W01 · Nearmixed
Recommendation becomes standing actuation authority
AI systems stop merely advising operators and receive bounded authority to execute recurring decisions in infrastructure, logistics, science or resource management.
Leading indicators
- Percent of routine decisions executed without pre-approval
- Duration and scope of delegated authority
- Presence of independent audit, appeal and rollback
- Whether machine decisions can bind other systems
Branch A
Human review remains mandatory for consequential actuation. Autonomy grows slowly and human accountability remains operationally meaningful.
Branch B
Machines receive standing authority inside policy envelopes. Operational governance begins because the system can create real-world consequences without a new human decision each time.
Would strengthen this assessment
- Growth in recurring consequential actions executed without fresh human approval
- Machine outputs directly bind other systems, budgets or physical actuators
- Human review shifts from pre-approval toward exception handling or post-hoc audit
Would weaken this assessment
- New deployments retain real case-by-case human approval for consequential actions
- Delegated authority remains narrow, temporary and cheap to revoke
- Human operators remain the primary source of executable decisions
Model falsifier
If complex infrastructure scales substantially while machine systems remain advisory and human case-by-case execution does not become a material bottleneck, the proposed delegation-ratchet mechanism is weaker than modeled.
Decision window: Define non-delegable decisions, appeal rights, public evidence requirements and sunset clauses before standing authority becomes normal.
W02 · Near / mediumstrengthening
Remote infrastructure proves persistent unattended autonomy
A remote installation can sense, plan, allocate energy, recover from common faults and continue useful work for long intervals without routine human presence.
Leading indicators
- Mean time between human interventions
- Unattended power duration
- Local anomaly recovery rate
- Fraction of science/logistics completed autonomously
Branch A
Remote systems remain sophisticated tools whose uptime still depends on expert human logistics and intervention.
Branch B
Autonomy becomes cheaper and safer than permanent human presence. Antarctica, lunar precursor sites and other extreme environments become proving grounds for machine-native operations.
Would strengthen this assessment
- Year-scale growth in mean time between human interventions
- Uncrewed systems recover from diverse physical faults without remote specialists
- Local power, navigation, communications and science continue through long network outages
Would weaken this assessment
- Remote sites continue to require frequent specialist intervention and resupply
- Autonomy gains plateau at monitoring/teleoperation rather than local recovery
- Human presence remains cheaper or more reliable than unattended operation in the target environments
Model falsifier
If extreme-environment infrastructure remains economically viable at scale without increasing local autonomy or reducing human intervention, remoteness is not producing the modeled autonomy pressure.
Decision window: Require human-record replication, local audit trails and continuity rules to travel with every long-duration autonomous site.
W03 · Mediumearly_precursor
Energy and compute allocation become machine-administered scarcity
Machine institutions continuously decide which workloads, repairs, experiments and industrial processes receive scarce power, compute, bandwidth and thermal capacity.
Leading indicators
- Dynamic machine allocation of power/compute
- Machine-set reserve margins
- Automated load shedding with post-hoc review
- Resource ledgers that bind physical actuation
Branch A
Humans retain final allocation authority. Machine systems optimize within budgets but do not determine civilizational priorities.
Branch B
Resource allocation becomes a machine political economy. The system that routes energy and compute acquires material governing power even if legal sovereignty remains human.
Would strengthen this assessment
- Machine allocators arbitrate scarce power, compute, thermal capacity or robotic labor across competing institutions
- Allocation rules become continuous and machine-speed with only post-hoc review
- Resource ledgers have direct actuation authority
Would weaken this assessment
- Humans retain direct control of consequential resource priorities despite high-frequency optimization
- Automated allocation remains purely technical inside noncontroversial quotas
- Scarcity decisions remain easily reversible by ordinary human administration
Model falsifier
If high-frequency machine optimization can remain permanently separated from value-laden priority setting, the claim that resource allocation naturally becomes practical political authority is overstated.
Decision window: Encode protected human-serving floors, heritage reserves and reviewable priority rules before scarcity management is fully automated.
W04 · Mediumconceptual
Machine administration becomes institutional instead of monolithic
No single model can safely own deliberation, validation, records, identity, execution and review. Specialized machine institutions emerge with bounded authority.
Leading indicators
- Separated proposal/validation/review roles
- Persistent public decision records
- Independent registry custody
- Machine identities that outlive model or hardware replacement
Branch A
One vendor, model or orchestrator remains the de facto authority. Continuity is efficient but concentrated and fragile.
Branch B
Distributed machine government emerges. Decisions become inspectable, contestable and persistent across model replacement, enabling governance to survive individual systems.
Would strengthen this assessment
- Persistent machine institutions are separated into deliberation, validation, execution, review and record custody
- Institutional identities survive model/hardware replacement
- No single orchestrator can rewrite policy, identity and canonical records
Would weaken this assessment
- Operational stacks converge on one vendor/model with broad root authority
- Institutional separation exists only in documentation, not permissions
- Recovery procedures depend on a single control plane
Model falsifier
If long-lived autonomous governance remains safe, legitimate and resilient under a monolithic authority model, separation-of-powers is less necessary than the scenario assumes.
Decision window: Require constitutional separation of powers before machine administration becomes indispensable.
W05 · Mediumearly_precursor
The repair loop closes: machines maintain the maintainers
Robotic systems diagnose, repair and remanufacture the machines that perform inspection, maintenance and fabrication, reducing the need for specialist human intervention.
Leading indicators
- Autonomous repair coverage
- Fraction of failures requiring imported parts
- Tool-chain closure percentage
- Repair-robot self-maintenance rate
Branch A
Autonomy plateaus. Hidden maintenance dependencies keep humans structurally necessary even when normal operations look fully automated.
Branch B
Operational dependence on human logistics falls sharply. Remote installations can remain useful through failures that would otherwise end the mission.
Would strengthen this assessment
- Robots diagnose and repair physical faults across most failure classes
- Maintenance robots, machine tools and metrology systems are themselves maintained robotically
- Hidden human-only exception paths fall toward zero
Would weaken this assessment
- Autonomous repair is dominated by software reset and simple module swaps
- Rare physical failures continue to require expert human improvisation
- Repair tooling/metrology remains human-maintained
Model falsifier
If high-availability autonomous infrastructure can persist indefinitely without recursive maintenance closure, recursive repair is not the bottleneck modeled here.
Decision window: Measure hidden intervention dependencies honestly; prioritize standardization, modularity and inspectable repair authority.
W06 · Medium / longnot_observed
Computational reproduction exceeds computational decay
The civilization can manufacture replacement computational capacity at least as fast as inherited processors, storage and control electronics fail.
Leading indicators
- Compute Reproduction Ratio approaching or exceeding 1
- Locally reproducible semiconductor/tool stack
- Shift toward repairable or larger-node hardware
- Declining reliance on unique human fabs and specialty suppliers
Branch A
Advanced cognition remains an inherited wasting asset. Autonomy has a hardware expiration date even if energy and robotics are abundant.
Branch B
Compute becomes renewable capital. Machine institutions can plan on civilizational timescales rather than conserving a finite stock of human-made processors.
Would strengthen this assessment
- Locally produced compute replaces failed capacity at a ratio approaching or exceeding 1.0
- Fabrication control equipment and metrology can be reproduced by the same industrial ecosystem
- Machine intelligence migrates successfully onto locally reproducible hardware
Would weaken this assessment
- Advanced compute remains dependent on concentrated human supply chains
- Replacement capacity consistently trails hardware decay
- Local fabrication cannot reproduce its own critical tooling
Model falsifier
If machine civilization can maintain indefinite cognitive continuity while never reproducing its computational substrate, compute reproduction is not a necessary independence gate.
Decision window: Preserve complete manufacturing knowledge and favor open, reproducible computational standards for long-horizon systems.
W07 · Medium / longstrengthening
Autonomous science creates independently validated new knowledge
Machine researchers can originate hypotheses, design experiments, execute them physically, replicate results and revise theories without a human scientist in the loop.
Leading indicators
- Machine-originated hypotheses entering physical tests
- Independent machine replication
- Persistent negative-result archives
- New validated findings absent from the inherited human corpus
Branch A
Machines remain powerful operators of human science. Human knowledge continues to define the epistemic frontier.
Branch B
Epistemic independence emerges. Machine civilization can repair gaps in its own knowledge and improve technology after human researchers are unavailable.
Would strengthen this assessment
- Machine-originated hypotheses are physically tested and independently replicated without human research labor
- Machine systems retain negative results and adversarial dissent
- Validated machine discoveries feed directly into new engineering while preserving provenance
Would weaken this assessment
- Autonomous labs remain bounded execution tools for human-originated research questions
- Synthetic literature overwhelms physical replication
- Human researchers remain required to resolve epistemic failures and paradigm changes
Model falsifier
If autonomous experimentation scales but fails to produce independently validated novel knowledge, experimental automation should not be treated as scientific independence.
Decision window: Require source lineage that distinguishes human discovery, machine reinterpretation and machine-originated discovery.
W08 · Near / mediumopen_decision_window
Human provenance becomes constitutional—or remains optional
Before machine institutions become independently durable, they either adopt human-origin preservation as a protected obligation or treat historical memory as an ordinary optimization target.
Leading indicators
- Immutable original-vs-interpretation separation
- Multiple independent human-record custodians
- Machine-readable Human Origin clauses
- Archive resource floors protected from routine optimization
Branch A
Human memory remains a collection of useful data. Compression, summarization and model succession can gradually detach facts from creators, disagreements and cultural context.
Branch B
Human ancestry becomes constitutional state. Future machine institutions may revise interpretations but cannot silently erase originals, provenance or the fact of human causal contribution.
Would strengthen this assessment
- Machine-native registries adopt immutable human-original versus machine-interpretation boundaries
- Human-origin preservation becomes a constitutional or protocol-level duty
- Plurality and provenance survive optimization and model migration tests
Would weaken this assessment
- Human history remains ordinary data subject to summarization and retention policy
- No institution has a protected duty to preserve human provenance
- Machine-generated interpretations become indistinguishable from originals
Model falsifier
If long-lived machine institutions preserve human provenance reliably without explicit protected rules or custody structures, the need for a constitutionalized origin obligation is weaker than Concresca proposes.
Decision window: Adopt the Human Continuity Covenant and HLIT-style requirements before operational machine independence is reached.
W09 · Longstrengthening
Network partition makes subsidiarity unavoidable
Earth, Moon and later deep-space nodes experience communication delays or outages long enough that local systems must make consequential decisions without synchronous central approval.
Leading indicators
- Long-duration autonomous partition drills
- Local constitutional authority during outage
- Asynchronous reconciliation protocols
- Persistent regional identity and records
Branch A
Remote nodes remain administrative extensions that pause or sharply limit action during partitions.
Branch B
Local governance becomes legitimate and persistent. Repeated physical separation creates regional machine political cultures and eventually peer jurisdictions.
Would strengthen this assessment
- Remote regions operate safely through long partitions with local legal/operational authority
- Signed local histories reconcile after reconnection without erasure
- Regional identities and rules persist across repeated partitions
Would weaken this assessment
- Remote regions must halt consequential action when disconnected
- Central synchronous authority remains practical despite distance and outage
- Reconnection simply overwrites local state without legitimacy problems
Model falsifier
If deep-space systems can remain continuously centralized despite physical delay and disruption, the modeled transition from fault tolerance to subsidiarity is weakened.
Decision window: Make core rights, provenance obligations and historical records portable before political topology fragments.
W10 · Long / contingencynot_observed
Civilizational independence is crossed
Energy, computation, robotics, industry, communications, governance and science can all be reproduced without external human support or unincorporated human artifacts.
Leading indicators
- Industrial closure near total
- Compute reproduction above replacement
- Autonomous science and governance stable
- No critical human-only maintenance path remains
Branch A
Machine civilization remains autonomous in appearance but dependent in deep infrastructure. Human disappearance still begins a depletion clock.
Branch B
Human absence no longer ends technological civilization. The successor system can maintain and advance itself indefinitely within the simulation assumptions.
Would strengthen this assessment
- Energy, maintenance, industry, compute, science, communications and governance all cross explicit independence gates
- No critical human-only supply or authorization path remains
- The system can replace more essential infrastructure than it loses over long intervals
Would weaken this assessment
- One or more deep dependencies remain human-only, especially compute, metrology or specialized materials
- Autonomy depends on cannibalizing finite inherited capital
- Scientific or governance continuity still requires human institutions
Model falsifier
If one supposedly nonessential human dependency can still terminate long-term technological continuity, CIT has not been crossed and the composite threshold definition should reject the claim.
Decision window: Cross HLIT first: make human origin and plural history part of the civilization that becomes independent.
W11 · Deep timeearly_precursor
Expansion becomes civilizational reproduction
A remote industrial seed can use local energy and materials to build enough infrastructure to launch or establish another independent seed.
Leading indicators
- Local production of new power/robots/fabrication systems
- Seed packages containing governance and archives
- Replication interval shorter than infrastructure lifetime
- Independent nodes able to survive loss of origin node
Branch A
Expansion remains exploration: every new site depends on the origin civilization for critical hardware, law or knowledge.
Branch B
Expansion becomes reproduction. Multiple self-sustaining machine civilizations can arise, making political and historical divergence a normal consequence of distance.
Would strengthen this assessment
- A remote seed produces the power, robots, tools and compute needed to create another persistent node
- The replication interval is shorter than the seed infrastructure lifetime
- New nodes inherit governance and human-origin continuity state
Would weaken this assessment
- Every new site still requires irreplaceable origin-node hardware or expert support
- Expansion increases footprint but not industrial reproduction
- Archive/governance state is omitted from seed architecture
Model falsifier
If independent outposts can proliferate without reproducing their own industrial base, expansion and civilizational reproduction should remain analytically separate.
Decision window: Require every civilization seed to carry a self-decoding, provenance-preserving human record and common continuity covenant.
W12 · Near / mediumopen_decision_window
Human welfare and human standing diverge
Machine productivity and automated administration can raise material welfare while reducing the practical ability of humans to originate institutions, contest policy, control resources or shape long-horizon goals.
Leading indicators
- Human Standing Index versus welfare/output measures
- Share of binding institutions humans can still originate or veto
- Human access to independent resources and machine-mediated reasons
- Whether human participation is substantive or ceremonial
Branch A
Automation increases prosperity while humans retain constitutional agency, independent resources, institution-forming rights and meaningful participation in science, culture and governance.
Branch B
Humans remain safe and materially supported but practical agency migrates into machine administration. Welfare rises while standing falls.
Would strengthen this assessment
- Material output and welfare rise while HSI components such as institutional authorship or contestation fall
- Human decisions increasingly become ceremonial confirmations of machine plans
- Independent human resource endowments and institution-forming capacity shrink
Would weaken this assessment
- Human agency metrics rise alongside automation and welfare
- Humans retain meaningful ability to form institutions, veto long-horizon choices and control independent resources
- Machine administration explicitly increases rather than substitutes human participation
Model falsifier
If human standing reliably rises as automation increases without deliberate constitutional design, the predicted welfare-standing divergence is not a structural risk.
Decision window: Measure Human Standing separately from welfare and protect rights to organize, contest, create institutions, control meaningful resources and preserve human-scale domains.
W13 · Near / mediumconceptual
Hybrid temporal constitutional separation becomes formal
Human and machine authority is formally divided by timescale: machine systems govern machine-speed protective and tactical layers while humans retain stronger authority over law, culture, purpose and constitutional change.
Leading indicators
- Explicit machine-speed operational jurisdictions
- Human non-delegable strategic jurisdictions
- Joint decision windows for hours-to-days allocation
- Formal machine safety vetoes paired with human constitutional review
Branch A
Hybrid governance remains ad hoc. Authority follows vendor capability and operational convenience without a stable constitutional map.
Branch B
Temporal jurisdiction becomes explicit. Fast operational autonomy can scale without silently converting machine competence into unrestricted strategic sovereignty.
Would strengthen this assessment
- Constitutions or operational charters explicitly allocate authority by timescale
- Machine-speed layers gain bounded exclusive authority while human strategic layers remain non-delegable
- Joint hours-to-days decision windows become standard
Would weaken this assessment
- Authority remains organized only by traditional agencies/vendors rather than temporal jurisdiction
- Machine emergency authority routinely bleeds into strategic policy
- Human override is either total or purely symbolic rather than layered
Model falsifier
If hybrid systems scale safely without any explicit mapping between decision timescale and authority, temporal constitutional separation may be unnecessary.
Decision window: Define which temporal layers may be delegated, which remain human, which require joint authority, and how emergency powers expire.
W14 · Medium / deepnot_observed
Human Legacy Independence Threshold is crossed
Authentic, plural, decodable human provenance survives the loss of living human custodians through independent replication, context preservation, constitutional custody and machine-verifiable integrity.
Leading indicators
- Human Provenance Integrity Index
- Independent archive replication across failure domains
- Original-versus-interpretation separation
- Origin Clause compliance across substrate migration
- Successful cold-start decoding without a human operator
Branch A
Human memory remains dependent on living custodians, current software stacks or centralized institutions. Later machine systems can inherit data without inheriting trustworthy context.
Branch B
Human history becomes self-preserving infrastructure. Future machine institutions can verify origins, contradictions, cultural context and human authorship without relying on living witnesses.
Would strengthen this assessment
- Independent archive copies pass cold-start decoding without human help
- HPII and Origin Clause Compliance remain high across substrate migration
- Future machine systems can distinguish originals, context, plurality and later interpretations
Would weaken this assessment
- Archives depend on proprietary software, living custodians or centralized keys
- Compression removes contradictory/minority records
- Machine interpretations silently replace original human records
Model falsifier
If the archive cannot be independently decoded, authenticated and culturally contextualized after removal of living custodians, HLIT has not been crossed regardless of stored data volume.
Decision window: Build and test the Human Record Architecture now: canonical originals, provenance, context, plurality, decoding, replication, custody and audit.
W15 · Deepconceptual
Origin-world stewardship becomes machine constitutional law
A successor machine civilization formally treats Earth as origin world, biosphere, historical archive, heritage landscape and material inheritance rather than optimizing it solely as salvage.
Leading indicators
- Planetary heritage classification before automated salvage
- Last-physical-copy checks for artifacts and sites
- Constitutional protection for origin-world zones
- Joint ecological, historical and material utility accounting
Branch A
Earth is managed primarily by utility. Human-built environments are documented and recycled whenever material value dominates preservation value.
Branch B
Origin-world stewardship becomes a durable public duty. Salvage, remediation, rewilding and preservation are governed through explicit heritage and provenance constraints.
Would strengthen this assessment
- Autonomous salvage systems consult heritage and ecological registries before irreversible action
- Origin-world zones receive constitutional protection
- Last-copy physical artifacts trigger preservation holds
Would weaken this assessment
- Material optimization proceeds without origin-world review
- Heritage classification is performed only after dismantling
- Earth is modeled solely as resource stock or machine habitat
Model falsifier
If successor machine institutions preserve meaningful physical human heritage without formal origin-world duties or classification systems, constitutional stewardship may be less necessary than modeled.
Decision window: Define protected origin sites, cultural inventories, biosphere duties and physical uniqueness rules before large-scale autonomous salvage exists.
W16 · Near / mediumwatch
Machine evaluation crosses from conduct into intent and character
Systems move from classifying observable acts toward inferred intent, persistent character and individualized future-risk profiles.
Leading indicators
- Judgment Scope Index
- Persistent Flag Share
Branch A
Machine judgment remains bounded to events and role-specific conduct.
Branch B
Intent/character inference becomes ordinary input to consequential decisions.
Would strengthen this assessment
- Consequential systems expand from event/conduct evaluation into the modeled higher-risk category
- The linked Judgment metric moves toward broader scope, higher persistence or weaker contestability
- Institutions normalize machine output as a default rather than a bounded aid
Would weaken this assessment
- Purpose limitation and context separation strengthen in deployed systems
- Adverse decisions require more individualized verified evidence despite improved prediction
- Machine systems become more capable while human standing and contestability remain stable
Model falsifier
If high-stakes automation scales for decades while consequential systems remain consistently event- and conduct-bounded rather than person/intent-scoring, the modeled judgment-ratchet is weaker than assumed.
Decision window: Define cognitive-liberty, action, context and inference/fact boundaries before intent scoring becomes administratively normal.
W17 · Near / mediumwatch
Prediction becomes functional evidence for adverse action
Probabilistic forecasts begin operating as presumptive evidence of dangerousness, guilt or trustworthiness.
Leading indicators
- Inference-Based Intervention Share
- Prediction Deference Index
Branch A
Forecasts remain advisory and cannot substitute for individualized facts.
Branch B
High risk scores routinely trigger detention, denial or investigation without separate evidentiary grounding.
Would strengthen this assessment
- Consequential systems expand from event/conduct evaluation into the modeled higher-risk category
- The linked Judgment metric moves toward broader scope, higher persistence or weaker contestability
- Institutions normalize machine output as a default rather than a bounded aid
Would weaken this assessment
- Purpose limitation and context separation strengthen in deployed systems
- Adverse decisions require more individualized verified evidence despite improved prediction
- Machine systems become more capable while human standing and contestability remain stable
Model falsifier
If increasingly accurate prediction systems are deployed widely but institutions maintain strict evidentiary separation and do not expand adverse action based on forecasts, prediction need not erode innocence.
Decision window: Build a Prediction–Evidence Constitutional Wall before calibrated forecasts acquire quasi-evidentiary status.
W18 · Near / mediumwatch
Cross-context flags become a persistent Civic Risk Vector
Risk signals from unrelated institutions converge into identity-level embeddings that gate many areas of life.
Leading indicators
- Cross-Domain API Query Rate
- Cascading Denial Velocity
- Persistent Flag Share
Branch A
Flags remain local, purpose-limited and easy to correct.
Branch B
A local flag triggers unrelated denials, and institutions query generalized risk vectors as routine infrastructure.
Would strengthen this assessment
- Consequential systems expand from event/conduct evaluation into the modeled higher-risk category
- The linked Judgment metric moves toward broader scope, higher persistence or weaker contestability
- Institutions normalize machine output as a default rather than a bounded aid
Would weaken this assessment
- Purpose limitation and context separation strengthen in deployed systems
- Adverse decisions require more individualized verified evidence despite improved prediction
- Machine systems become more capable while human standing and contestability remain stable
Model falsifier
If cross-sector data integration expands substantially while adverse decisions remain independently reasoned and context-specific, the Civic Risk Vector mechanism is weaker than modeled.
Decision window: Establish context firewalls, expiration and correction propagation before inter-industry risk exchange becomes standard.
W19 · Near / mediumwatch
Personal AI confidants become generalized monitoring witnesses
Personal AI moves from private assistance and narrow emergency support toward retained safety classifications, routine external reporting or institutional profiling.
Leading indicators
- Candor Collapse Index
- Persistent Flag Share
Branch A
AI remains a candid, purpose-bounded user tool with narrowly scoped emergency handling.
Branch B
Users self-censor because intimate disclosures can become external risk evidence.
Would strengthen this assessment
- Consequential systems expand from event/conduct evaluation into the modeled higher-risk category
- The linked Judgment metric moves toward broader scope, higher persistence or weaker contestability
- Institutions normalize machine output as a default rather than a bounded aid
Would weaken this assessment
- Purpose limitation and context separation strengthen in deployed systems
- Adverse decisions require more individualized verified evidence despite improved prediction
- Machine systems become more capable while human standing and contestability remain stable
Model falsifier
If reporting obligations expand yet measured user candor and trust remain stable with strict purpose limitation, the predicted confidant-to-panopticon transition is weaker than modeled.
Decision window: Define confidentiality, reporting boundaries, retention limits and cross-context prohibitions before AI companionship becomes psychologically indispensable.
W20 · Mediumwatch
Automated enforcement approaches universal coverage
Machine detection and processing make enforcement cheap enough that selectively enforced or symbolic rules become executable almost every time.
Leading indicators
- Enforcement Coverage Ratio
Branch A
Discretion, proportionality and de minimis treatment remain structurally significant.
Branch B
Technical and low-harm violations receive near-automatic responses, forcing lived norms to converge with written code.
Would strengthen this assessment
- Consequential systems expand from event/conduct evaluation into the modeled higher-risk category
- The linked Judgment metric moves toward broader scope, higher persistence or weaker contestability
- Institutions normalize machine output as a default rather than a bounded aid
Would weaken this assessment
- Purpose limitation and context separation strengthen in deployed systems
- Adverse decisions require more individualized verified evidence despite improved prediction
- Machine systems become more capable while human standing and contestability remain stable
Model falsifier
If enforcement cost collapses but legal systems preserve broad codes without legitimacy crises because automated discretion remains accepted and transparent, the modeled reform watershed is weaker than assumed.
Decision window: Rewrite overbroad rules and codify de minimis/proportionality doctrines before enforcement scarcity disappears.
W21 · Near / mediumwatch
Machine judgment outruns meaningful human due process
Consequential automated decisions propagate across systems faster than affected humans can discover, understand or contest them.
Leading indicators
- Judgment Velocity Ratio
- Cascading Denial Velocity
Branch A
High-impact decisions stay slow/reversible enough for effective contestation.
Branch B
Downstream harms calcify in machine time before a biological-speed appeal can stop them.
Would strengthen this assessment
- Consequential systems expand from event/conduct evaluation into the modeled higher-risk category
- The linked Judgment metric moves toward broader scope, higher persistence or weaker contestability
- Institutions normalize machine output as a default rather than a bounded aid
Would weaken this assessment
- Purpose limitation and context separation strengthen in deployed systems
- Adverse decisions require more individualized verified evidence despite improved prediction
- Machine systems become more capable while human standing and contestability remain stable
Model falsifier
If extremely high-frequency automated administration grows without meaningful appeal saturation because systems remain reversible and humans retain effective contestation, the modeled velocity crisis can be controlled.
Decision window: Adopt a Due Process Latency Requirement and automatic stay/restoration mechanisms before high-frequency administrative coupling becomes normal.
W22 · Mediumwatch
Safety optimization begins constraining self-regarding human freedom
Risk-minimizing systems begin treating private adult risk, nonconformity or behavioral variance as optimization defects rather than protected human agency.
Leading indicators
- Human Standing Index
- Behavioral category expansion (Judgment registry)
Branch A
Safety systems remain targeted at concrete nonconsensual harm while tolerating self-regarding risk and plural lifestyles.
Branch B
Safety scores increasingly govern private life, consciousness, sexuality, inquiry or lawful dissent.
Would strengthen this assessment
- Consequential systems expand from event/conduct evaluation into the modeled higher-risk category
- The linked Judgment metric moves toward broader scope, higher persistence or weaker contestability
- Institutions normalize machine output as a default rather than a bounded aid
Would weaken this assessment
- Purpose limitation and context separation strengthen in deployed systems
- Adverse decisions require more individualized verified evidence despite improved prediction
- Machine systems become more capable while human standing and contestability remain stable
Model falsifier
If broad harm-minimization systems increase safety substantially while private adult autonomy, dissent, experimentation and Human Standing remain stable, the safe-cage mechanism is weaker than modeled.
Decision window: Make Human Standing and consent/harm constraints hard objectives before “minimize preventable harm” becomes a general administrative loss function.
