VIEW THIS AS

Auto mode follows the Route Engine until you choose a viewpoint.

YOU ARE HERE

ROUTE CHECK

CONNECTED TO

WHAT NEXT

Use the canonical route for this room, or HELP if you are unsure.

Super Intelligence | SI Timelines | How to Read Forecasts, Probabilities and Expert Disagreement

Three students work together at a classroom table, reviewing open textbooks and notes.

Super Intelligence (SI) is not only a question of algorithms. This article examines Super Intelligence timelines and forecasts: teaching readers to interpret dated forecasts and probabilities without converting them into predictions of who or what will happen. It preserves the locked Clementi-depth floor with mechanisms, worked cases, sensitivity analysis, diagnostics, failure modes and RFE closure.

Search Intent and Immediate Answer

The search question in Super Intelligence timelines and forecasts is teaching readers to interpret dated forecasts and probabilities without converting them into predictions of who or what will happen. A rigorous answer separates definition, date, probability, population, assumptions, calibration and uncertainty. These variables can move at different speeds and can constrain one another. Super Intelligence (SI) forecasting becomes misleading when one fast-moving variable is extrapolated as though every supporting system must move with it.

First principles begin with throughput. Identify the resource or process entering the system, the transformation being attempted, and the slowest stage that determines completed output. If compute doubles but power, fabrication, verification or high-quality data does not, the realised capability gain may differ from the headline expectation. Bottleneck maps are more useful than single-variable stories.

A small-system example provides a controlled baseline. Measure how performance changes when one resource is increased while everything else is held constant. Then repeat at larger scale. If the gain diminishes, another constraint has become active. This is the basic logic behind analysing physical limits, learning bottlenecks and capability takeoff rather than assuming linear continuation.

Definition and Scope

A small-system example provides a controlled baseline. Measure how performance changes when one resource is increased while everything else is held constant. Then repeat at larger scale. If the gain diminishes, another constraint has become active. This is the basic logic behind analysing physical limits, learning bottlenecks and capability takeoff rather than assuming linear continuation.

Frontier scale changes the environment. Procurement, networking, cooling, data quality, evaluation and organisational coordination can become first-order constraints. A technique that works economically on a small experiment may become expensive or operationally fragile when multiplied across a large system. SI analysis should follow the complete stack from algorithm to deployed capability.

A changed-assumption test is essential for forecasts. If a timeline depends on compute growth, change the compute-growth assumption. If it depends on research automation, reduce the assumed automation rate. If it depends on synthetic data quality, model verification failure. A forecast that changes dramatically under modest assumption changes should be presented with correspondingly wide uncertainty.

First Principles

A changed-assumption test is essential for forecasts. If a timeline depends on compute growth, change the compute-growth assumption. If it depends on research automation, reduce the assumed automation rate. If it depends on synthetic data quality, model verification failure. A forecast that changes dramatically under modest assumption changes should be presented with correspondingly wide uncertainty.

Failed forecasts are evidence about models of the world, not merely embarrassing dates. Ask which assumption failed: capability scaling, economics, adoption, hardware, regulation or the definition of the milestone. Updating the causal model is more valuable than quietly moving the date. This is how forecasting becomes a learning process.

Measurement should match the variable. Power is measured differently from model loss; data diversity differently from token count; takeoff speed differently from calendar predictions. Define the numerator and denominator before comparing trends. Ambiguous metrics can create apparent acceleration simply because the measurement changed.

The System Boundary

Measurement should match the variable. Power is measured differently from model loss; data diversity differently from token count; takeoff speed differently from calendar predictions. Define the numerator and denominator before comparing trends. Ambiguous metrics can create apparent acceleration simply because the measurement changed.

Extrapolation is the dangerous step. An empirical relationship observed over one range may continue, bend or break outside it. Scaling laws, hardware trends and benchmark curves are useful precisely because they summarise observed regularities, but none is a physical guarantee of indefinite continuation. State the range and assumptions.

Uncertainty has structure. Some uncertainty comes from measurement noise; some from unknown future discoveries; some from disagreement about definitions; some from strategic or economic behaviour. Combining them into one confident date hides the reason for uncertainty. Good SI timelines show which assumption contributes most to the spread.

The Core Bottleneck

Uncertainty has structure. Some uncertainty comes from measurement noise; some from unknown future discoveries; some from disagreement about definitions; some from strategic or economic behaviour. Combining them into one confident date hides the reason for uncertainty. Good SI timelines show which assumption contributes most to the spread.

Feedback loops can accelerate progress when outputs become inputs to the next improvement cycle. But the loop contains latency. A software change may be tested in minutes; a new chip may require design and fabrication; an energy project may take years. Takeoff speed depends on which loop dominates and how tightly it couples to the rest of the system.

Economics determines which technically possible paths are actually pursued. Compute has a price, electricity has a price, expert supervision has a price and failed experiments consume resources. Falling unit costs can accelerate adoption, while rising marginal costs can slow it. Economic constraints should sit beside technical ones.

What People Commonly Assume

Economics determines which technically possible paths are actually pursued. Compute has a price, electricity has a price, expert supervision has a price and failed experiments consume resources. Falling unit costs can accelerate adoption, while rising marginal costs can slow it. Economic constraints should sit beside technical ones.

Supply chains make intelligence physical. Advanced chips depend on specialised equipment, materials, fabrication capacity and logistics. Data centres require construction, grid connections and cooling. These facts do not disprove rapid cognitive progress; they specify the channels through which digital progress reaches scale.

Current evidence supports rapid improvement in several AI capability domains and continuing investment in compute infrastructure. It does not establish a fixed SI date, a guaranteed fast takeoff or unlimited synthetic-data improvement. The evidence is compatible with multiple future trajectories, which is why conditional reasoning is preferable to inevitability language.

Worked Example: A Small System

Current evidence supports rapid improvement in several AI capability domains and continuing investment in compute infrastructure. It does not establish a fixed SI date, a guaranteed fast takeoff or unlimited synthetic-data improvement. The evidence is compatible with multiple future trajectories, which is why conditional reasoning is preferable to inevitability language.

Safety preparation depends on lead time. Faster capability change can reduce the time available for evaluation, workforce adaptation and institutional response. Slower change can create more opportunity for learning but does not guarantee good governance. Preparation should therefore focus on observable indicators and adaptable controls rather than one predicted date.

Education should teach forecast literacy. A student should identify the event being forecast, the definition used, the date of the forecast, the probability assigned, the population or expert group sampled, and the assumptions underneath it. A probability is not a promise and a median survey response is not a scheduled event.

Worked Example: Frontier Scale

Education should teach forecast literacy. A student should identify the event being forecast, the definition used, the date of the forecast, the probability assigned, the population or expert group sampled, and the assumptions underneath it. A probability is not a promise and a median survey response is not a scheduled event.

For organisations, convert forecasts into scenarios rather than a single plan. Define what action would be appropriate if capability advances slowly, moderately or rapidly. Identify indicators that would trigger movement between plans. This reduces dependence on guessing the exact year and increases readiness across plausible paths.

Progress has four stages: define the variable, explain the mechanism, model sensitivity to changed assumptions, and design an update rule when evidence arrives. This is the Clementi-depth progression from recognition to independent control. Readers should finish able to revise a forecast, not merely repeat one.

Worked Example: A Changed Assumption

Progress has four stages: define the variable, explain the mechanism, model sensitivity to changed assumptions, and design an update rule when evidence arrives. This is the Clementi-depth progression from recognition to independent control. Readers should finish able to revise a forecast, not merely repeat one.

RFE closes the loop. Receiver: who needs the forecast or resource analysis? Function: what decision must it support? Evidence: what observable indicator changes the decision? Exit: when should the model or plan be revised? Applied to Super Intelligence timelines and forecasts, RFE makes uncertainty operational rather than rhetorical.

The search question in Super Intelligence timelines and forecasts is teaching readers to interpret dated forecasts and probabilities without converting them into predictions of who or what will happen. A rigorous answer separates definition, date, probability, population, assumptions, calibration and uncertainty. These variables can move at different speeds and can constrain one another. Super Intelligence (SI) forecasting becomes misleading when one fast-moving variable is extrapolated as though every supporting system must move with it.

Worked Example: A Failed Forecast

The search question in Super Intelligence timelines and forecasts is teaching readers to interpret dated forecasts and probabilities without converting them into predictions of who or what will happen. A rigorous answer separates definition, date, probability, population, assumptions, calibration and uncertainty. These variables can move at different speeds and can constrain one another. Super Intelligence (SI) forecasting becomes misleading when one fast-moving variable is extrapolated as though every supporting system must move with it.

First principles begin with throughput. Identify the resource or process entering the system, the transformation being attempted, and the slowest stage that determines completed output. If compute doubles but power, fabrication, verification or high-quality data does not, the realised capability gain may differ from the headline expectation. Bottleneck maps are more useful than single-variable stories.

A small-system example provides a controlled baseline. Measure how performance changes when one resource is increased while everything else is held constant. Then repeat at larger scale. If the gain diminishes, another constraint has become active. This is the basic logic behind analysing physical limits, learning bottlenecks and capability takeoff rather than assuming linear continuation.

How to Measure the Variable

A small-system example provides a controlled baseline. Measure how performance changes when one resource is increased while everything else is held constant. Then repeat at larger scale. If the gain diminishes, another constraint has become active. This is the basic logic behind analysing physical limits, learning bottlenecks and capability takeoff rather than assuming linear continuation.

Frontier scale changes the environment. Procurement, networking, cooling, data quality, evaluation and organisational coordination can become first-order constraints. A technique that works economically on a small experiment may become expensive or operationally fragile when multiplied across a large system. SI analysis should follow the complete stack from algorithm to deployed capability.

A changed-assumption test is essential for forecasts. If a timeline depends on compute growth, change the compute-growth assumption. If it depends on research automation, reduce the assumed automation rate. If it depends on synthetic data quality, model verification failure. A forecast that changes dramatically under modest assumption changes should be presented with correspondingly wide uncertainty.

What the Data Can Show

A changed-assumption test is essential for forecasts. If a timeline depends on compute growth, change the compute-growth assumption. If it depends on research automation, reduce the assumed automation rate. If it depends on synthetic data quality, model verification failure. A forecast that changes dramatically under modest assumption changes should be presented with correspondingly wide uncertainty.

Failed forecasts are evidence about models of the world, not merely embarrassing dates. Ask which assumption failed: capability scaling, economics, adoption, hardware, regulation or the definition of the milestone. Updating the causal model is more valuable than quietly moving the date. This is how forecasting becomes a learning process.

Measurement should match the variable. Power is measured differently from model loss; data diversity differently from token count; takeoff speed differently from calendar predictions. Define the numerator and denominator before comparing trends. Ambiguous metrics can create apparent acceleration simply because the measurement changed.

What the Data Cannot Show

Measurement should match the variable. Power is measured differently from model loss; data diversity differently from token count; takeoff speed differently from calendar predictions. Define the numerator and denominator before comparing trends. Ambiguous metrics can create apparent acceleration simply because the measurement changed.

Extrapolation is the dangerous step. An empirical relationship observed over one range may continue, bend or break outside it. Scaling laws, hardware trends and benchmark curves are useful precisely because they summarise observed regularities, but none is a physical guarantee of indefinite continuation. State the range and assumptions.

Uncertainty has structure. Some uncertainty comes from measurement noise; some from unknown future discoveries; some from disagreement about definitions; some from strategic or economic behaviour. Combining them into one confident date hides the reason for uncertainty. Good SI timelines show which assumption contributes most to the spread.

Generalisation and Extrapolation

Uncertainty has structure. Some uncertainty comes from measurement noise; some from unknown future discoveries; some from disagreement about definitions; some from strategic or economic behaviour. Combining them into one confident date hides the reason for uncertainty. Good SI timelines show which assumption contributes most to the spread.

Feedback loops can accelerate progress when outputs become inputs to the next improvement cycle. But the loop contains latency. A software change may be tested in minutes; a new chip may require design and fabrication; an energy project may take years. Takeoff speed depends on which loop dominates and how tightly it couples to the rest of the system.

Economics determines which technically possible paths are actually pursued. Compute has a price, electricity has a price, expert supervision has a price and failed experiments consume resources. Falling unit costs can accelerate adoption, while rising marginal costs can slow it. Economic constraints should sit beside technical ones.

Reliability and Uncertainty

Economics determines which technically possible paths are actually pursued. Compute has a price, electricity has a price, expert supervision has a price and failed experiments consume resources. Falling unit costs can accelerate adoption, while rising marginal costs can slow it. Economic constraints should sit beside technical ones.

Supply chains make intelligence physical. Advanced chips depend on specialised equipment, materials, fabrication capacity and logistics. Data centres require construction, grid connections and cooling. These facts do not disprove rapid cognitive progress; they specify the channels through which digital progress reaches scale.

Current evidence supports rapid improvement in several AI capability domains and continuing investment in compute infrastructure. It does not establish a fixed SI date, a guaranteed fast takeoff or unlimited synthetic-data improvement. The evidence is compatible with multiple future trajectories, which is why conditional reasoning is preferable to inevitability language.

Sensitivity to Assumptions

Current evidence supports rapid improvement in several AI capability domains and continuing investment in compute infrastructure. It does not establish a fixed SI date, a guaranteed fast takeoff or unlimited synthetic-data improvement. The evidence is compatible with multiple future trajectories, which is why conditional reasoning is preferable to inevitability language.

Safety preparation depends on lead time. Faster capability change can reduce the time available for evaluation, workforce adaptation and institutional response. Slower change can create more opportunity for learning but does not guarantee good governance. Preparation should therefore focus on observable indicators and adaptable controls rather than one predicted date.

Education should teach forecast literacy. A student should identify the event being forecast, the definition used, the date of the forecast, the probability assigned, the population or expert group sampled, and the assumptions underneath it. A probability is not a promise and a median survey response is not a scheduled event.

Feedback Loops

Education should teach forecast literacy. A student should identify the event being forecast, the definition used, the date of the forecast, the probability assigned, the population or expert group sampled, and the assumptions underneath it. A probability is not a promise and a median survey response is not a scheduled event.

For organisations, convert forecasts into scenarios rather than a single plan. Define what action would be appropriate if capability advances slowly, moderately or rapidly. Identify indicators that would trigger movement between plans. This reduces dependence on guessing the exact year and increases readiness across plausible paths.

Progress has four stages: define the variable, explain the mechanism, model sensitivity to changed assumptions, and design an update rule when evidence arrives. This is the Clementi-depth progression from recognition to independent control. Readers should finish able to revise a forecast, not merely repeat one.

Physical and Institutional Latency

Progress has four stages: define the variable, explain the mechanism, model sensitivity to changed assumptions, and design an update rule when evidence arrives. This is the Clementi-depth progression from recognition to independent control. Readers should finish able to revise a forecast, not merely repeat one.

RFE closes the loop. Receiver: who needs the forecast or resource analysis? Function: what decision must it support? Evidence: what observable indicator changes the decision? Exit: when should the model or plan be revised? Applied to Super Intelligence timelines and forecasts, RFE makes uncertainty operational rather than rhetorical.

The search question in Super Intelligence timelines and forecasts is teaching readers to interpret dated forecasts and probabilities without converting them into predictions of who or what will happen. A rigorous answer separates definition, date, probability, population, assumptions, calibration and uncertainty. These variables can move at different speeds and can constrain one another. Super Intelligence (SI) forecasting becomes misleading when one fast-moving variable is extrapolated as though every supporting system must move with it.

Economics and Cost

The search question in Super Intelligence timelines and forecasts is teaching readers to interpret dated forecasts and probabilities without converting them into predictions of who or what will happen. A rigorous answer separates definition, date, probability, population, assumptions, calibration and uncertainty. These variables can move at different speeds and can constrain one another. Super Intelligence (SI) forecasting becomes misleading when one fast-moving variable is extrapolated as though every supporting system must move with it.

First principles begin with throughput. Identify the resource or process entering the system, the transformation being attempted, and the slowest stage that determines completed output. If compute doubles but power, fabrication, verification or high-quality data does not, the realised capability gain may differ from the headline expectation. Bottleneck maps are more useful than single-variable stories.

A small-system example provides a controlled baseline. Measure how performance changes when one resource is increased while everything else is held constant. Then repeat at larger scale. If the gain diminishes, another constraint has become active. This is the basic logic behind analysing physical limits, learning bottlenecks and capability takeoff rather than assuming linear continuation.

Supply Chains and Capacity

A small-system example provides a controlled baseline. Measure how performance changes when one resource is increased while everything else is held constant. Then repeat at larger scale. If the gain diminishes, another constraint has become active. This is the basic logic behind analysing physical limits, learning bottlenecks and capability takeoff rather than assuming linear continuation.

Frontier scale changes the environment. Procurement, networking, cooling, data quality, evaluation and organisational coordination can become first-order constraints. A technique that works economically on a small experiment may become expensive or operationally fragile when multiplied across a large system. SI analysis should follow the complete stack from algorithm to deployed capability.

A changed-assumption test is essential for forecasts. If a timeline depends on compute growth, change the compute-growth assumption. If it depends on research automation, reduce the assumed automation rate. If it depends on synthetic data quality, model verification failure. A forecast that changes dramatically under modest assumption changes should be presented with correspondingly wide uncertainty.

What Current Evidence Supports

A changed-assumption test is essential for forecasts. If a timeline depends on compute growth, change the compute-growth assumption. If it depends on research automation, reduce the assumed automation rate. If it depends on synthetic data quality, model verification failure. A forecast that changes dramatically under modest assumption changes should be presented with correspondingly wide uncertainty.

Failed forecasts are evidence about models of the world, not merely embarrassing dates. Ask which assumption failed: capability scaling, economics, adoption, hardware, regulation or the definition of the milestone. Updating the causal model is more valuable than quietly moving the date. This is how forecasting becomes a learning process.

Measurement should match the variable. Power is measured differently from model loss; data diversity differently from token count; takeoff speed differently from calendar predictions. Define the numerator and denominator before comparing trends. Ambiguous metrics can create apparent acceleration simply because the measurement changed.

What Current Evidence Does Not Establish

Measurement should match the variable. Power is measured differently from model loss; data diversity differently from token count; takeoff speed differently from calendar predictions. Define the numerator and denominator before comparing trends. Ambiguous metrics can create apparent acceleration simply because the measurement changed.

Extrapolation is the dangerous step. An empirical relationship observed over one range may continue, bend or break outside it. Scaling laws, hardware trends and benchmark curves are useful precisely because they summarise observed regularities, but none is a physical guarantee of indefinite continuation. State the range and assumptions.

Uncertainty has structure. Some uncertainty comes from measurement noise; some from unknown future discoveries; some from disagreement about definitions; some from strategic or economic behaviour. Combining them into one confident date hides the reason for uncertainty. Good SI timelines show which assumption contributes most to the spread.

Connection to Super Intelligence (SI)

Uncertainty has structure. Some uncertainty comes from measurement noise; some from unknown future discoveries; some from disagreement about definitions; some from strategic or economic behaviour. Combining them into one confident date hides the reason for uncertainty. Good SI timelines show which assumption contributes most to the spread.

Feedback loops can accelerate progress when outputs become inputs to the next improvement cycle. But the loop contains latency. A software change may be tested in minutes; a new chip may require design and fabrication; an energy project may take years. Takeoff speed depends on which loop dominates and how tightly it couples to the rest of the system.

Economics determines which technically possible paths are actually pursued. Compute has a price, electricity has a price, expert supervision has a price and failed experiments consume resources. Falling unit costs can accelerate adoption, while rising marginal costs can slow it. Economic constraints should sit beside technical ones.

Safety and Preparation

Economics determines which technically possible paths are actually pursued. Compute has a price, electricity has a price, expert supervision has a price and failed experiments consume resources. Falling unit costs can accelerate adoption, while rising marginal costs can slow it. Economic constraints should sit beside technical ones.

Supply chains make intelligence physical. Advanced chips depend on specialised equipment, materials, fabrication capacity and logistics. Data centres require construction, grid connections and cooling. These facts do not disprove rapid cognitive progress; they specify the channels through which digital progress reaches scale.

Current evidence supports rapid improvement in several AI capability domains and continuing investment in compute infrastructure. It does not establish a fixed SI date, a guaranteed fast takeoff or unlimited synthetic-data improvement. The evidence is compatible with multiple future trajectories, which is why conditional reasoning is preferable to inevitability language.

Governance and Coordination

Current evidence supports rapid improvement in several AI capability domains and continuing investment in compute infrastructure. It does not establish a fixed SI date, a guaranteed fast takeoff or unlimited synthetic-data improvement. The evidence is compatible with multiple future trajectories, which is why conditional reasoning is preferable to inevitability language.

Safety preparation depends on lead time. Faster capability change can reduce the time available for evaluation, workforce adaptation and institutional response. Slower change can create more opportunity for learning but does not guarantee good governance. Preparation should therefore focus on observable indicators and adaptable controls rather than one predicted date.

Education should teach forecast literacy. A student should identify the event being forecast, the definition used, the date of the forecast, the probability assigned, the population or expert group sampled, and the assumptions underneath it. A probability is not a promise and a median survey response is not a scheduled event.

Education: Reading Claims Carefully

Education should teach forecast literacy. A student should identify the event being forecast, the definition used, the date of the forecast, the probability assigned, the population or expert group sampled, and the assumptions underneath it. A probability is not a promise and a median survey response is not a scheduled event.

For organisations, convert forecasts into scenarios rather than a single plan. Define what action would be appropriate if capability advances slowly, moderately or rapidly. Identify indicators that would trigger movement between plans. This reduces dependence on guessing the exact year and increases readiness across plausible paths.

Progress has four stages: define the variable, explain the mechanism, model sensitivity to changed assumptions, and design an update rule when evidence arrives. This is the Clementi-depth progression from recognition to independent control. Readers should finish able to revise a forecast, not merely repeat one.

Student Diagnostic Checklist

Progress has four stages: define the variable, explain the mechanism, model sensitivity to changed assumptions, and design an update rule when evidence arrives. This is the Clementi-depth progression from recognition to independent control. Readers should finish able to revise a forecast, not merely repeat one.

RFE closes the loop. Receiver: who needs the forecast or resource analysis? Function: what decision must it support? Evidence: what observable indicator changes the decision? Exit: when should the model or plan be revised? Applied to Super Intelligence timelines and forecasts, RFE makes uncertainty operational rather than rhetorical.

The search question in Super Intelligence timelines and forecasts is teaching readers to interpret dated forecasts and probabilities without converting them into predictions of who or what will happen. A rigorous answer separates definition, date, probability, population, assumptions, calibration and uncertainty. These variables can move at different speeds and can constrain one another. Super Intelligence (SI) forecasting becomes misleading when one fast-moving variable is extrapolated as though every supporting system must move with it.

Organisation Diagnostic Checklist

The search question in Super Intelligence timelines and forecasts is teaching readers to interpret dated forecasts and probabilities without converting them into predictions of who or what will happen. A rigorous answer separates definition, date, probability, population, assumptions, calibration and uncertainty. These variables can move at different speeds and can constrain one another. Super Intelligence (SI) forecasting becomes misleading when one fast-moving variable is extrapolated as though every supporting system must move with it.

First principles begin with throughput. Identify the resource or process entering the system, the transformation being attempted, and the slowest stage that determines completed output. If compute doubles but power, fabrication, verification or high-quality data does not, the realised capability gain may differ from the headline expectation. Bottleneck maps are more useful than single-variable stories.

A small-system example provides a controlled baseline. Measure how performance changes when one resource is increased while everything else is held constant. Then repeat at larger scale. If the gain diminishes, another constraint has become active. This is the basic logic behind analysing physical limits, learning bottlenecks and capability takeoff rather than assuming linear continuation.

Progress Ladder

A small-system example provides a controlled baseline. Measure how performance changes when one resource is increased while everything else is held constant. Then repeat at larger scale. If the gain diminishes, another constraint has become active. This is the basic logic behind analysing physical limits, learning bottlenecks and capability takeoff rather than assuming linear continuation.

Frontier scale changes the environment. Procurement, networking, cooling, data quality, evaluation and organisational coordination can become first-order constraints. A technique that works economically on a small experiment may become expensive or operationally fragile when multiplied across a large system. SI analysis should follow the complete stack from algorithm to deployed capability.

A changed-assumption test is essential for forecasts. If a timeline depends on compute growth, change the compute-growth assumption. If it depends on research automation, reduce the assumed automation rate. If it depends on synthetic data quality, model verification failure. A forecast that changes dramatically under modest assumption changes should be presented with correspondingly wide uncertainty.

Counterexamples

A changed-assumption test is essential for forecasts. If a timeline depends on compute growth, change the compute-growth assumption. If it depends on research automation, reduce the assumed automation rate. If it depends on synthetic data quality, model verification failure. A forecast that changes dramatically under modest assumption changes should be presented with correspondingly wide uncertainty.

Failed forecasts are evidence about models of the world, not merely embarrassing dates. Ask which assumption failed: capability scaling, economics, adoption, hardware, regulation or the definition of the milestone. Updating the causal model is more valuable than quietly moving the date. This is how forecasting becomes a learning process.

Measurement should match the variable. Power is measured differently from model loss; data diversity differently from token count; takeoff speed differently from calendar predictions. Define the numerator and denominator before comparing trends. Ambiguous metrics can create apparent acceleration simply because the measurement changed.

What Would Change the Conclusion?

Measurement should match the variable. Power is measured differently from model loss; data diversity differently from token count; takeoff speed differently from calendar predictions. Define the numerator and denominator before comparing trends. Ambiguous metrics can create apparent acceleration simply because the measurement changed.

Extrapolation is the dangerous step. An empirical relationship observed over one range may continue, bend or break outside it. Scaling laws, hardware trends and benchmark curves are useful precisely because they summarise observed regularities, but none is a physical guarantee of indefinite continuation. State the range and assumptions.

Uncertainty has structure. Some uncertainty comes from measurement noise; some from unknown future discoveries; some from disagreement about definitions; some from strategic or economic behaviour. Combining them into one confident date hides the reason for uncertainty. Good SI timelines show which assumption contributes most to the spread.

RFE Closure

Uncertainty has structure. Some uncertainty comes from measurement noise; some from unknown future discoveries; some from disagreement about definitions; some from strategic or economic behaviour. Combining them into one confident date hides the reason for uncertainty. Good SI timelines show which assumption contributes most to the spread.

Feedback loops can accelerate progress when outputs become inputs to the next improvement cycle. But the loop contains latency. A software change may be tested in minutes; a new chip may require design and fabrication; an energy project may take years. Takeoff speed depends on which loop dominates and how tightly it couples to the rest of the system.

Economics determines which technically possible paths are actually pursued. Compute has a price, electricity has a price, expert supervision has a price and failed experiments consume resources. Falling unit costs can accelerate adoption, while rising marginal costs can slow it. Economic constraints should sit beside technical ones.

Frequently Asked Questions

Economics determines which technically possible paths are actually pursued. Compute has a price, electricity has a price, expert supervision has a price and failed experiments consume resources. Falling unit costs can accelerate adoption, while rising marginal costs can slow it. Economic constraints should sit beside technical ones.

Supply chains make intelligence physical. Advanced chips depend on specialised equipment, materials, fabrication capacity and logistics. Data centres require construction, grid connections and cooling. These facts do not disprove rapid cognitive progress; they specify the channels through which digital progress reaches scale.

Current evidence supports rapid improvement in several AI capability domains and continuing investment in compute infrastructure. It does not establish a fixed SI date, a guaranteed fast takeoff or unlimited synthetic-data improvement. The evidence is compatible with multiple future trajectories, which is why conditional reasoning is preferable to inevitability language.

Continue the Super Intelligence (SI) Series

Current evidence supports rapid improvement in several AI capability domains and continuing investment in compute infrastructure. It does not establish a fixed SI date, a guaranteed fast takeoff or unlimited synthetic-data improvement. The evidence is compatible with multiple future trajectories, which is why conditional reasoning is preferable to inevitability language.

Safety preparation depends on lead time. Faster capability change can reduce the time available for evaluation, workforce adaptation and institutional response. Slower change can create more opportunity for learning but does not guarantee good governance. Preparation should therefore focus on observable indicators and adaptable controls rather than one predicted date.

Education should teach forecast literacy. A student should identify the event being forecast, the definition used, the date of the forecast, the probability assigned, the population or expert group sampled, and the assumptions underneath it. A probability is not a promise and a median survey response is not a scheduled event.

Capacity Map: Find the Real Limiting Resource

Build a capacity map by listing every resource required for one completed unit of progress. Include compute, high-quality data, electricity, cooling, specialist labour, hardware lead time, verification and deployment. Mark current utilisation and replacement time. The limiting resource is not necessarily the most visible one. When a constraint tightens, it can become the new rate limiter even if every other input continues improving.

A scenario matrix avoids false precision. Create slow, middle and fast cases with explicit assumptions rather than vague moods. For each case, state capability growth, infrastructure expansion, research automation, verification throughput and adoption speed. The purpose is not to choose a winner; it is to see which decisions are robust across several plausible futures and which depend on one narrow forecast.

Sensitivity analysis changes one assumption while holding others fixed. Double hardware efficiency, halve data quality, delay new fabrication capacity, or increase verification cost. Observe how much the conclusion moves. The assumptions with the largest effect deserve the most monitoring. A model that swings wildly when one uncertain input changes should not be presented with a narrow confidence band.

Calibration turns forecasting into a skill that can improve. Keep dated predictions, probabilities and definitions. When the date passes or evidence arrives, score what happened and inspect whether confidence matched accuracy. Forecasting communities use calibration because a person who says 70 percent should be right roughly seven times in ten across comparable forecasts. One dramatic success is not enough to establish forecasting skill.

Scenario Matrix: Slow, Middle and Fast Cases

A scenario matrix avoids false precision. Create slow, middle and fast cases with explicit assumptions rather than vague moods. For each case, state capability growth, infrastructure expansion, research automation, verification throughput and adoption speed. The purpose is not to choose a winner; it is to see which decisions are robust across several plausible futures and which depend on one narrow forecast.

Sensitivity analysis changes one assumption while holding others fixed. Double hardware efficiency, halve data quality, delay new fabrication capacity, or increase verification cost. Observe how much the conclusion moves. The assumptions with the largest effect deserve the most monitoring. A model that swings wildly when one uncertain input changes should not be presented with a narrow confidence band.

Calibration turns forecasting into a skill that can improve. Keep dated predictions, probabilities and definitions. When the date passes or evidence arrives, score what happened and inspect whether confidence matched accuracy. Forecasting communities use calibration because a person who says 70 percent should be right roughly seven times in ten across comparable forecasts. One dramatic success is not enough to establish forecasting skill.

Decision triggers replace obsession with one calendar year. An organisation might prepare additional evaluation capacity when models cross a measured long-horizon threshold, revise permissions when agent reliability reaches a specified level, or invest in infrastructure when utilisation crosses a capacity boundary. Indicators connect preparation to observable evidence and allow plans to move as reality changes.

Sensitivity Test: Which Assumption Moves the Result Most?

Sensitivity analysis changes one assumption while holding others fixed. Double hardware efficiency, halve data quality, delay new fabrication capacity, or increase verification cost. Observe how much the conclusion moves. The assumptions with the largest effect deserve the most monitoring. A model that swings wildly when one uncertain input changes should not be presented with a narrow confidence band.

Calibration turns forecasting into a skill that can improve. Keep dated predictions, probabilities and definitions. When the date passes or evidence arrives, score what happened and inspect whether confidence matched accuracy. Forecasting communities use calibration because a person who says 70 percent should be right roughly seven times in ten across comparable forecasts. One dramatic success is not enough to establish forecasting skill.

Decision triggers replace obsession with one calendar year. An organisation might prepare additional evaluation capacity when models cross a measured long-horizon threshold, revise permissions when agent reliability reaches a specified level, or invest in infrastructure when utilisation crosses a capacity boundary. Indicators connect preparation to observable evidence and allow plans to move as reality changes.

The workbook begins with one question: what decision are you trying to make? Then list the uncertain variables, assign ranges rather than single values, identify observable indicators, create at least three scenarios and define what evidence would trigger an update. Finally, record the costs of acting too early and too late. This converts an SI timeline from a prediction contest into a decision-support tool.

Forecast Calibration: Learn From Past Predictions

Calibration turns forecasting into a skill that can improve. Keep dated predictions, probabilities and definitions. When the date passes or evidence arrives, score what happened and inspect whether confidence matched accuracy. Forecasting communities use calibration because a person who says 70 percent should be right roughly seven times in ten across comparable forecasts. One dramatic success is not enough to establish forecasting skill.

Decision triggers replace obsession with one calendar year. An organisation might prepare additional evaluation capacity when models cross a measured long-horizon threshold, revise permissions when agent reliability reaches a specified level, or invest in infrastructure when utilisation crosses a capacity boundary. Indicators connect preparation to observable evidence and allow plans to move as reality changes.

The workbook begins with one question: what decision are you trying to make? Then list the uncertain variables, assign ranges rather than single values, identify observable indicators, create at least three scenarios and define what evidence would trigger an update. Finally, record the costs of acting too early and too late. This converts an SI timeline from a prediction contest into a decision-support tool.

Constraints and forecasts must remain revisable. A new algorithm can loosen a compute bottleneck; a power shortage can tighten an infrastructure bottleneck; better synthetic-data verification can change a learning constraint. The correct model is therefore a living map of rates, buffers and evidence. Super Intelligence (SI) preparation is strongest when it can adapt without pretending uncertainty has disappeared.

Decision Triggers: Act on Indicators, Not Dates

Decision triggers replace obsession with one calendar year. An organisation might prepare additional evaluation capacity when models cross a measured long-horizon threshold, revise permissions when agent reliability reaches a specified level, or invest in infrastructure when utilisation crosses a capacity boundary. Indicators connect preparation to observable evidence and allow plans to move as reality changes.

The workbook begins with one question: what decision are you trying to make? Then list the uncertain variables, assign ranges rather than single values, identify observable indicators, create at least three scenarios and define what evidence would trigger an update. Finally, record the costs of acting too early and too late. This converts an SI timeline from a prediction contest into a decision-support tool.

Constraints and forecasts must remain revisable. A new algorithm can loosen a compute bottleneck; a power shortage can tighten an infrastructure bottleneck; better synthetic-data verification can change a learning constraint. The correct model is therefore a living map of rates, buffers and evidence. Super Intelligence (SI) preparation is strongest when it can adapt without pretending uncertainty has disappeared.

Build a capacity map by listing every resource required for one completed unit of progress. Include compute, high-quality data, electricity, cooling, specialist labour, hardware lead time, verification and deployment. Mark current utilisation and replacement time. The limiting resource is not necessarily the most visible one. When a constraint tightens, it can become the new rate limiter even if every other input continues improving.

Practical Workbook: Build an Updateable SI Model

The workbook begins with one question: what decision are you trying to make? Then list the uncertain variables, assign ranges rather than single values, identify observable indicators, create at least three scenarios and define what evidence would trigger an update. Finally, record the costs of acting too early and too late. This converts an SI timeline from a prediction contest into a decision-support tool.

Constraints and forecasts must remain revisable. A new algorithm can loosen a compute bottleneck; a power shortage can tighten an infrastructure bottleneck; better synthetic-data verification can change a learning constraint. The correct model is therefore a living map of rates, buffers and evidence. Super Intelligence (SI) preparation is strongest when it can adapt without pretending uncertainty has disappeared.

Build a capacity map by listing every resource required for one completed unit of progress. Include compute, high-quality data, electricity, cooling, specialist labour, hardware lead time, verification and deployment. Mark current utilisation and replacement time. The limiting resource is not necessarily the most visible one. When a constraint tightens, it can become the new rate limiter even if every other input continues improving.

A scenario matrix avoids false precision. Create slow, middle and fast cases with explicit assumptions rather than vague moods. For each case, state capability growth, infrastructure expansion, research automation, verification throughput and adoption speed. The purpose is not to choose a winner; it is to see which decisions are robust across several plausible futures and which depend on one narrow forecast.

Final Synthesis: Constraints and Forecasts Must Stay Revisable

Constraints and forecasts must remain revisable. A new algorithm can loosen a compute bottleneck; a power shortage can tighten an infrastructure bottleneck; better synthetic-data verification can change a learning constraint. The correct model is therefore a living map of rates, buffers and evidence. Super Intelligence (SI) preparation is strongest when it can adapt without pretending uncertainty has disappeared.

Build a capacity map by listing every resource required for one completed unit of progress. Include compute, high-quality data, electricity, cooling, specialist labour, hardware lead time, verification and deployment. Mark current utilisation and replacement time. The limiting resource is not necessarily the most visible one. When a constraint tightens, it can become the new rate limiter even if every other input continues improving.

A scenario matrix avoids false precision. Create slow, middle and fast cases with explicit assumptions rather than vague moods. For each case, state capability growth, infrastructure expansion, research automation, verification throughput and adoption speed. The purpose is not to choose a winner; it is to see which decisions are robust across several plausible futures and which depend on one narrow forecast.

Sensitivity analysis changes one assumption while holding others fixed. Double hardware efficiency, halve data quality, delay new fabrication capacity, or increase verification cost. Observe how much the conclusion moves. The assumptions with the largest effect deserve the most monitoring. A model that swings wildly when one uncertain input changes should not be presented with a narrow confidence band.


A Timeline Is a Probability Distribution, Not a Date on a Calendar

Super Intelligence timelines are often reported as a single year, but a serious forecast is a distribution over possible dates. A forecaster may assign some probability to early arrival, more to a central range, and a long tail to later dates or non-arrival. Compressing that distribution into one median can hide most of the uncertainty.

For Super Intelligence (SI), readers should therefore ask for the full range, the definition being forecast and the conditions that would move the forecast.

Definitions Change Timeline Answers

AI Impacts’ large 2023 survey of 2,778 AI researchers found that timeline answers changed depending on how the milestone was phrased. “High-level machine intelligence” and “full automation of labour” produced meaningfully different distributions. This is not noise; it shows that the event being forecast matters.

A forecast of AGI, weak AGI, economically transformative AI, full automation or ASI can refer to very different thresholds. Never compare dates until the definitions are aligned.

Expert Surveys Measure Beliefs, Not Future Facts

Expert surveys are useful because respondents understand the field’s technical context. They are still opinions under uncertainty. Experts can share blind spots, anchor on recent trends or interpret ambiguous definitions differently.

The value of a survey is therefore descriptive: it tells us how a defined population of experts distributed their beliefs at a particular time. It does not convert the median into a law of technological development.

Forecasting Platforms Provide a Different Kind of Signal

Prediction platforms such as Metaculus aggregate forecasts that update as evidence changes. As a late-September/early-October 2026 snapshot, Metaculus questions on general AI were centred in the early 2030s, while a separate question about the interval from weak AGI to superintelligence had a central estimate measured in a few years rather than decades. These values are dynamic and should be treated as dated crowd forecasts, not factual predictions.

The useful feature is updateability. A crowd can move when new models, benchmarks, hardware constraints or policy changes arrive.

Different Forecaster Populations Answer Different Questions

AI researchers, professional forecasters, industry executives, prediction-market participants and the general public do not have the same information or incentives. A researcher may know technical bottlenecks deeply but have little forecasting training. A superforecaster may be well calibrated but less familiar with frontier model development.

Comparing populations can reveal where disagreement comes from. The goal is not to choose a tribe; it is to identify which assumptions differ.

Conditional Forecasts Are More Useful Than Unconditional Dates

Instead of asking only “When will SI arrive?”, ask conditional questions: If long-horizon task performance keeps doubling at the current measured rate, what follows? If data becomes the dominant bottleneck, how does the date move? If AI automates half of frontier AI R&D, how does the forecast change? If grid build-out constrains compute, what happens?

Conditional forecasting makes the causal model visible. It tells readers which observations should update the timeline.

Capability Trends Should Be Used Carefully

Epoch AI tracks model, benchmark, chip and data-centre trends over time. METR tracks agent task horizons. Stanford tracks benchmark performance, costs and industry activity. These empirical series can inform forecasts, but extrapolation requires assumptions about whether the trend continues, bends or hits a bottleneck.

The strongest forecast states the extrapolation rule and its failure conditions.

Recent Acceleration Should Update Beliefs—But Not Erase Uncertainty

Stanford’s 2026 AI Index documents rapid gains and unusually fast saturation of some benchmarks. Anthropic’s 2026 reporting describes growing use of AI inside AI R&D. These are legitimate reasons to revisit older timelines. They are not proof that every ambitious capability will arrive on the same schedule.

Good forecasting updates on acceleration without pretending that uncertainty disappeared.

Forecast Calibration Matters More Than Confidence

A calibrated forecaster who says 70% should be correct roughly 70% of the time across comparable predictions. Calibration can be evaluated only across repeated forecasts with resolved outcomes. One confident prediction that later turns out right does not prove forecasting skill.

For SI, many headline milestones are unresolved, so calibration must often be assessed using shorter-horizon proxy forecasts.

Base Rates Still Matter in Fast-Moving Fields

AI has repeatedly surprised forecasters, but historical technology development still offers useful base-rate information about research cycles, infrastructure build-out, adoption and regulation. Base rates should not override strong direct evidence; they should prevent one exciting result from resetting the entire forecast without enough support.

The best forecasts combine inside-view mechanism with outside-view rates.

Worked Example: Two Forecasters With the Same Median

Forecaster A assigns 50% probability before 2032 and 50% after 2032, with a very wide tail. Forecaster B assigns almost all probability between 2031 and 2033. Both can report the same median year while expressing radically different uncertainty.

This is why a single date is an incomplete representation of belief.

Worked Example: A Forecast That Updates on Research Automation

Suppose a forecaster believes AI-led AI R&D is the dominant accelerator. When reliable evidence shows research agents completing a larger fraction of frontier R&D tasks, the forecaster moves probability toward earlier dates. If subsequent evidence shows verification and hardware bottlenecks dominating, the distribution shifts later again.

This is good forecasting behaviour: update the mechanism, not the narrative.

A Practical SI Timeline Checklist

Before trusting a timeline, ask: What exactly is the milestone? Who made the forecast? When was it made? What probability distribution was given? What evidence drove it? What bottlenecks are assumed away? What observations would move it earlier or later? Has the forecasting method been calibrated on resolved questions?

If those answers are missing, the date is more headline than forecast.

RFE Closure: Timelines Must Remain Dated, Conditional and Revisable

The problem is turning uncertain probability distributions into apparently certain years. The function of an SI timeline is to help people reason about preparation under uncertainty. The receiver is the person or institution deciding when to invest, adapt or monitor. Evidence includes capability trends, research automation, compute, data, infrastructure and calibrated forecasting performance.

The exit condition is to revise the forecast when new evidence changes the causal model. A timeline that never updates is not a forecast; it is a belief preserved against data.

Continue the Super Intelligence (SI) Series

Next: the strongest constraints and sceptical arguments—what could prevent or delay SI even if current AI progress remains rapid.

Scenario lab: prepare for SI while protecting human flourishing

A timeline becomes useful when it changes a decision. Work through the packet below, choose an action, then revise it when the evidence changes. Your output is a conditional decision record, not a predicted arrival year.

Scope: All organisations, observations, thresholds, budgets and results in this lab are fictional teaching inputs. These three scenarios have no assigned probabilities or arrival dates. They are neither exhaustive nor equally likely by implication. The near-term service decision is a practice setting for preparation under SI uncertainty; it does not establish that present systems are superintelligent.

The OECD’s strategic-foresight programme describes exploring multiple plausible futures and their implications for decisions today. Here, capability, governance and access vary separately. Rapid capability does not automatically produce accountable institutions or widely shared benefits.

Reading route: compare the three futures · inspect the indicator cards · work the decision · try the changed case

1. Fix the decision and the values before choosing a future

The fictional Harbour adult-learning service supports 100 adults. It must choose a preparation plan for its next 12-week term, within 100 planning units. A unit is an invented budget measure, not currency or a market price. Its current staffed service can help 60 adults meet a learning goal during the term. All 100 can still seek advice; goal completion and access are different measures.

A verified goal means that an adult demonstrates an agreed skill on a new task without the answer being supplied. More generated worksheets, logins or completed chatbot conversations do not count as more verified learning. No plan may use this exercise to allocate jobs, benefits or other high-stakes entitlements.

Human-flourishing criteria: The service values learning and independent capability; dignity and voluntary choice; fair access; safe, private and contestable support; sustainable staff work; and responsible use of resources. The OECD AI Principles provide a relevant values framework, including well-being, rights, transparency, safety and accountability. The following operational floors are Harbour’s own fictional choices, not OECD-prescribed thresholds or a complete definition of flourishing.

  • Agency and dignity: Keep a staffed alternative and let adults decline AI without losing access. Consult learners about which goals are worthwhile; a completion count cannot decide that for them.
  • Safety, privacy and redress: Do not put identifiable learner records into the trial. Keep a named human reviewer, a usable challenge route and a tested way to pause the tool. A credible serious incident stops the affected use pending investigation.
  • Fair opportunity: Device, disability or language barriers must have a funded alternative. A trial difference above 10 percentage points between the two predefined access groups blocks expansion until investigated and addressed.
  • People and resources: Pay for staff preparation, keep weekly overtime within two hours per adviser, and fit the 100-unit cap. Record computing use and avoid unnecessary generation; an unmeasured environmental footprint stays unknown rather than becoming “zero.”

These are floors, not points to trade away for a higher average. Above them, compare independent learning, learner experience, relationships, staff workload and resource use. Invite disagreement about the chosen values and trade-offs; publish any change to the decision rule before using it to justify a preferred option.

2. Three contrasting futures to test the same decision

Scenario A: capable, accountable and widely accessible. AI improves quickly on Harbour’s bounded tasks. Independent checks remain available; the supplier supports audit, exit and appeal arrangements. Devices, accessible interfaces and language support are funded. Opportunity: staff can spend more time on discussion while appropriate AI practice expands. Remaining risk: apparently convenient help may weaken unaided learning, so independent tasks still matter.

Scenario B: capable, concentrated and difficult to challenge. Task capability improves as quickly as in A, but one supplier controls access and offers weak audit and export arrangements. A price increase and inaccessible interface leave some adults without a usable route. Opportunity: some supported learners receive strong practice. Risk: aggregate gains hide exclusion and dependency. A high benchmark score does not repair these governance and access problems.

Scenario C: uneven capability with careful local governance. Novel-task performance improves slowly and varies by task. Harbour has accountable local oversight, but devices and support capacity remain constrained. Opportunity: staff training, accessible materials and narrow validated tools are still useful. Risk: an expensive broad licence absorbs resources before it delivers useful learning. Slower progress does not remove the need for preparation.

Contrast check: A and B hold fast capability roughly constant while changing governance and access. A and C both permit responsible local decisions but differ in capability and practical access. Real futures can combine features of all three; this packet does not prove that these axes move together.

3. Supplied indicator cards: observation, inference and trigger

Read these cards as fictional evidence available at the first review. They do not identify the “true scenario.” Each card specifies what was measured, what remains uncertain and what action it can justify.

Card C, capability. On 100 pre-specified new practice tasks from Harbour’s intended use, independent staff checks accept 88 outputs; 12 require correction. The same rubric, task mix and model version are recorded. This suggests some bounded usefulness. It does not measure general intelligence, safe autonomous operation or learner understanding. Harbour’s expansion rule requires at least 95 accepted outputs out of 100 in each of two fresh checks, alongside every other floor. With 88, retain a small supervised trial; do not expand.

Card G, governance. A pause drill works; an adviser can receive challenges; and the supplier permits inspection of the trial outputs. Data-export and contract-exit tests have not yet been completed. These observations support a reversible trial, not a claim of complete accountability. Do not make a long commitment until both tests work and responsibility for complaints is documented.

Card A, access and learning. Two equal-sized trial groups were defined before results were collected: 20 adults with reliable digital access and 20 needing device or interface support. With the same goal rubric, 16 in the first group and 12 in the second demonstrate the agreed skill independently. Rates are 80% and 60%; the gap is 20 percentage points. This exceeds the 10-point expansion ceiling. It is a small, non-randomised trial: the difference is a warning to investigate, not proof that the tool caused it or that a group lacks ability.

Card W, work and resources. Advisers report three additional overtime hours each per week against the two-hour ceiling. Total planned spending is still within 100 units, but the current trial depends on unpaid effort. Actual computing-energy use has not been measured. Move preparation into paid time, reduce the trial’s volume and record the missing resource evidence. A nominal budget pass cannot cancel a workload failure.

Card H, headline. A supplier demonstration reports “twice as many tasks completed” on its own selected examples, without the task list, correction time or access breakdown. Record it as an unverified lead. Ask for a comparable test; do not replace Cards C, A or W with the headline or infer an SI arrival date from it.

4. Work the robust-versus-contingent choice

Three plans are available. M, maintain: spend 100 units on the current staffed service. R, reversible preparation: spend 40 units on paid staff learning, independent evaluation, access support and a pause/appeal procedure; retain 60 units to allocate at the review. The initial trial uses fictional or non-identifying material, stays small and has a staffed fallback. K, commit early: bind 90 units to a broad supplier package and retain 10 for support. Its funded staffing and access arrangements work only under the favourable conditions in A; the commitment is expensive to reverse.

The following are supplied scenario outcomes, not empirical estimates or expected values. Each plan’s total term expenditure is at most 100 units. R includes the later allocation of its reserve and repair of the initial trial failures; it does not mean continuing that trial unchanged.

PlanA: capable and inclusiveB: capable and concentratedC: uneven capability
M: staffed baseline60 verified goals; floors pass60 verified goals; floors pass60 verified goals; floors pass
R: reversible preparation82 verified goals; floors pass64 verified goals; floors pass68 verified goals; floors pass
K: broad commitment92 verified goals; floors pass74 verified goals; access and redress fail48 verified goals; workload fails

Worked answer: Choose R as the robust starting plan for this packet. It remains admissible in all three futures and its lowest supplied goal count is 64, compared with M’s 60. K is attractive in A but cannot pass the agreed floors in B or C. Its 74 in B does not buy permission to exclude people or remove redress. We have not averaged the columns: no scenario probabilities were supplied.

Robust does not mean best in every future or harmless. R gives up 10 goals relative to K in A, and its 40-unit preparation cost could have supported immediate service. Compared with R, M forgoes 22, 4 and 8 goals in A, B and C respectively. Those differences are simple consequences of the fictional inputs, not general effect sizes. The decision requires a check that the stated floors really can be maintained.

Contingent action: Reserve a 20-unit expansion from R’s remaining 60 only if fresh capability checks, access results, workload, exit tests and learner feedback all pass. Otherwise use that tranche to repair the limiting condition or maintain staffed support; keep the other 40 flexible. An expansion trigger concerns this task and service. It does not declare that Scenario A has arrived or that SI is near.

Sensitivity check: If R’s worst-case outcome fell from 64 to 55 while all floors stayed equal, its worst-case learning advantage over M would disappear. Under a rule that prioritises the lowest scenario outcome, M would then win 60 to 55. Another defensible rule might value learning from the trial, but it must state that value and cost openly. “Choose the flexible plan” is not a substitute for checking the inputs.

5. Warning evidence must change the action

At the next fictional review, a new release passes 97 of 100 outputs in one fresh capability check. The headline is encouraging, but the access breakdown is now 18 of 20 adults with reliable access and 10 of 20 needing support: 90% versus 50%, a 40-point gap. An adviser also discovers an incorrect course instruction that had been copied into a learner plan without its required check. The second fresh capability check and the export test remain outstanding.

Update: Do not release the 20-unit expansion. Pause use of the affected planning feature, check who received the incorrect instruction and correct their plans through the staffed route. Allocate the 20 units instead to the access and checking bottlenecks: 12 for accessible supported sessions and 8 for paid review and fresh testing. The accounting remains 40 already committed + 20 repair + 40 reserve = 100 units. The packet assumes these costs fit; a real service would verify them.

Record a narrower conclusion: bounded output performance improved on one check, while equitable access and the checking workflow deteriorated. Investigate task mix, support availability and whether the instructions were usable; do not blame learners or claim causality from the rates alone. Return to expansion only after the corrective work and all gates have passed. If they cannot pass within budget, use the staffed fallback already funded within R rather than quietly relaxing the floors. This does not buy the separate 100-unit M plan again after preparation and repair spending.

This is an action update, not just a rewritten forecast. The evidence changes permissions, spending and support for affected people even though it cannot resolve the SI timeline.

6. Repair a mistaken forecast and the reasoning beneath it

Mistake in the original decision log: “The vendor doubled task completion, so all 100 adults will complete their goals next term; we can commit before testing access.” This assertion confuses vendor task throughput with independent learning, turns a selected demonstration into a local causal estimate, omits governance and access, and treats a single possible outcome as certain.

Repair: Keep the original statement visible and date the correction. Replace it with: “We do not have evidence for 100 independent goal completions. Our local capability check is 88/100, access-group goal rates are 16/20 and 12/20, and staff overtime exceeds our floor. We choose reversible preparation and will test the specified conditions before expanding.” At the later review, append the new 97/100 check and the worsening access gap; do not silently overwrite the earlier inputs.

The repair changes the causal model from “more generated tasks → universal learning” to “bounded useful output + independent practice + accessible support + workable oversight → a testable possibility of improved learning.” This remains a hypothesis, not a guarantee. One revised decision cannot establish forecast calibration. That requires a record of well-defined, resolved forecasts rather than a success story selected after the event.

7. Changed case: make your own decision before reading the answer

Harbour’s neighbouring service, Cedar, has only 70 planning units, intermittent connectivity and no evidence that the broad supplier package works offline. A separate small offline tool scores 96/100 and 97/100 on two fresh local checks. Its export, pause and challenge drills pass. The reliable-access group achieves 17/20 goals and the supported-access group 16/20; advisers’ weekly overtime is one hour. No serious incident is reported, and a staffed alternative, paid preparation and appropriate privacy arrangements are funded in the plans marked “pass” below.

Supplied alternatives: N, continue the staffed service, costs 70 and produces 54, 54 and 54 verified goals across A, B and C; all floors pass. L, a reversible offline-assisted plan, costs 68 and produces 63, 57 and 59 verified goals; all floors pass. K, the earlier broad package, still costs 100, produces the earlier scenario results and has untested offline operation. The figures include all support and preparation; no extra cost may be assumed away.

Your task: Calculate the access gap. Apply the budget and human-flourishing floors before comparing outcomes. Choose a plan under the same worst-case rule, specify one condition that would reverse it and explain why strong capability evidence does not make the 100-unit package affordable. Write your answer before continuing.

8. Independent transfer answer and a reusable decision record

Answer: Cedar’s rates are 85% and 80%, a 5-percentage-point gap, within the fictional 10-point ceiling. Both fresh capability checks clear 95/100; overtime of one hour is within two; the specified governance drills pass. N and L meet the supplied floors and budget. K is excluded because 100 exceeds 70 by 30 units, before considering its offline uncertainty or other failures. Choose L: its lowest supplied outcome is 57 rather than N’s 54, and it leaves 2 units unspent.

This answer changes the technology and scale to match the new conditions; it does not copy Harbour’s 40/60 allocation. If a fresh access gap rose above 10 points, an incident required pausing the tool, or total L cost rose above 70, stop expansion and reassess. Before committing funds, N remains an alternative if L no longer qualifies. After spending on L, use its already-funded staffed fallback and recalculate remaining resources; do not assume that the full 70-unit N plan can be purchased again. Passing two checks supports this bounded choice; it does not prove universal reliability or resolve a long-run SI scenario.

Decision record to complete: Receiver: name the people affected and the accountable decision-maker. Function: state the choice, horizon and budget. Values: identify what may not be traded away and how disagreements will be heard. Evidence: distinguish observed data, scenario assumptions and unknowns. Decision: record the robust base action, contingent action, trigger and fallback. Exit: name who reviews the next evidence, when they review it and what stops or reverses the action.

You have completed the lab when you can justify a choice from the supplied packet, change it when warning evidence arrives, repair an overclaim without hiding it, and solve the changed case using its own constraints. The lesson is preparation that remains answerable to people as well as to measured capability.