-jevbench/ Sat, 10 Oct 2026 00:00:00 GMT We ran decider:4b at q4_k_m, q8_0 and bf16 on JevBench's 231 public items. q8_0 is six items ahead, the 2B gap is reversed, and neither clears the noise. https://ardana.ai/blog/mixture-of-experts-decision-models/ https://ardana.ai/blog/mixture-of-experts-decision-models/ Sat, 10 Oct 2026 00:00:00 GMT Two mixture-of-experts models on JevBench's 231 public items through ardana: what 3 to 4 billion active parameters decide and what the file costs in RAM. https://ardana.ai/blog/decision-model-in-browser-tab/ https://ardana.ai/blog/decision-model-in-browser-tab/ Fri, 09 Oct 2026 00:00:00 GMT What ardana.ai/playground downloads into your tab, how onnxruntime-web runs decider:0.8b on WebGPU or WASM, and why the tab answers exactly like the server. https://ardana.ai/blog/instruct-model-versus-decision-model/ https://ardana.ai/blog/instruct-model-versus-decision-model/ Fri, 09 Oct 2026 00:00:00 GMT We sent the same typed questions to gemma-4, qwen3.5 and decider through ardana, shuffled the options, and measured where a zero-shot instruct model breaks. https://ardana.ai/blog/local-decision-model-cascade/ https://ardana.ai/blog/local-decision-model-cascade/ Fri, 09 Oct 2026 00:00:00 GMT How to put ardana serve and decider:2b in front of a chat model, which returned field to threshold, and how to check the threshold on your own labelled tickets. https://ardana.ai/blog/quantization-calibration-decider-2b/ https://ardana.ai/blog/quantization-calibration-decider-2b/ Fri, 09 Oct 2026 00:00:00 GMT We ran JevBench's public items against decider:2b at q4_k_m, q8_0 and bf16 through ardana serve on a CPU and compared accuracy, Brier score and calibration. https://ardana.ai/blog/running-decision-models-locally/ https://ardana.ai/blog/running-decision-models-locally/ Fri, 09 Oct 2026 00:00:00 GMT A support ticket through ardana run, the documented requests for Ollama 0.35 and the llama.cpp server, what each returns, and which models each one can run. https://ardana.ai/blog/decision-model-answers-with-probabilities/ https://ardana.ai/blog/decision-model-answers-with-probabilities/ Thu, 08 Oct 2026 00:00:00 GMT How Ardana's choice, score and noul questions turn a support ticket into a probability per option, and how an agent thresholds that instead of parsing text.