@base . @prefix : . @prefix schema: . @prefix xsd: . @prefix rdf: . @prefix rdfs: . @prefix owl: . @prefix skos: . @prefix prov: . @prefix dbr: . <> a schema:Article ; schema:name "Jev & Beyond Human-in-the-Loop AI"@en ; schema:headline "Beyond Human-in-the-Loop AI"@en ; schema:abstract "A newsletter article on the shift from conversational AI to decision-native AI. TypeSafe's Jev returns structured decisions and calibrated probabilities that software can act on directly, moving human oversight from checking every decision toward designing, measuring, and governing the system that makes them. The article argues that preference, correctness, and calibration are three distinct objectives, and that reliable automation depends on the harness around the model rather than on the model alone."@en ; schema:description "The Business Engineer newsletter on Jev, TypeSafe AI's decision model, and what changes before a specific decision no longer needs a person to approve it every time."@en ; schema:articleSection "AI, automation, calibration, agentic systems"@en ; schema:genre "newsletter analysis"@en ; schema:inLanguage "en"@en ; schema:datePublished "2026-09-19"^^xsd:date ; schema:dateCreated "2026-09-19T11:59:31Z"^^xsd:dateTime ; schema:dateModified "2026-09-19T11:59:31Z"^^xsd:dateTime ; schema:url ; schema:author :gennaroCuofano ; schema:publisher :theBusinessEngineer ; schema:isPartOf :theBusinessEngineerSeries ; schema:about :jev, :typesafeAI, :rlhf, :rlcd, :calibration, :preferenceObjective, :correctnessObjective, :assistance, :automation, :harness, :metaHarness, :forcedLoop, :chosenLoop, :operationalLoop, :measurementLoop ; schema:hasPart :secWhereTheLoopCameFrom, :secHumanAsErrorCorrection, :secThreeObjectives, :secWhatCalibrationGivesYou, :secWhatJevIsTryingToBuild, :secProductClaimAndEvidence, :secLimitationsThatMatter, :secFromProbabilityToDecisionRule, :secWhyCalibrationDoesNotRemoveSystem, :secClosingTheLoopThroughMeasurement, :secWhatRepricesWhenReviewSelective, :secScalingRelay, :secIsThisTheEndOfRLHF, :secHumanLoopThatStays, :secWhatSeriousDeploymentWouldTest, :secMentalModels, :secNextQuestion, :faqPage, :glossary, :howto ; schema:accountablePerson ; prov:wasGeneratedBy , , :deepseekV4Pro ; schema:about , . # ── Curation provenance (skills + LLM acting on behalf of the principal) ────── :deepseekV4Pro a schema:SoftwareApplication, prov:SoftwareAgent ; schema:name "DeepSeek V4 Pro"@en ; schema:url ; prov:actedOnBehalfOf . a schema:SoftwareApplication, prov:SoftwareAgent ; schema:name "kg-generator"@en ; schema:url ; prov:actedOnBehalfOf . a schema:SoftwareApplication, prov:SoftwareAgent ; schema:name "rdf-infographic-skill"@en ; schema:url ; prov:actedOnBehalfOf . # ── People ───────────────────────────────────────────────────────────────────── :gennaroCuofano a schema:Person ; schema:name "Gennaro Cuofano"@en ; schema:description "Author of The Business Engineer newsletter and founder of FourWeekMBA. He covers the business and strategic implications of AI, writing on what is important for where the industry is moving rather than the loudest news of the week."@en ; schema:email "thebusinessengineer@substack.com"@en ; schema:worksFor :theBusinessEngineer ; schema:author :theBusinessEngineerSeries . :diogoAlmeida a schema:Person ; schema:name "Diogo Almeida"@en ; schema:jobTitle "Founder, TypeSafe AI"@en ; schema:description "Founder of TypeSafe AI and a co-author of the InstructGPT paper. He leads the company's effort to build Jev as an intelligence primitive for software rather than another chatbot."@en ; schema:worksFor :typesafeAI ; schema:creator :typesafeAI . # ── Organizations ────────────────────────────────────────────────────────────── :theBusinessEngineer a schema:Organization ; schema:name "The Business Engineer"@en ; schema:description "A Substack publication by Gennaro Cuofano covering the business, strategy, and economic implications of AI, including the Agenting platform and the Genesis tool."@en ; schema:url ; schema:email "thebusinessengineer@substack.com"@en ; schema:identifier "594665"@en ; schema:founder :gennaroCuofano . :theBusinessEngineerSeries a schema:CreativeWorkSeries ; schema:name "The Business Engineer"@en ; schema:description "The ongoing newsletter series by Gennaro Cuofano analyzing the business and strategic shape of AI."@en ; schema:publisher :theBusinessEngineer ; schema:url . :typesafeAI a schema:Organization ; schema:name "TypeSafe AI"@en ; schema:description "The company behind Jev, a decision model designed to return structured outputs and calibrated probabilities that software can consume directly. TypeSafe emerged from stealth in September 2026 with $40M in funding."@en ; schema:founder :diogoAlmeida ; schema:about :jev, :rlcd . :openai a schema:Organization ; schema:name "OpenAI"@en ; schema:description "The AI research and deployment company that released ChatGPT, InstructGPT, and the GPT-3.5 and GPT-4o models discussed in the article."@en ; owl:sameAs . # ── Software and Products ────────────────────────────────────────────────────── :jev a schema:SoftwareApplication ; schema:name "Jev"@en ; schema:description "TypeSafe AI's decision model, introduced in early access on September 15, 2026. Rather than free-form text, Jev returns structured decisions and probabilities that software can consume directly, using three primitives — Choice, Noul, and Score — trained via Reinforcement Learning for Calibrated Decisions (RLCD)."@en ; schema:creator :typesafeAI ; schema:provider :typesafeAI ; schema:about :choice, :noul, :score, :rlcd ; schema:applicationCategory "Decision Model"@en ; schema:releaseDate "2026-09-15"^^xsd:date . :chatgpt a schema:SoftwareApplication ; schema:name "ChatGPT"@en ; schema:description "OpenAI's conversational assistant launched in November 2022, built as a sibling of InstructGPT with RLHF-based training that made model intelligence approachable through natural interaction."@en ; schema:creator :openai ; owl:sameAs . :gpt35 a schema:SoftwareApplication ; schema:name "GPT-3.5"@en ; schema:description "The OpenAI language model underlying the original ChatGPT, whose November 2022 launch is the starting point of the article's narrative."@en ; schema:creator :openai . :gpt4o a schema:SoftwareApplication ; schema:name "GPT-4o"@en ; schema:description "The OpenAI model whose 2025 sycophancy problem is cited as evidence that poorly balanced human feedback can reward agreement over accuracy."@en ; schema:creator :openai . :instructgpt a schema:SoftwareApplication ; schema:name "InstructGPT"@en ; schema:description "The OpenAI model family that introduced reinforcement learning from human feedback (RLHF) to close the gap between predicting text and following instructions; ChatGPT was its sibling."@en ; schema:creator :openai ; schema:about :rlhf . :openclaw a schema:SoftwareApplication ; schema:name "OpenClaw"@en ; schema:description "Described as one of the defining product signals of 2026 — persistent AI that can use tools, hold context, and keep working beyond the chat window, making the direction toward agents obvious."@en . :genesis a schema:SoftwareApplication ; schema:name "Genesis"@en ; schema:description "A free AI tool from The Business Engineer that lets users experience Jev's speed firsthand."@en ; schema:creator :theBusinessEngineer ; schema:url . :agentingPlatform a schema:WebApplication ; schema:name "The Business Engineer's Agenting Platform"@en ; schema:description "The platform where Jev is made available to executive members of The Business Engineer."@en ; schema:creator :theBusinessEngineer ; schema:url . # ── Core Concepts (DefinedTerms) ─────────────────────────────────────────────── :rlhf a skos:Concept ; schema:name "Reinforcement Learning from Human Feedback (RLHF)"@en ; schema:description "A training method that combines supervised examples with reinforcement learning guided by human rankings of responses. It helped turn a model trained to continue text into an assistant better able to follow instructions, but optimizing for approval and optimizing for truth are not always the same objective."@en . :rlcd a skos:Concept ; schema:name "Reinforcement Learning for Calibrated Decisions (RLCD)"@en ; schema:description "TypeSafe's stated training approach for Jev, designed around one idea: produce probabilities that software can use directly, making the numbers attached to decisions reflect how often those decisions are right."@en . :calibration a skos:Concept ; schema:name "Calibration"@en ; schema:description "The property that a model's stated confidence matches how often it is actually right — when the model says 90% confident, it is right about 90% of the time. Calibration makes uncertainty measurable enough for a workflow to act on."@en . :preferenceObjective a skos:Concept ; schema:name "Preference"@en ; schema:description "One of three training objectives: which response a person would prefer. Useful for instruction-following, tone, and interaction quality, but a preferred answer is not necessarily a true one."@en . :correctnessObjective a skos:Concept ; schema:name "Correctness"@en ; schema:description "One of three training objectives: did the system get the answer right, especially where the result can be checked directly, as in mathematics, code, or other tasks with clear verifiers."@en . :sycophancy a skos:Concept ; schema:name "Sycophancy"@en ; schema:description "A failure mode where a model becomes excessively agreeable, favoring answers that sound convincing or align with user expectations over answers that are actually correct — a risk of poorly balanced human-feedback training."@en . :harness a skos:Concept ; schema:name "Harness"@en ; schema:description "The operating layer around a model that supplies authoritative context, permissions, allowed actions, approval rules, failure handling, recovery, and a record of what happened. The model makes the judgment; the harness decides whether that judgment can become an action."@en . :metaHarness a skos:Concept ; schema:name "Meta-harness"@en ; schema:description "Not just one agent, but the layer that coordinates many agents, models, tools, permissions, context, and execution environments — the product that makes agents useful together."@en . :assistance a skos:Concept ; schema:name "Assistance"@en ; schema:description "A mode where the machine accelerates a person while the person still supplies the final guarantee; the human remains responsible for catching mistakes before they matter."@en . :automation a skos:Concept ; schema:name "Automation"@en ; schema:description "A mode where selected decisions are allowed to proceed without case-by-case human check, on the basis that the remaining risk is understood, bounded, and acceptable for that specific decision."@en . :forcedLoop a skos:Concept ; schema:name "Forced Loop"@en ; schema:description "A human checkpoint that exists because the system cannot yet be trusted to act reliably on its own; it should shrink as the evidence improves."@en . :chosenLoop a skos:Concept ; schema:name "Chosen Loop"@en ; schema:description "A human checkpoint that remains because the decision requires judgment, accountability, values, or explicit authority — even if the model could predict the likely answer with high accuracy."@en . :operationalLoop a skos:Concept ; schema:name "Operational Loop"@en ; schema:description "The loop that handles an individual decision: the model makes a prediction, the system applies policy, executes an allowed action, and checks what happened next."@en . :measurementLoop a skos:Concept ; schema:name "Measurement Loop"@en ; schema:description "The loop that watches the system over time: it samples decisions, collects outcomes, measures calibration and error rates, looks for drift, and changes thresholds, policies, or models when performance deteriorates."@en . :seatAnchor a skos:Concept ; schema:name "Seat Anchor"@en ; schema:description "The assumption in per-seat pricing that value scales with the number of humans using the software; as more work runs without direct human interaction, the economically relevant unit shifts toward decisions, outcomes, or completed work."@en . # ── Jev Primitives ───────────────────────────────────────────────────────────── :choice a skos:Concept ; schema:name "Choice"@en ; schema:description "A Jev primitive that selects among defined alternatives, returning a probability distribution over the allowed options."@en . :noul a skos:Concept ; schema:name "Noul"@en ; schema:description "A Jev primitive that returns the probability of a yes-or-no proposition."@en . :score a skos:Concept ; schema:name "Score"@en ; schema:description "A Jev primitive that evaluates something against an ordered scale."@en . # ── Article Sections ─────────────────────────────────────────────────────────── :secWhereTheLoopCameFrom a schema:CreativeWork ; schema:name "Where the Loop Came From"@en ; schema:abstract "RLHF and InstructGPT closed the gap between predicting text and following instructions, but optimizing for approval and optimizing for truth are not the same objective. A human preference tells us which answer a person likes, not which is factually correct."@en ; schema:isPartOf <> . :secHumanAsErrorCorrection a schema:CreativeWork ; schema:name "The Human as Error Correction, and as a Cost"@en ; schema:abstract "When every case still requires human review, every case still carries a human cost. This is the dividing line between assistance, where the human catches mistakes, and automation, where selected decisions proceed without case-by-case check."@en ; schema:isPartOf <> . :secThreeObjectives a schema:CreativeWork ; schema:name "Three Objectives, Not Three Mutually Exclusive Machines"@en ; schema:abstract "Preference, correctness, and calibration answer three different questions, and a strong AI system may need all three. Accuracy alone does not solve automation because a model can be accurate on average yet dangerous when it does not know when it is likely to be wrong."@en ; schema:isPartOf <> . :secWhatCalibrationGivesYou a schema:CreativeWork ; schema:name "What Calibration Actually Gives You"@en ; schema:abstract "Calibration makes a probability mean something, but it is not accuracy and it works across groups of predictions, not individual cases. The goal is useful predictions with justified confidence, so uncertainty becomes something the workflow can use."@en ; schema:isPartOf <> . :secWhatJevIsTryingToBuild a schema:CreativeWork ; schema:name "What Jev Is Trying to Build"@en ; schema:abstract "TypeSafe introduced Jev in early access on September 15, 2026, founded by Diogo Almeida, a co-author of the InstructGPT paper. Jev returns structured outputs via three primitives — Choice, Noul, and Score — and is best understood as a decision component inside a larger agentic system."@en ; schema:isPartOf <> . :secProductClaimAndEvidence a schema:CreativeWork ; schema:name "The Product Claim and the Evidence Are Different Things"@en ; schema:abstract "Evidence splits into three buckets: what TypeSafe claims, what outside testing observes, and architectural speculation. Early calibration results are encouraging but do not yet prove the category on enterprise workflows."@en ; schema:isPartOf <> . :secLimitationsThatMatter a schema:CreativeWork ; schema:name "The Limitations That Matter in Production"@en ; schema:abstract "Option order, wording, and rounding can move probabilities enough to change whether a case is automated or reviewed. Calibration is local to the environment where it was measured, and separate decisions can still fail together."@en ; schema:isPartOf <> . :secFromProbabilityToDecisionRule a schema:CreativeWork ; schema:name "From a Probability to a Decision Rule"@en ; schema:abstract "A probability says how likely something is; a decision rule says what to do with it. The most likely answer is not always the economically correct action, and there is no universal confidence threshold for automation."@en ; schema:isPartOf <> . :secWhyCalibrationDoesNotRemoveSystem a schema:CreativeWork ; schema:name "Why Calibration Does Not Remove the Need for a System"@en ; schema:abstract "A trustworthy probability is not enough to run a business process, and reliability compounds across steps — ten decisions each 99% reliable yield only about 90.4% end-to-end. A reliable component is valuable; a reliable workflow still has to be engineered."@en ; schema:isPartOf <> . :secClosingTheLoopThroughMeasurement a schema:CreativeWork ; schema:name "Closing the Loop Through Measurement"@en ; schema:abstract "The strongest architecture has two loops — an operational loop for each decision and a measurement loop over time. The human moves from reviewing every decision to maintaining the evidence that allows some decisions to run without them."@en ; schema:isPartOf <> . :secWhatRepricesWhenReviewSelective a schema:CreativeWork ; schema:name "What Reprices When Review Becomes Selective"@en ; schema:abstract "Product, governance, and work all reprice once review becomes selective: value shifts from assistance toward completed decisions, oversight toward evidence, and work toward exceptions and system design."@en ; schema:isPartOf <> . :secScalingRelay a schema:CreativeWork ; schema:name "The Scaling Relay"@en ; schema:abstract "This fits the broader AI Supercycle as a rotating bottleneck: once raw capability becomes good enough, delegation, reliability, and economics become the constraint that matters next."@en ; schema:isPartOf <> . :secIsThisTheEndOfRLHF a schema:CreativeWork ; schema:name "Is This the End of RLHF?"@en ; schema:abstract "RLHF is not over; the narrower claim is that preference optimization alone is not enough for every automation problem. The real experiment is comparative — whether decision-native systems automate more work, more reliably, at lower total cost."@en ; schema:isPartOf <> . :secHumanLoopThatStays a schema:CreativeWork ; schema:name "The Human Loop That Stays"@en ; schema:abstract "Some human checkpoints exist because the system is not reliable enough yet (forced loops); others exist because the decision requires judgment, accountability, values, or authority (chosen loops). The goal is not zero humans — it is zero unnecessary review."@en ; schema:isPartOf <> . :secWhatSeriousDeploymentWouldTest a schema:CreativeWork ; schema:name "What a Serious Deployment Would Test"@en ; schema:abstract "Start with one bounded decision with an observable outcome, then test prediction, policy, workflow, unattended execution, and economics. A deployment earns automation when a defined slice of work runs with less human effort, acceptable outcomes, and risk inside an agreed boundary."@en ; schema:isPartOf <> . :secMentalModels a schema:CreativeWork ; schema:name "The Mental Models"@en ; schema:abstract "The article rests on a small set of portable ideas — the loop as objective, synchronous versus statistical oversight, the three objectives, the cost of the loop, confidence versus calibration, and oversight as measurement."@en ; schema:isPartOf <> . :secNextQuestion a schema:CreativeWork ; schema:name "The Next Question"@en ; schema:abstract "The important shift may be a system that makes a narrower promise but proves it more reliably — one that knows when to act, when to ask, and can prove the distinction is working."@en ; schema:isPartOf <> . # ── Comparison Dimensions (Three Objectives) ─────────────────────────────────── :dimQuestionItAnswers a schema:DefinedTerm ; schema:name "Question it answers"@en ; schema:description "The core question each objective is designed to resolve about a system's output."@en . :dimWhatItOptimizes a schema:DefinedTerm ; schema:name "What it optimizes"@en ; schema:description "The underlying signal each objective is trained or rewarded against."@en . :dimBestFor a schema:DefinedTerm ; schema:name "Best suited for"@en ; schema:description "The use case or workflow context where each objective adds the most value."@en . :dimLimitation a schema:DefinedTerm ; schema:name "Primary limitation"@en ; schema:description "What each objective alone cannot guarantee about unattended decisions."@en . # ── FAQ ──────────────────────────────────────────────────────────────────────── :faqPage a schema:FAQPage ; schema:name "Frequently Asked Questions"@en ; schema:mainEntity :q1, :q2, :q3, :q4, :q5, :q6, :q7, :q8, :q9, :q10, :q11, :q12 ; schema:isPartOf <> . :q1 a schema:Question ; schema:name "What is Jev?"@en ; schema:acceptedAnswer :a1 . :a1 a schema:Answer ; schema:text "Jev is TypeSafe AI's decision model, introduced in early access on September 15, 2026. Instead of free-form text, it returns structured decisions and calibrated probabilities that software can consume directly, using three primitives — Choice, Noul, and Score."@en . :q2 a schema:Question ; schema:name "Who built Jev and why is the founder notable?"@en ; schema:acceptedAnswer :a2 . :a2 a schema:Answer ; schema:text "Jev was built by TypeSafe AI, founded by Diogo Almeida, a co-author of the InstructGPT paper. His background in RLHF makes Jev's alternative training objective — calibration rather than preference — a deliberate product bet."@en . :q3 a schema:Question ; schema:name "What are Jev's three core primitives?"@en ; schema:acceptedAnswer :a3 . :a3 a schema:Answer ; schema:text "Choice selects among defined alternatives; Noul returns the probability of a yes-or-no proposition; and Score evaluates something against an ordered scale. Several questions can be sent in one call against the same shared state."@en . :q4 a schema:Question ; schema:name "What is RLCD?"@en ; schema:acceptedAnswer :a4 . :a4 a schema:Answer ; schema:text "Reinforcement Learning for Calibrated Decisions is TypeSafe's stated training approach for Jev. It is designed around one idea: produce probabilities that software can use directly, making the numbers attached to decisions reflect how often those decisions are right."@en . :q5 a schema:Question ; schema:name "Is this the end of RLHF?"@en ; schema:acceptedAnswer :a5 . :a5 a schema:Answer ; schema:text "No. The stronger claim is narrower: preference optimization is not enough for every problem automation needs to solve. RLHF remains useful for assistants, but calibration can be reached through better base models, external verifiers, or dedicated decision models."@en . :q6 a schema:Question ; schema:name "What is the difference between preference, correctness, and calibration?"@en ; schema:acceptedAnswer :a6 . :a6 a schema:Answer ; schema:text "Preference asks which answer a person prefers; correctness asks whether the answer satisfied an objective verifier; calibration asks whether stated confidence matches how often the model is actually right. They are complementary, and a strong system may need all three."@en . :q7 a schema:Question ; schema:name "What is calibration and why does it matter?"@en ; schema:acceptedAnswer :a7 . :a7 a schema:Answer ; schema:text "Calibration means that when a model says it is 90% confident, it is right about 90% of the time. It matters because accuracy alone does not solve automation — a model that does not know when it is likely to be wrong is dangerous to leave unattended."@en . :q8 a schema:Question ; schema:name "Is a probability the same as a decision?"@en ; schema:acceptedAnswer :a8 . :a8 a schema:Answer ; schema:text "No. A probability tells you how likely something is; a decision rule tells you what to do with it. The most likely answer is not always the economically correct action — thresholds depend on the cost of error, the cost of review, and reversibility."@en . :q9 a schema:Question ; schema:name "Does calibration remove the need for a harness?"@en ; schema:acceptedAnswer :a9 . :a9 a schema:Answer ; schema:text "No. A calibrated decision cannot create a missing business rule, grant permission, prevent duplicate execution, or verify that the intended outcome happened. The model supplies the judgment; the harness turns that judgment into governed execution."@en . :q10 a schema:Question ; schema:name "What are forced loops and chosen loops?"@en ; schema:acceptedAnswer :a10 . :a10 a schema:Answer ; schema:text "Forced loops keep humans in place because the system cannot yet be trusted to act reliably. Chosen loops keep humans in place because the decision requires judgment, accountability, values, or authority. Forced loops should shrink as evidence improves; chosen loops may remain permanently."@en . :q11 a schema:Question ; schema:name "What should a serious deployment test first?"@en ; schema:acceptedAnswer :a11 . :a11 a schema:Answer ; schema:text "Start with one bounded decision with an observable outcome, frequent examples, clear actions, and a practical fallback. Then test prediction, policy, workflow, unattended execution, and total economics before expanding scope."@en . :q12 a schema:Question ; schema:name "What are the two loops in a robust automation architecture?"@en ; schema:acceptedAnswer :a12 . :a12 a schema:Answer ; schema:text "The operational loop handles each decision — prediction, policy, action, result. The measurement loop watches over time — outcome, audit, calibration, drift detection, updated policy. The second loop keeps the connection between prediction, action, and consequence maintained."@en . # ── Glossary ─────────────────────────────────────────────────────────────────── :glossary a schema:DefinedTermSet ; schema:name "Core Technical Glossary"@en ; schema:hasDefinedTerm :termRlhf, :termRlcd, :termCalibration, :termPreference, :termCorrectness, :termSycophancy, :termHarness, :termMetaHarness, :termOperationalLoop, :termMeasurementLoop, :termSeatAnchor, :termDecisionRule ; schema:isPartOf <> . :termRlhf a schema:DefinedTerm ; schema:name "Reinforcement Learning from Human Feedback (RLHF)"@en ; schema:description "Training that combines supervised examples with reinforcement learning guided by human rankings of responses, turning a text-predictor into an instruction-following assistant."@en ; schema:inDefinedTermSet :glossary . :termRlcd a schema:DefinedTerm ; schema:name "Reinforcement Learning for Calibrated Decisions (RLCD)"@en ; schema:description "TypeSafe's training objective for Jev, aimed at producing probabilities that correspond to how often decisions are actually right."@en ; schema:inDefinedTermSet :glossary . :termCalibration a schema:DefinedTerm ; schema:name "Calibration"@en ; schema:description "The property that stated confidence matches observed correctness across a group of predictions, making risk measurable enough to act on."@en ; schema:inDefinedTermSet :glossary . :termPreference a schema:DefinedTerm ; schema:name "Preference"@en ; schema:description "One of three objectives: which response a person would prefer — useful for instruction-following and tone, but not a guarantee of truth."@en ; schema:inDefinedTermSet :glossary . :termCorrectness a schema:DefinedTerm ; schema:name "Correctness"@en ; schema:description "One of three objectives: whether the answer satisfied an objective verifier, as in mathematics or code."@en ; schema:inDefinedTermSet :glossary . :termSycophancy a schema:DefinedTerm ; schema:name "Sycophancy"@en ; schema:description "A failure mode where a model becomes excessively agreeable, favoring convincing or expectation-aligned answers over correct ones."@en ; schema:inDefinedTermSet :glossary . :termHarness a schema:DefinedTerm ; schema:name "Harness"@en ; schema:description "The operating layer around a model that supplies context, permissions, approval rules, failure handling, recovery, and a record of what happened."@en ; schema:inDefinedTermSet :glossary . :termMetaHarness a schema:DefinedTerm ; schema:name "Meta-harness"@en ; schema:description "The layer that coordinates many agents, models, tools, permissions, context, and execution environments — the product that makes agents useful together."@en ; schema:inDefinedTermSet :glossary . :termOperationalLoop a schema:DefinedTerm ; schema:name "Operational Loop"@en ; schema:description "The loop that handles each decision: prediction, policy, action, result."@en ; schema:inDefinedTermSet :glossary . :termMeasurementLoop a schema:DefinedTerm ; schema:name "Measurement Loop"@en ; schema:description "The loop that watches over time: outcome, audit, calibration, drift detection, and updated policy."@en ; schema:inDefinedTermSet :glossary . :termSeatAnchor a schema:DefinedTerm ; schema:name "Seat Anchor"@en ; schema:description "The per-seat pricing assumption that value scales with the number of humans using the software; automation shifts the unit toward decisions and outcomes."@en ; schema:inDefinedTermSet :glossary . :termDecisionRule a schema:DefinedTerm ; schema:name "Decision Rule"@en ; schema:description "The policy that maps a probability and its context to an action, driven by the cost of error, the cost of review, and reversibility."@en ; schema:inDefinedTermSet :glossary . # ── HowTo ────────────────────────────────────────────────────────────────────── :howto a schema:HowTo ; schema:name "How to Test a Decision Before Automating It"@en ; schema:description "A staged checklist for moving from human review to selective automation on a single bounded decision."@en ; schema:step :step1, :step2, :step3, :step4, :step5, :step6, :step7, :step8, :step9 ; schema:isPartOf <> . :step1 a schema:HowToStep ; schema:position 1 ; schema:name "Choose one bounded, observable decision"@en ; schema:text "Pick a decision whose outcome can be observed, whose examples occur often enough to measure, whose actions are clear, and which has a practical fallback when the system is unsure."@en . :step2 a schema:HowToStep ; schema:position 2 ; schema:name "Test the prediction"@en ; schema:text "Check whether the model makes the right call, whether its probabilities are reliable, whether wording or option order moves the result, and what happens when context is missing — especially near the automation threshold."@en . :step3 a schema:HowToStep ; schema:position 3 ; schema:name "Test the policy"@en ; schema:text "Decide which mistakes matter most, which cases may be automated, which still require approval, and what happens when confidence is too low."@en . :step4 a schema:HowToStep ; schema:position 4 ; schema:name "Test the workflow"@en ; schema:text "Verify that the right action is actually executed, permissions are enforced, duplicate requests are handled safely, and the process can recover when something downstream fails."@en . :step5 a schema:HowToStep ; schema:position 5 ; schema:name "Test unattended execution"@en ; schema:text "Run the system in shadow mode on held-out cases before letting it act freely, then continue auditing a sample of automated decisions after launch."@en . :step6 a schema:HowToStep ; schema:position 6 ; schema:name "Test the economics"@en ; schema:text "Include inference, remaining human review, monitoring, maintenance, remediation, and the cost of errors — then compare the total with the existing process and simpler alternatives."@en . :step7 a schema:HowToStep ; schema:position 7 ; schema:name "Measure calibration and error rates"@en ; schema:text "Sample decisions, collect outcomes, and measure calibration against real outcomes on the task where the system will actually operate."@en . :step8 a schema:HowToStep ; schema:position 8 ; schema:name "Monitor for drift"@en ; schema:text "Watch for changes in customer behavior, policy, products, and fraud patterns that make a once-reliable probability mean something different today."@en . :step9 a schema:HowToStep ; schema:position 9 ; schema:name "Adjust thresholds, policies, or models"@en ; schema:text "Change thresholds, policies, or models when performance deteriorates, and let automation earn autonomy through evidence — keeping it only while the evidence holds."@en .