Super Intelligence and Civilisation OS: A Proposed Lens for Repair, Buffers and Human Capability

Super Intelligence changes more than the speed of producing answers. It can change who understands a process, who can repair it, how replacements learn, and how long an institution can continue when its usual assistance is unavailable. eduKateSG’s proposed Civilisation OS lens brings those questions into the same conversation. Its value should be judged by whether it helps people notice important dependencies and make better-supported decisions.

This article is an applied reading exercise, not a claim that Civilisation OS is a universally validated science. The framework’s operating-system labels are eduKateSG’s conceptual vocabulary. They are not installed software, agreed scientific units, or evidence that a society has a measurable countdown to collapse. We will use a complete fictional institution, calculate a few narrowly defined quantities, repair a misleading conclusion, and then test the reasoning on a different institution.

Super Intelligence is the series title. Artificial superintelligence, or ASI, refers here to a hypothetical system with broadly superhuman intellectual capabilities. None of the fictional cases establishes that present systems are ASI, predicts an arrival date, or assumes that machine capability determines human priorities. All institutions, people, records, timings and operating choices in the cases are invented for teaching. The header image is illustrative and does not depict the fictional institutions.

1. Begin with the service people need to preserve

Imagine an institution that becomes exceptionally good at producing explanations. Its staff answer more questions, its documents look clearer, and its queues shorten. These are meaningful improvements. Now imagine that only one person knows how the explanations are checked, trainees have stopped practising the underlying reasoning, and the fallback instructions exist only inside the unavailable service. The improvement in output has not answered the question of continuity.

It would also be a mistake to reverse the story into a universal warning against automation. An institution might use assistance to free experienced staff for teaching, make procedures easier to find, produce realistic practice cases, and shorten the time needed to diagnose an error. It could become more productive and more repairable together. Which account is true depends on the arrangement, the evidence and the particular function being discussed.

The first question is therefore concrete: what must continue, for whom, and under which conditions? A community learning centre might promise a usable feedback brief for each scheduled learner. An archive might promise accurate catalogue records in two languages. A transport operator would have a different, much more consequential service boundary. These functions should not be treated as interchangeable examples of a single generic score.

Within this article, continuity means meeting the fictional institution’s stated service requirement during a defined interruption. It does not mean perfect performance, survival of a nation, or freedom from all harm. Repair means restoring the specified function and verifying that the relevant defect has been addressed. Human capability means a person’s demonstrated ability to perform a particular part of the work under stated conditions. Each definition has a boundary.

These boundaries make the proposed lens useful. Rather than asking whether the institution is intelligent in general, we can ask whether a learner receives correct feedback, whether a tutor can recognise a bad explanation, whether a replacement can take over, and whether there is time to recover before commitments are missed. An impressive system description earns its place only when it improves these more ordinary questions.

Back to contents · Next: 2. What Civilisation OS proposes, and what this article accepts

2. What Civilisation OS proposes, and what this article accepts

The canonical Civilisation OS framework proposes a loop connecting Mind, Education, Governance, Production, Constraint, diagnosis and repair. Its introductory What is Civilisation OS? page explains the intended movement between capability, coordination, material outcomes and limits. This article uses those names as prompts for examining relationships within a bounded institution.

Mind directs attention to judgement and the conditions under which people can think. Education directs attention to how capability is learned and renewed. Governance directs attention to purposes, authority, incentives and correction. Production directs attention to the service being delivered. Constraint directs attention to the resources and conditions that restrict what is possible. Diagnosis and repair ask how problems become visible and how a response changes the underlying process.

The framework also proposes a Human Regenerative Lattice, or HRL: people in roles, the relationships between those roles, and the learning pathways that produce future competence. Its lattice-buffer language asks how much time remains before a lost capability prevents a required function. Those are productive questions to investigate. The terminology itself does not supply measurements or establish a universal causal theory.

Some canonical passages use stronger language about physics, inevitability, scoring and prediction. Here, those statements are treated as proposals or heuristics that would require independent definition and validation before being used as scientific claims. We do not assume universal zero-to-ten thresholds, average unlike layer scores into a physical quantity, or infer a six-to-twenty-four-month societal trajectory from this exercise.

That qualification follows the Super Intelligence master guide’s distinction between eduKateSG operating models, demonstrated capability and hypothetical futures. A framework can be worth exploring without claiming authority it has not earned. Readers can retain its useful questions while rejecting an unsupported equation, a premature prediction, or an interpretation that fits the story better than the records.

The goal is not to rename every existing discipline. A learning study, a staffing analysis and a service-recovery exercise each have their own methods. Civilisation OS is used here to place their questions alongside one another. Whether this combination improves decisions beyond simpler approaches remains an empirical question, not a conclusion supplied by the framework’s name.

Back to contents · Next: 3. Keep four kinds of statement visibly separate

3. Keep four kinds of statement visibly separate

An observation describes something recorded under specified conditions. For example, twelve fictional trainees attempted a task without generative assistance and eight met the stated criterion. The observation includes the denominator, the conditions and the criterion. It does not automatically tell us why four trainees did not meet it, whether they would improve next week, or whether the criterion captures everything that matters.

A calculation combines defined quantities using a stated relationship. If a centre has 180 usable reserve briefs and consumes them at a net rate of 60 briefs per working day, the simplified reserve lasts three working days. The units cancel correctly. The answer still depends on assumptions: every reserve brief must be usable, daily demand must follow the stipulated pattern, and no other constraint may stop the service first.

A normative priority says what the institution should protect. Giving each scheduled learner adequate feedback, preserving staff learning time, or avoiding unfair exclusion are choices about purposes and obligations. Numbers can illuminate their consequences, but cannot select them on their own. A faster system is not automatically a better educational system if speed is purchased by abandoning an important learning purpose.

A hypothetical scenario asks what would follow if a specified condition were true. If future AI could reliably perform a broader range of repair tasks, some staffing needs might change. If all fallback services depended on the same infrastructure, an interruption could still affect them together. Neither conditional statement is evidence that the proposed future already exists or that its probability has been measured.

Confusion appears when a sentence silently moves between these categories. “The automated team produced more, so human capability is no longer necessary” moves from an output observation to a broad value judgement. “Repair scored seven and drift scored five, so collapse is impossible” turns subjective labels into a predictive law. “ASI could solve this, therefore we need no buffer” treats a hypothetical possibility as an available resource.

A disciplined analysis marks the transition instead. It reports the output increase, states the remaining uncertainty about independent skill, explains why the institution values continuity, and tests a conditional interruption. This is not excessive caution. It is how a reader can reproduce the conclusion and see which new fact would change it.

Back to contents · Next: 4. Human capability is a pathway, not a headcount

4. Human capability is a pathway, not a headcount

Counting employees tells us little about which functions can continue. Eight staff members may include one specialist, several people who depend on that specialist, and a colleague who is technically qualified but unavailable on the relevant day. The same number can describe a robust team or a fragile one. Capability has to be connected to a task, a time, an authority and an actual opportunity to act.

For the fictional cases, a capability pathway has five stages. Someone encounters examples; practises the underlying work; receives feedback; demonstrates the task under appropriate conditions; and later helps another person learn it. These stages need not occur in a rigid sequence. A novice can contribute useful work while still being supervised. But completed outputs should not be mistaken for evidence that all five stages are occurring.

AI assistance can enter at every stage. It may generate practice variations, explain a difficult passage, organise a mentor’s feedback, or help a qualified worker examine an unusual case. It may also make it easy to submit plausible work while skipping the reasoning the exercise was meant to develop. The relevant distinction is the learner’s activity and later capability, not merely whether a machine was present.

Replacement is similarly specific. A new colleague may learn routine processing quickly while needing much longer to recognise rare exceptions. A procedure may be easy to read but difficult to enact because it omits the judgement used by experienced staff. A mentor may have excellent knowledge but no protected time to transfer it. In the HRL vocabulary, these are questions about the regeneration pathway; in ordinary language, they are questions about whether the next person can really take over.

We should resist turning this into an assessment of human worth. A person who cannot perform one fallback task is not a failed person or an inferior member of the institution. Roles, training opportunities, disability accommodations, experience and available support affect performance. The practical response is to clarify what support or redesign the function needs, not to attach a dramatic civilisational label to an individual.

The same restraint applies to experienced workers. Asking them to document, supervise, verify and repair every system can create an impossible burden. Their knowledge is valuable, but it does not create extra hours. A credible capability pathway includes the resources needed to teach and practise, rather than assuming that competence can be transferred at no cost.

Back to contents · Next: 5. External research can support a bounded question

5. External research can support a bounded question

One relevant primary study is Bastani and colleagues’ field experiment on GPT-4-based mathematics support in a high school in Turkey. The researchers compared different assistance designs and later unassisted performance. They reported that benefits during assisted practice did not necessarily become better performance without the tool, while the tutor design mattered. This supports keeping assisted output and independent learning separate. It does not establish the result for every age, subject, institution or AI system, and it does not validate Civilisation OS. See the primary research article.

The paper’s existence should not be used as a shortcut to diagnose the fictional trainees below. Their records are invented and their conditions differ. We use the research to justify a question worth asking: what can the learner do after assistance changes? The answer for a particular institution would require evidence from that institution, collected with an appropriate design and respect for the people involved.

For recovery planning, NIST’s Contingency Planning Guide for Federal Information Systems, updated Revision 1 discusses assessing operational priorities, recovery requirements and exercises. Its scope is information-system contingency planning. It offers a relevant example of defining the function and recovery conditions rather than assuming a backup exists because a document says so. It does not endorse eduKateSG’s terminology or establish a universal institutional survival threshold.

Our calculations below are simpler than a professional continuity analysis. They use stipulated demand, reserve quantities and restoration times so the reader can inspect the reasoning. They do not account for every practical complication, and they should not be adopted as a safety, legal, clinical or infrastructure operating standard. Their educational purpose is to expose assumptions that a headline productivity comparison would otherwise hide.

This separation prevents validation by association. A credible paper about learning does not prove a larger model of civilisation. An established recovery guide does not convert a metaphor into physics. A numerical example does not demonstrate predictive accuracy merely because its arithmetic is correct. Each source supports a bounded claim; each inference has to earn its own support.

It is equally important not to demand that a single study answer every question before acting thoughtfully. An institution can protect time for practice, ask people to explain a process, and rehearse a modest interruption without pretending these choices settle the future of AI. Proportionate learning can begin before grand theories are resolved.

Back to contents · Next: 6. The Larch packet: purpose, people and comparable work

6. The Larch packet: purpose, people and comparable work

Larch Civic Learning Centre is a fictional adult-learning institution. It supports 120 adult learners with eight tutors and two coordinators. Twelve of the adult learners also take part in a supervised tutor-development pathway. They are trainees, not additional independent fallback staff. The centre’s service promise for the exercise is to supply 100 usable feedback briefs per working day during the specified interruption.

A feedback brief addresses one submitted practice task. For this packet, briefs have a common length limit, a common checking rubric and one assigned task category. A routine brief concerns a familiar task with a standard checking path. An exception brief requires interpretation of an unusual response or conflicting evidence. These categories are assigned before quality checking and do not change between the comparison periods.

The centre records two consecutive four-week periods, each containing twenty working days. In the first period, tutors work with reference material and ordinary editing tools. In the second, they add an AI drafting assistant and a revised work allocation. We will call these the reference period and the accelerated period. Both use 240 staff-hours counted across drafting, checking and the routine correction work included in the packet’s time boundary.

The 240 hours are not the centre’s entire payroll or every learner’s study time. They define the comparison activity. Time spent creating the experiment’s records is stipulated outside that boundary in both periods. This makes the arithmetic possible but limits economic conclusions. A real evaluation would need to account for setup, procurement, training, supervision and other relevant costs rather than quietly omitting them.

The quality record is a complete audit of the listed briefs, using the same fictional rubric. Each affected brief contributes at most one counted defect, even if it contains several mistakes. A counted defect makes the brief unusable until corrected. The packet does not distinguish severity within this count; that limitation will matter. The staff-hours and output quantities are supplied facts within the fictional exercise, not estimates generated by a model.

This comparison is observational, not randomised. The work mix changes, the periods occur at different times, and work allocation changes along with the assistant. We can describe the recorded differences. We cannot isolate an AI-only causal effect or claim that Larch represents every educational institution.

Back to contents · Next: 7. The Larch packet: outputs, errors and learning observations

7. The Larch packet: outputs, errors and learning observations

In the reference period, Larch completes 2,400 briefs: 1,800 routine and 600 exception briefs. The audit finds 18 defective routine briefs and 42 defective exception briefs, for 60 affected briefs overall. In the accelerated period, Larch completes 3,000 briefs: 2,700 routine and 300 exception briefs. The audit finds 27 defective routine briefs and 48 defective exception briefs, for 75 affected briefs overall.

The correction ledger is separate from the output audit. At the start of the reference period, 40 defective briefs from earlier work remain unresolved. During the reference period, the 60 newly identified defects enter the ledger and 72 defects are corrected and independently rechecked. During the accelerated period, 75 newly identified defects enter and 60 are corrected and independently rechecked. There are no duplicated entries or reopened corrections in this simplified ledger.

The twelve trainees attempt an independent explanation-and-checking task before the accelerated period and an equivalent-form task afterwards. Eight meet the stated criterion before and six afterwards. The criterion requires an adequate explanation, correct use of the supplied evidence and identification of a deliberately flawed answer. The tasks allow ordinary reference notes but no generative assistant. No individual scores, attendance details or explanations for the difference are supplied.

The staff fallback record is different again. Before the accelerated arrangement, four tutors had recently demonstrated the full offline feedback procedure in a drill. At the later drill, two tutors demonstrate it successfully; the remaining six have no current successful demonstration in that drill record. The packet does not say that all six have permanently lost the skill. It tells us how much current evidence of fallback competence is available.

For the specified outage, a measured team exercise supplies the usable fallback rates: 70 briefs per working day under the reference arrangement and 40 under the accelerated arrangement. These are team rates, not a rate assigned to every tutor. The exercise already includes coordination and checking within its narrow scope. Do not multiply either number by eight.

No record in this packet proves that assistance caused the trainee result, that staff are generally less capable, or that all error types are equally harmful. The distinction between what is supplied and what is missing is part of the task. A good response carries these limits into its recommendation rather than hiding them in a footnote.

Back to contents · Next: 8. The Larch packet: interruption, reserve and repair conditions

8. The Larch packet: interruption, reserve and repair conditions

At the beginning of working day one, a hypothetical fault makes the normal drafting process unavailable. Larch has 180 already checked, case-matched reserve briefs that can be used for the scheduled demand. For this simplified exercise they do not expire, none is duplicated, and all are available offline. They are completed service units for known tasks, not generic templates that still need an unspecified amount of work.

Demand is 100 briefs per working day. Work and demand are treated as flowing evenly during each day, so fractional days can be calculated. The reference fallback supplies 70 usable briefs per day; the accelerated fallback supplies 40. There is no overtime, demand cancellation, borrowing from another institution, or additional reserve production beyond those fallback rates during the outage.

The restoration path takes four working days from the start of the interruption, including detecting the relevant fault, preparing the correction, validating it and returning the process to service. Normal service therefore becomes available at the start of working day five. Four days is a stipulated scenario duration supported by one fictional rehearsal; it is not a guarantee about every future incident or a universal repair time.

The service requirement is also stipulated: Larch should meet all 100 daily briefs throughout those four working days. Missing that requirement is called a service shortfall in this article. It is not called societal collapse. The institution might respond to a shortfall in many ways, but the first calculation excludes those responses so that both arrangements face the same stated test.

For a possible redesign, the centre has eight discretionary staff-hours available in the next week. Preparing and checking 120 additional case-matched reserve briefs would take four of those hours, using material and tasks already available. Four hours can also be allocated to supervised fallback practice. The packet supplies no result from that future practice; additional competence cannot be credited before it is demonstrated.

These assumptions are deliberately visible because each could change the answer. If reserve briefs cannot match actual demand, the buffer is overstated. If restoration includes an uncounted validation stage, the time is understated. If staff are diverted to urgent work elsewhere, fallback capacity is overstated. A useful framework encourages these objections instead of protecting its initial verdict.

Back to contents · Next: 9. Worked answer: acknowledge the productivity gain

9. Worked answer: acknowledge the productivity gain

Under the defined activity boundary, reference throughput is 2,400 briefs divided by 240 staff-hours, or 10 briefs per staff-hour. Accelerated throughput is 3,000 divided by 240, or 12.5 briefs per staff-hour. The recorded increase is 2.5 briefs per hour. Relative to the reference value, that is 2.5 divided by 10, giving a 25% increase.

We should not obscure that improvement simply because later findings are less comfortable. More usable educational feedback can matter to learners and staff. A critical analysis that refuses to acknowledge gains is no more reliable than a promotional analysis that ignores weaknesses. The first responsibility is to describe each measured result accurately.

Subtracting the defective briefs gives 2,340 initially usable briefs in the reference period and 2,925 in the accelerated period. Dividing by the same staff-hours gives 9.75 and 12.1875 initially usable briefs per staff-hour. Their ratio is again 1.25. In this particular packet, excluding initially defective briefs preserves the 25% aggregate improvement because both aggregate defect proportions are the same.

This is not yet a complete quality-adjusted economic analysis. The count treats every usable brief as equivalent despite differences in task category, depth or learner need. It does not place a value on missed exception work, later misunderstandings or the cost of carrying unresolved defects. “Initially usable briefs per recorded staff-hour” is the precise description. “Overall institutional intelligence rose by 25%” would be an unsupported transformation.

The result also does not tell us whether the assistant alone produced the gain. The accelerated period includes more routine work and fewer exception briefs. If routine work is easier, a different mix could increase average output even without a better tool. We would need comparable task distributions, a stronger study design or a carefully specified adjustment to separate these explanations.

The honest first conclusion is therefore positive but narrow: Larch recorded more output, including more initially usable output, per counted staff-hour in the accelerated period. Keep that result. Then continue to the capability, error and interruption questions rather than forcing a single verdict too early.

Back to contents · Next: 10. Worked answer: map the relationships without inventing scores

10. Worked answer: map the relationships without inventing scores

The Production question concerns the increase from 2,400 to 3,000 briefs and the kinds of briefs being delivered. The Education question concerns what the twelve trainees are learning and how tutors retain their checking ability. The Governance question concerns why work was allocated differently, who accepts a corrected brief, and whether the service promise remains appropriate. These are related questions with different evidence.

The Mind prompt should be used carefully. We have no evidence about staff mental health, motives or general judgement. The packet gives task performances and staffing conditions, not diagnoses of people. A responsible analysis might ask whether the drill allowed sufficient time, whether instructions were clear and whether competing duties affected performance. It should not infer a damaged mindset from two counts.

Constraint directs attention to the eight available discretionary hours, the daily demand, the four-day restoration path and the usable reserve. These limits prevent an easy but empty recommendation such as “increase output, train everyone and maintain unlimited backup.” The institution has to choose a feasible allocation. Adding the names of more layers does not expand the available time.

Diagnosis appears in the complete quality audit, the correction ledger and the drills. Repair appears when a defect is actually corrected and rechecked, not merely when a plan is written. In HRL terms, the staff fallback record and trainee pathway reveal whether the institution can reproduce a needed role. In ordinary terms, someone needs to know the work, show that they can do it, and have time to help the next person learn.

The most important connection is a hypothesis: if more routine work displaces exception practice and mentoring, current output could improve while replacement capability weakens. The observations make that hypothesis worth testing. They do not prove the entire chain. The trainees’ task forms, the timetable, individual experience and the reasons for the changed work mix would help distinguish it from alternatives.

A useful mapping ends with questions that can be answered, not with a coloured dashboard whose numbers came from intuition. We can ask which exceptions were deferred, what staff did during the freed hours, and whether a fresh practice design improves independent performance. That is a stronger output than announcing that Larch is in an invented civilisational phase.

Back to contents · Next: 11. Worked answer: calculate a buffer in compatible units

11. Worked answer: calculate a buffer in compatible units

The reserve is measured in usable briefs. The drawdown is measured in usable briefs per working day. Under the reference arrangement, demand exceeds fallback production by 100 minus 70, or 30 briefs per day. Dividing 180 briefs by 30 briefs per day gives six working days of reserve coverage under the simplified constant-flow assumptions.

Under the accelerated arrangement, the gap is 100 minus 40, or 60 briefs per day. Dividing the same 180 briefs by 60 briefs per day gives three working days. The reserve quantity has not changed, but the time it buys has halved because the fallback gap has doubled. This is why a stock quantity and a time buffer should not be used as synonyms.

The four-day restoration path fits within the reference buffer: six minus four equals two working days of modelled margin. Under the accelerated arrangement, three minus four equals a negative one-working-day margin. With restoration available at the start of day five, the accelerated team exhausts the reserve at the end of day three and faces a shortfall on day four.

The day-four shortfall is 60 briefs under these assumptions. Fallback still supplies 40 of the required 100. It would be wrong to say that the whole service has ceased or that 100 briefs are missing. It would also be wrong to spread the missing 60 across the earlier days and claim every learner received slightly less service. The timing of the unmet commitment matters.

The arithmetic is meaningful because it compares compatible quantities. It is not a formula for civilisation. The reserve must be usable for this demand, the fallback rate must apply under the interruption, and the restoration estimate must include all necessary stages. An earlier deadline, a demand surge or a loss of the people performing fallback could make this model inadequate.

The result is nevertheless instructive. The accelerated arrangement is better on the recorded aggregate throughput measure and worse on this interruption test. Neither observation cancels the other. The institution now has a concrete redesign question: can it preserve useful assistance while restoring enough continuity margin and human capability?

Back to contents · Next: 12. Worked answer: separate repair flow from unresolved stock

12. Worked answer: separate repair flow from unresolved stock

The correction ledger answers a different question from the interruption reserve. At the beginning of the reference period, 40 defective briefs are unresolved. Adding 60 new defects and subtracting 72 verified corrections leaves 28. The ledger has improved by twelve briefs during that period. The count says nothing about which individual learner experienced the longest delay, but it establishes the change in unresolved stock.

The accelerated period begins with those 28 unresolved briefs. Adding 75 new defects and subtracting 60 verified corrections leaves 43. The unresolved stock has grown by fifteen. Even though the aggregate defect proportion remained at 2.5%, the larger output produced more counted defects and the ledger recorded fewer verified corrections. A stable percentage did not produce a stable backlog.

Because both periods contain four weeks, the reference inflow is 15 defective briefs per week and the verified correction flow is 18 per week. The accelerated inflow is 18.75 per week and the verified correction flow is 15 per week. Subtracting these flows is legitimate within this ledger because the quantities refer to the same counted unit and the same time interval.

It is tempting to say that this proves the broad slogan “repair must exceed drift.” What it actually proves is a bookkeeping identity under the packet’s assumptions: an unresolved stock rises when additions exceed removals. That identity does not validate a universal measure of social repair or a collapse threshold. The name assigned to the ledger does not extend its explanatory reach.

Even this narrow ledger needs interpretation. One severe error may matter more than ten minor errors. Closing old cases quickly might leave a few difficult cases waiting for a long time. A low count may conceal defects that were never detected. The packet rules eliminate some complications to teach the arithmetic, but a real institution would need measures of severity, age, recurrence and detection quality as well.

The practical conclusion is that Larch needs to investigate correction capacity alongside output. It should not declare the arrangement successful from throughput alone, or unsuccessful merely because a single count grew. The next question is whether the changed allocation created an avoidable repair bottleneck and what a feasible correction would cost.

Back to contents · Next: 13. A reassuring average can hide the weaker service

13. A reassuring average can hide the weaker service

An analyst writes: “The defect rate remained at 2.5% while output rose by 25%. The new arrangement therefore preserved quality.” The first two numerical statements are correct. The conclusion is too broad. An aggregate proportion can remain unchanged because the mixture of work changes, even when one important category deteriorates.

In the reference period, routine defects are 18 divided by 1,800, or 1%. Exception defects are 42 divided by 600, or 7%. In the accelerated period, routine defects are 27 divided by 2,700, still 1%. Exception defects are 48 divided by 300, or 16%. The exception proportion has risen by nine percentage points, even though the combined proportion has remained at 2.5%.

The work mix explains how this can happen arithmetically. Routine briefs account for 75% of reference output and 90% of accelerated output. The accelerated arrangement does more of the lower-defect category and less of the higher-defect category. That shift makes the aggregate look stable while obscuring the worsening result in the smaller exception category.

For illustration, apply the accelerated category proportions to the reference mix. Seventy-five per cent weighted at 1% plus twenty-five per cent weighted at 16% gives 4.75%. This is a standardised comparison using a chosen common mix, not the observed accelerated overall proportion. It helps reveal the composition effect. It does not prove what would actually happen if the team processed that alternative workload.

The corrected statement is more useful: “Aggregate initially defective output remained at 2.5%, but the exception category worsened from 7% to 16% while its share of output fell. We should investigate exception selection, difficulty, supervision and checking before claiming preserved quality.” This sentence keeps the genuine productivity result and names the evidence that weakens the original conclusion.

The framework’s contribution is to keep the exception pathway visible when the production headline looks attractive. It is not needed to calculate the percentages. Basic statistical reasoning supplies that correction. If Civilisation OS encourages an analyst to ignore these ordinary methods in favour of an overall phase score, the lens is making the work worse rather than better.

Back to contents · Next: 14. Repair an invalid comparison of scores and rates

14. Repair an invalid comparison of scores and rates

Now consider a second flawed report: “Larch’s repair capability is eight out of ten, while drift is six defects per week. Eight is greater than six, so the institution is stable.” The numbers do not describe the same kind of quantity. One is an undefined rating; the other is a counted flow. Their numerical ordering cannot support the claimed conclusion.

Changing the labels does not fix the problem. A subjective score called a repair rate is still a subjective score unless its units, measurement procedure and time basis are defined. A difference between two zero-to-ten ratings may be a discussion aid, but it should not be presented as physical distance from failure. Arithmetic can be mechanically possible and conceptually meaningless at the same time.

The repair begins by asking what the report is trying to establish. If the question concerns unresolved defective briefs, use the ledger’s additions and verified removals per week. If it concerns uninterrupted delivery, compare usable reserve coverage with the duration of a specific restoration scenario. If it concerns human learning, examine task performance and the conditions of practice. These questions need separate answers.

For Larch, the valid ledger comparison is 18.75 newly identified defective briefs per week against 15 verified corrections per week in the accelerated period. The unresolved stock therefore grows by 3.75 briefs per week on average across the four-week record. The valid interruption comparison is three working days of reserve coverage against four working days to restoration. Neither result requires an invented overall score.

There is room for qualitative judgement. The records do not answer whether the exception mistakes are unacceptable or how much independent capability the centre should preserve. Those decisions require explanation, consultation and domain knowledge. Calling them qualitative does not make them arbitrary. It makes clear where reasons and priorities, rather than common physical units, do the work.

A repaired report should also preserve its own uncertainty. “This arrangement fails the stipulated four-day continuity test” is defensible. “Larch will collapse soon” is not. An institution can miss one service commitment and still have many available responses. A dramatic label can conceal those responses and make a recoverable problem appear inevitable.

Back to contents · Next: 15. Propose a feasible repair without claiming it already worked

15. Propose a feasible repair without claiming it already worked

Larch has a useful option between abandoning assistance and accepting the current arrangement unchanged. It can use four of its eight discretionary staff-hours to prepare and check 120 additional case-matched reserve briefs. The usable reserve would then rise from 180 to 300. This is a planned change with a supplied production requirement, not an observation that the reserve has already been built.

If the reserve is completed and verified, the accelerated fallback gap remains 60 briefs per day. Three hundred divided by sixty gives five working days of coverage. Against the same four-day restoration scenario, that creates one working day of modelled margin. The immediate service shortfall in the original scenario is removed without claiming that the staff’s independent capability has improved.

The remaining four hours can be allocated to supervised fallback practice. That decision addresses the capability pathway rather than merely accumulating completed work. Tutors could attempt exception briefs with the assistant unavailable, compare their reasoning with a checked reference, and practise the handover procedure. Whether this improves performance is an outcome to examine, not a number to insert into the plan in advance.

A common planning error would be to count a hoped-for fallback rate of 60 briefs per day immediately. With a 300-brief reserve, that assumed rate would yield 7.5 working days of coverage. But the packet only supports the measured rate of 40 under the accelerated arrangement. The larger figure belongs in a conditional scenario until a suitable exercise demonstrates it.

The eight-hour allocation also has an opportunity cost. Those hours cannot simultaneously be spent producing additional ordinary briefs, clearing the correction backlog and mentoring trainees. The plan should identify which activities are deferred and whether the centre’s priorities justify that trade-off. A recommendation that assigns the same hour to three different repairs is infeasible, however humane its language sounds.

This modest repair is not a complete approval of the accelerated arrangement. The exception-category result and growing correction stock still require investigation. It does, however, demonstrate the intended use of the lens: connect a specific weakness to a feasible response, state what the response can change, and retain the questions it leaves unresolved.

Back to contents · Next: 16. Write the decision so another person can challenge it

16. Write the decision so another person can challenge it

A defensible Larch recommendation might read as follows. “Retain a bounded use of the drafting assistant for the work where the recorded benefits are useful, while addressing the exception pathway and continuity gap before expanding dependence. Build and verify the additional reserve using four available hours. Use the other four hours for supervised fallback practice. Do not credit improved fallback performance until it has been demonstrated.”

The explanation should include the trade-off. The accelerated period recorded a 25% increase in initially usable briefs per counted staff-hour. Its aggregate defect proportion concealed a weaker exception result. The correction stock grew from 28 to 43. Its current reserve covers three working days against a four-day restoration scenario. These are distinct findings, not ingredients for an average institutional score.

The value judgement should be equally visible. This recommendation gives weight to continuity, adequate exception feedback and continued human learning, rather than maximising the current output count alone. A stakeholder who prioritises an urgent temporary backlog might accept a different short-term allocation. That disagreement should be discussed as a disagreement about consequences and priorities, not hidden behind claims of mathematical inevitability.

The recommendation also needs a reversal condition. If reserve preparation proves slower than stated, the eight-hour plan must be revised. If a repeat exercise finds lower fallback capacity, the continuity calculation must be recomputed. If the exception result is traced to a change in case difficulty rather than the work arrangement, the response should target the actual cause. The plan should remain answerable to evidence.

A different cautious decision could be defensible: temporarily return exception work to a more supervised arrangement while retaining assistance for routine drafting. The packet does not give enough information to calculate the effect of that split on total staff-hours, so it should be proposed for evaluation rather than described as a guaranteed solution. Being specific about missing information is more useful than pretending one design dominates every other design.

The final decision belongs to the responsible institution and the people affected by its priorities. A conceptual framework helps make the decision legible. It does not acquire the authority to make educational, employment or public-service commitments simply because it has organised the evidence into named layers.

Back to contents · Next: 17. What the Larch observations still cannot establish

17. What the Larch observations still cannot establish

The trainee result is eight successful demonstrations before and six afterwards among the same twelve people. Without the individual paired records, we cannot say that exactly two trainees deteriorated while everyone else stayed the same. Several people could have changed in each direction. The group totals describe a net difference, not each person’s learning trajectory.

Even with paired records, several explanations would remain. The task forms might not be equally difficult. Some trainees might have missed practice. The revised timetable might have changed sleep, preparation or mentor access. The assistant could have supported some activities and displaced others. A causal study would need a design capable of distinguishing the effect of interest from these alternatives.

The staff drill requires similar care. Two current successful demonstrations do not prove that only two staff members possess any relevant knowledge. A person might understand the process but need a prompt to locate the correct offline record. Another might complete the task accurately but too slowly for the stipulated service rate. These are different problems and call for different support.

The correction ledger also omits the experience of those waiting. Forty-three unresolved briefs could include many recent minor matters or a smaller set of long-standing serious ones, depending on the counting scheme. The packet deliberately compresses this complexity. A responsible reader should ask for age and severity distributions before treating the total as an adequate account of harm.

The buffer model assumes steady demand and perfectly usable reserve. A real centre might receive demand in bursts, encounter mismatched tasks, or lose access to a room at the same time as the assistant. It might also be able to postpone a non-urgent activity with learners’ agreement, reducing demand without harming the essential service. Both adverse and favourable departures from the model matter.

These limitations are not reasons to abandon analysis. They are the next questions generated by analysis. The point is to stop at the right claim: the supplied arrangement has a demonstrated weakness under the stipulated scenario, and several plausible explanations deserve examination. No record here establishes inevitable institutional decline, a universal threshold, or the future of civilisation.

Back to contents · Next: 18. Try a second institution before reading the answer

18. Try a second institution before reading the answer

Merehaven Community Archive is a second fictional institution. It prepares catalogue records so visitors can find material in a general collection and a local-language collection. Eight staff members share the work, and six adult trainees learn catalogue checking. A generative assistant is introduced for draft descriptions, while staff continue to decide how records are classified and accepted.

Across a reference four-week period, the archive produces 1,000 records using 200 counted staff-hours; 40 records contain a counted defect. In a later assisted four-week period, it produces 1,500 records using 200 counted staff-hours; 30 contain a counted defect. The same fictional acceptance rubric is used. The record does not supply a task-category breakdown for these production periods, so you cannot rule out a work-mix effect.

On equivalent-form independent checking tasks, four of the six trainees meet the criterion before and five afterwards. Notes are allowed; generative assistance is unavailable during the check. The packet again supplies no individual paired records or causal comparison. The improvement is an observation that deserves attention, not proof that every aspect of learning improved because of AI.

During the specified interruption, daily demand is 50 general-collection records and 20 local-language records. Offline fallback can supply 45 general records and five local-language records per working day. The archive has 60 usable, case-matched general reserve records and 40 usable, case-matched local-language reserve records. Records cannot substitute across the two categories.

The restoration scenario lasts three working days from the start of the interruption, with normal service available at the start of day four. Demand and output flow evenly. A recent drill verified one additional arrangement: a bilingual cataloguer can shift ten records per day of their output from the general collection to the local-language collection without changing total staff-hours or the acceptance standard. The archive authorises this reallocation, and the bilingual cataloguer is available throughout the three-day interruption, so the shift can begin immediately in the scenario.

Before continuing, answer four questions. What has improved in the production record? Does a pooled reserve calculation protect both collections? Which permitted staffing arrangement gives better continuity? What different recommendation might still be defensible, and what evidence would distinguish it? Use the supplied facts rather than importing Larch’s conclusion into the new case.

Back to contents · Next: 19. Transfer answer: recognise a different pattern of gains

19. Transfer answer: recognise a different pattern of gains

Merehaven’s reference throughput is 1,000 records divided by 200 staff-hours, or five records per staff-hour. Its assisted throughput is 1,500 divided by 200, or 7.5 records per staff-hour. The recorded increase is 50%. The initially defective proportion declines from 40 divided by 1,000, or 4%, to 30 divided by 1,500, or 2%.

Initially usable records increase from 960 to 1,470. Per counted staff-hour, that is 4.8 versus 7.35. The increase in this measure is 2.55 divided by 4.8, or 53.125%. This differs from the 50% increase in total output because the counted defect proportion also improves. A careful answer states which measure it uses rather than alternating between percentages as though they were identical.

The trainee count also moves in a favourable direction, from four successful demonstrations to five among six trainees. That matters. The proposed lens should not force a story of capability erosion whenever AI appears. Merehaven has observations consistent with a more useful arrangement, even though the packet cannot isolate the cause or establish long-term durability.

There are still unanswered questions. The production periods may contain different kinds of work. The defect count may hide severity differences. A short independent task may not represent all the judgement required in catalogue work. We have no correction-stock ledger comparable to Larch’s, so we cannot claim that Merehaven’s backlog of unresolved defects fell merely because fewer new defects were counted.

The correct transfer therefore carries the method, not the verdict. Recognise the favourable output, error and independent-performance observations. Preserve the uncertainty about their causes and scope. Then examine the specific continuity dependencies. A framework that yields “AI makes every institution fragile” regardless of the supplied evidence would fail this exercise.

This case also shows why human capability and assisted productivity should not be treated as necessarily competing quantities. A work arrangement might improve both. The packet does not explain how that happened, but it leaves room for hypotheses such as better examples, more effective feedback or freed mentor time. Those hypotheses could guide a future inquiry without being promoted into findings.

Back to contents · Next: 20. Transfer answer: the pooled buffer is misleading

20. Transfer answer: the pooled buffer is misleading

An analyst pools Merehaven’s reserves and fallback output. Total demand is 70 records per day, total fallback output is 50, and the reserve contains 100 records. The calculation gives 100 divided by 20, or five working days. Against three days to restoration, that appears to leave two days of margin.

The arithmetic is internally correct for a pool of interchangeable records. But the packet explicitly says the categories are not interchangeable. A general-collection record cannot satisfy a local-language catalogue request. Treating them as one pool assumes a substitution that the case rules do not allow. Compatible units are necessary for a valid calculation, but they are not sufficient when the items serve different functions.

Calculate the lanes separately. General demand exceeds fallback by 50 minus 45, or five records per day. Its reserve of 60 lasts twelve days. Local-language demand exceeds fallback by 20 minus five, or fifteen records per day. Its reserve of 40 lasts forty divided by fifteen, or about 2.67 working days.

The archive promises service in both collections. Under the initial staffing arrangement, the local-language lane runs short before the three-day restoration finishes. Across three days it needs 60 records, produces fifteen through fallback and begins with forty in reserve. That leaves five records unmet. The general lane still has spare reserve, but cannot use it to erase the local-language shortfall.

This is a different error from Larch’s misleading aggregate defect proportion, but the discipline is related. Preserve the distinctions that matter to the service. A plentiful resource in one lane does not automatically compensate for scarcity in another. An average can conceal a specific group’s unmet need even when the total looks comfortable.

The CivOS mapping can draw attention to the interaction between role capability, allocation rules and service output. The calculation itself remains a small, conditional inventory analysis. We should describe the five-record shortfall precisely rather than calling it collapse, and investigate the permitted reallocation before concluding that additional staff or a rejection of assistance is necessary.

Back to contents · Next: 21. Transfer answer: use the capability that is actually available

21. Transfer answer: use the capability that is actually available

The packet provides a verified reallocation: one bilingual cataloguer can shift ten records per day from general work to local-language work. After the shift, general fallback falls from 45 to 35 records per day, while local-language fallback rises from five to fifteen. Total fallback remains 50. No extra person, hour or machine capability has been invented.

The general lane now draws fifteen reserve records per day, because demand is 50 and fallback is 35. Its 60-record reserve lasts four working days. The local-language lane draws five per day, because demand is twenty and fallback is fifteen. Its forty-record reserve lasts eight working days. The combined service can therefore meet both requirements for four days under these assumptions.

Against the three-day restoration scenario, the binding lane has one day of margin. At the end of day three, the general reserve has fifteen records left and the local-language reserve has twenty-five. All stipulated demand has been met. This is a useful improvement even though total fallback output and total reserve have not changed.

The finding illustrates the difference between possessing capability and routing it where it is needed. Merehaven already had a person with a relevant demonstrated skill. The initial allocation concentrated too much fallback production in the better-buffered lane. A change in coordination makes the existing capacity more useful. There is no need to frame the problem as a failure of everyone’s learning or as proof that the assistant is harmful.

The reallocation still has boundaries. It depends on the bilingual cataloguer being available, authorised to make the shift, and able to maintain the acceptance standard. The packet supplies those conditions for this scenario. In a real institution they would require checking. A second person with the same capability could reduce dependence on one individual, but the case does not give the time or cost required to create that redundancy.

A reasonable recommendation is to continue the bounded assisted arrangement while documenting and practising the verified interruption allocation, and to gather better evidence about the longer-term learning and quality results. This recommendation differs from Larch’s because the evidence and available repair differ. That difference is what makes transfer an actual test of reasoning.

Back to contents · Next: 22. Two defensible readings, and the evidence that separates them

22. Two defensible readings, and the evidence that separates them

One reader may favour continued use at Merehaven. The recorded output and defect measures improved; the trainee result improved; and an available reallocation satisfies the specified three-day interruption. If the archive’s priority is to improve access while maintaining these service commitments, a bounded continuation with explicit conditions is defensible.

Another reader may recommend withholding approval for continued reliance on the assisted arrangement until a second interruption exercise and a named backup for the bilingual role are available. The restoration duration comes from a limited exercise, the present bilingual fallback depends on one person, and the learning evidence is small. This reader need not deny the observed gains: they give uncertainty greater weight before accepting continued dependence. The packet does not quantify the cost or service effect of that pause, so its feasibility must be checked rather than assumed.

The disagreement can be made productive by naming the evidence that would distinguish the positions. A restoration exercise under a different failure condition could test whether three days is a reasonable planning assumption. A drill with the bilingual cataloguer unavailable could show what happens to the local-language lane. A comparable production sample could separate work-mix changes from changes in the quality of the arrangement.

Repeated independent tasks with appropriate variation could reveal whether trainee capability persists and transfers. Records of who performs the checking and how much supervision is required could show whether the new output relies on an uncounted expert bottleneck. Consultation with users of both collections could clarify the consequences of delay and whether the stated service priorities match their needs.

Neither reading is entitled to invent a probability of failure from the packet. We do not know the frequency of interruptions, the distribution of restoration times, or the likelihood of the specialist’s absence. A scenario margin is not an annual reliability estimate. Decisions can still be made under uncertainty, but the uncertainty must remain visible rather than being replaced by a confident percentage.

This is an important standard for the proposed lens. It should allow reasonable people to disagree while exposing exactly where their evidence, assumptions or priorities differ. If every disagreement is dismissed as failure to understand the framework, the framework has stopped being a tool for inquiry and become a way of protecting a preferred story.

Back to contents · Next: 23. A local exercise cannot establish a law of civilisation

23. A local exercise cannot establish a law of civilisation

Larch and Merehaven are deliberately small. Their boundaries, categories and quantities are supplied in one packet. A society is not. It contains many institutions, contested purposes, informal care, political conflict, changing resources and people whose contributions cannot be reduced to a catalogue of tasks. Moving from a local service calculation to a claim about civilisation requires additional theory and evidence.

The fact that a reserve lasts three days under a constant-flow model does not imply that an entire society has three days of survivability. Institutions adapt, substitute, cooperate and sometimes fail for reasons outside the chosen boundary. A locally useful simplification can become misleading when transferred to a scale where its assumptions no longer hold.

Nor does the presence of feedback loops establish that all important relationships behave like an engineered control system. People interpret rules, contest goals and change their behaviour in response to being measured. A metric can affect the thing it measures. A target for fast correction might encourage staff to close easy cases and neglect difficult ones. A framework must leave room for those responses rather than treating people as passive components.

The language of regeneration also needs ethical restraint. People have lives and purposes beyond keeping an institution supplied with labour. Training, care, rest and participation matter in their own right, not merely because they increase a system’s replacement capacity. A human-centred use of the lens makes these priorities explicit and does not treat anyone as an interchangeable spare part.

Claims of predictive accuracy would need a much more demanding test than a convincing retrospective story. Analysts would have to define the outcome, specify the prediction before seeing it, document inputs, compare against reasonable alternatives, and examine errors across cases. Agreement after a crisis does not establish that a model could reliably have anticipated it.

This article provides no such validation of Civilisation OS. It demonstrates a way of asking connected questions and a few conditional calculations. That is a narrower contribution, but it can still be useful. The right standard is whether the lens improves attention and reasoning without encouraging unsupported certainty.

Back to contents · Next: 24. A hypothetical ASI changes the scenario, not the evidence rules

24. A hypothetical ASI changes the scenario, not the evidence rules

Suppose, as a thought experiment, a future system could explain complex exceptions more accurately than today’s best human team, train novices effectively, and identify faults that humans would miss. Such a system could improve several parts of the proposed loop at once. It would be unreasonable to assume that greater AI capability can only weaken human institutions.

The next question would concern the surrounding arrangement. Who defines the service goal? Which records establish that the system performs as claimed? What happens when a person disputes an outcome? Which infrastructure does the service depend on? Can the institution recover from a failure whose cause lies outside the model’s reasoning ability? These questions remain meaningful even in a highly capable-machine scenario.

Human fallback may also need a more nuanced role. It is not credible to promise that people can manually reproduce every future system’s speed and breadth. A fallback may instead preserve a smaller essential service, pause non-urgent work, route difficult cases to another institution, or maintain the ability to make legitimate decisions while technical recovery proceeds. The requirement should be defined honestly rather than defended as a slogan.

Human learning could remain important for participation, understanding, accountability and the ability to choose what the institution should do. Those are normative reasons, not predictions that humans will always be technically superior. Different communities may weigh them differently. A model of capability should not silently decide that people no longer need a voice because another system can generate a more sophisticated answer.

Conversely, preserving every historical task without examining its purpose could waste opportunities. If a tool reliably removes a tedious activity, education might redirect effort toward explanation, judgement, relationships or a different foundational skill. The question is which capabilities people should retain and why, not whether every old workflow must remain unchanged.

All these statements are conditional. They do not establish that ASI exists, specify when it will arrive, or show that any current institution can rely on it. Larch’s and Merehaven’s decisions should use the capabilities and evidence actually available to them. A speculative future can broaden planning questions; it should not be booked as present repair capacity.

Back to contents · Next: 25. Use a compact reasoning record instead of a grand score

25. Use a compact reasoning record instead of a grand score

A reader can apply this exercise with a short written record. Start with the function: what outcome must continue, for whom, and over what period? Then state the boundary: which people, tools, suppliers, tasks and resources are included? If an important dependency lies outside the boundary, name it rather than allowing it to disappear.

Next, describe the capability pathway. Identify who can perform the essential task, what evidence supports that judgement, who can check it, and how someone else learns it. Distinguish a current demonstration from a job title, a completed training session or an assumption that a helpful colleague will be available. Include the time needed to supervise and practise.

Record the output and defect measures with their denominators. Keep categories separate when their purposes, difficulty or consequences differ. Describe unresolved stock separately from incoming defects and verified corrections. If a measure is a subjective rating, label it as such and explain what it helps people discuss; do not disguise it as a physical unit.

For the interruption scenario, specify the unavailable function, usable reserve, fallback output, demand and restoration path. Calculate a time margin only when the quantities genuinely correspond. Test whether categories are interchangeable and whether a person or resource has been counted twice. State which assumption is most likely to change the result.

Finally, write the recommendation and its priority. Explain what benefit is being preserved, what weakness is being repaired, what the repair costs, and what evidence would cause a revision. Keep hypothetical improvements outside the observed record until they are demonstrated. This is enough to produce an accountable first analysis without building software, inventing dashboards or claiming to run a universal diagnostic engine.

The record should be understandable without the framework’s terminology. A colleague should be able to say, “We have three days of coverage and a four-day restoration scenario,” even if they have never heard of a lattice buffer. If translating the analysis into ordinary language makes it collapse, the conceptual language was doing too much work.

Back to contents · Next: 26. The useful question is what remains possible after the gain

26. The useful question is what remains possible after the gain

The strongest lesson from these cases is not a verdict for or against AI. Larch demonstrates that more output can coexist with weaker exception handling, a growing correction stock and a thinner interruption buffer. Merehaven demonstrates that favourable output and learning observations can coexist with a hidden allocation problem that a verified human capability can repair.

The same broad lens helps only if it permits those different answers. It directs attention to learning, coordination, production, constraints and correction, but the case evidence decides which relationships matter. A name does not establish causation. A score does not establish a threshold. A correct inventory calculation does not establish a theory of history.

For an educator, the practical question is whether assistance helps learners become more capable and whether the institution can still recognise and repair mistakes. For a manager, it is whether useful output gains are accompanied by a workable handover and recovery pathway. For a student, it is whether they can explain which parts of the conclusion came from observation, calculation, values and speculation.

These are demanding questions, but they do not require alarm. A weakness can reveal a repairable design choice. More reserve, a different allocation, a better practice task or clearer checking responsibility may change the outcome. Sometimes the evidence will favour retaining assistance; sometimes it will favour narrowing its role; sometimes the packet will not yet support a confident choice.

eduKateSG’s Civilisation OS is most responsibly used here as a proposed lens for making those relationships easier to examine. Its claims should remain open to correction, and its vocabulary should remain subordinate to the people and functions being discussed. The test is whether the next decision becomes more accurate, humane and revisable.

An intelligent institution should be able to benefit from a powerful tool and still ask how the benefit is being produced, who can carry it forward, and what happens when conditions change. That is the bridge between Super Intelligence and this proposed Civilisation OS lens: an answer matters, but so does the continuing capacity to understand, choose, teach and repair.

Back to contents · Next: Sources and further reading

Sources and further reading

Framework provenance: Civilisation OS supplies eduKateSG’s proposed loop, HRL and lattice-buffer vocabulary. What is Civilisation OS? introduces the framework and points to the canonical page. These are the framework’s own definitions, not independent validation of its predictive claims.

Series boundary: the Super Intelligence master guide distinguishes eduKateSG operating models from standard scientific definitions and hypothetical AGI/ASI scenarios from demonstrated present capabilities. It also provides the broader discussion of resilience, human fallback and repair from which this applied exercise proceeds.

Primary learning research: Hamsa Bastani and colleagues, Generative AI without guardrails can harm learning: Evidence from high school mathematics, PNAS, 2025. The article is relevant to the distinction between assisted performance and independent learning; its context and design limit generalisation. Consult the publication record and its updates when using specific findings.

Primary recovery guidance: Marianne Swanson and colleagues, NIST SP 800-34 Revision 1, Contingency Planning Guide for Federal Information Systems, including the November 2010 update. This supports a bounded discussion of recovery priorities and exercises within its information-system scope, not validation of Civilisation OS.

Case status: Larch Civic Learning Centre and Merehaven Community Archive, their participants, records, quantities and decisions are original fictional teaching materials in this article. Their calculations demonstrate conditional reasoning, not empirical institutional results or predictions about real learners, workers or communities.

Back to contents · Return to the Super Intelligence master guide