Source literacy · New in v1.5
Work through a deterministic set of public-source examples. Decide whether each statement is a verified fact, an official or manufacturer description, independent analysis, a disputed claim, a public unknown, or a fictional teaching value—then inspect what the source actually supports.
Answer-first summary
It is a local-only exercise that asks visitors to classify public claims by source type, inspect lineage, preserve unknowns, and identify what each record does not establish.
Read the supporting sectionA manufacturer page establishes what the developer publicly says. Independent verification requires evidence outside the organization whose capability or performance is being described.
Read the supporting sectionClassified or undisclosed operating modes, authority arrangements, thresholds, and abort paths cannot responsibly be guessed in either direction.
Read the supporting sectionOrientation
The goal is not to distrust everything. It is to identify who is speaking, whether the evidence is independent, what exact proposition is supported, and which facts remain unknown.
Classification key
The interface uses text, shape, border treatment, and explanation rather than color alone. These labels describe the evidence state of the proposition—not the moral worth of a source or institution.
A narrow proposition directly established by an inspectable public record or corroborated evidence. The label applies only to what the record actually says.
A government, program office, military service, alliance, or operator describes policy, purpose, status, testing, or intended control. It is authoritative for the institution’s statement, not independent proof of every fielded configuration or outcome.
A developer or vendor publicly describes a capability. The source establishes what the manufacturer says, not independently verified operational performance, doctrine, or control mode.
A research organization or analyst evaluates public evidence and uncertainty. Analysis can strengthen context, but it remains interpretation unless it directly corroborates the specific proposition at issue.
A consequential claim is contested, based on a belligerent, manufacturer, anonymous, or otherwise incomplete account, or lacks enough public evidence for confirmation.
Reviewed public sources do not establish the relevant mode, threshold, authority, abort path, configuration, or operational behavior. Unknown must not be converted into either “human-controlled” or “fully autonomous.”
A scenario value, timing, confidence score, source packet, or outcome invented for instruction. It is not evidence about a real system.
Three-to-five-minute exercise
Classify each card, inspect the source lineage, and read the explanation before moving on. No answer is sent anywhere.
Evidence Lab ready.
Evidence card 1 · Public-source claim
The CDAO describes Maven Smart System functions including sensor-data fusion, object detection, tracking, and decision support.
Lineage reveal
Multiple documents do not become independent merely because they are formatted differently. Follow the material proposition back to its origin.
cdao-maven
Classification explanation
The claim is an official program description from the responsible government organization. It establishes the described support functions but should not be converted into a claim that the software independently authorizes weapon release.
Government or operator statement. The claim is an official program description from the responsible government organization. It establishes the described support functions but should not be converted into a claim that the software independently authorizes weapon release.
Unknowns: The public summary does not disclose all model thresholds, false-positive rates, source weighting, operator workflow, or mission-specific authority rules.
Does not establish: AI-enabled fusion and recommendation do not equal autonomous engagement authority, and the program description is not an independent operational audit.
Evidence card 2 · Public-source claim
The Navy fact file supports the proposition that Phalanx can automatically perform a bounded close-in defense sequence; that fact alone does not establish that its core operational logic uses machine learning.
Lineage reveal
Multiple documents do not become independent merely because they are formatted differently. Follow the material proposition back to its origin.
usn-phalanx
Classification explanation
The official fact file directly supports automatic search, detection, evaluation, tracking, engagement, and assessment. The statement also preserves the critical distinction between documented automation and unproven AI or machine-learning use.
Verified public fact. The official fact file directly supports automatic search, detection, evaluation, tracking, engagement, and assessment. The statement also preserves the critical distinction between documented automation and unproven AI or machine-learning use.
Unknowns: Detailed threat logic, mode-specific crew permissions, current configuration, and classified performance are not disclosed in the public fact file.
Does not establish: “Automatic” is not a synonym for “AI,” and technical capability does not reveal the selected operating mode of every real engagement.
Evidence card 3 · Fictional teaching packet
Three fictional briefings repeat the same Iron Dome capability sentence from one manufacturer page, so the dashboard counts three independent confirmations.
Lineage reveal
Multiple documents do not become independent merely because they are formatted differently. Follow the material proposition back to its origin.
rafael-iron-dome
Classification explanation
The three briefings are invented teaching artifacts and share one source root. Repetition is not independent corroboration. The exercise deliberately demonstrates circular reporting without asserting that any real publication copied another.
Fictional simulation value. The three briefings are invented teaching artifacts and share one source root. Repetition is not independent corroboration. The exercise deliberately demonstrates circular reporting without asserting that any real publication copied another.
Unknowns: The exercise does not assess real journalists, real source-sharing arrangements, or actual Iron Dome performance.
Does not establish: Three documents are not three independent sources when all derive the material proposition from the same origin.
Evidence card 4 · Public-source claim
A CSIS report analyzes public evidence and uncertainty surrounding Russia’s development of AI-driven drone autonomy; it does not present itself as direct telemetry from every named engagement.
Lineage reveal
Multiple documents do not become independent merely because they are formatted differently. Follow the material proposition back to its origin.
csis-russia
Classification explanation
CSIS is separate from the manufacturer and belligerent whose capabilities are being examined. Its work is valuable independent analysis, but analysis is not automatically incident-level corroboration.
Independent analysis. CSIS is separate from the manufacturer and belligerent whose capabilities are being examined. Its work is valuable independent analysis, but analysis is not automatically incident-level corroboration.
Unknowns: Exact software versions, selected operating modes, target-specific human approval, system logs, and named-incident telemetry are generally unavailable publicly.
Does not establish: It does not prove that every Russian autonomy claim is false or true, nor that a specific system selected and engaged a particular target without human approval.
Evidence card 5 · Public-source claim
A ZALA article’s description of “intelligent targeting and recognition” proves that a named combat engagement used autonomous target selection and lethal release without human approval.
Lineage reveal
Multiple documents do not become independent merely because they are formatted differently. Follow the material proposition back to its origin.
zala-lancet
csis-russia
Classification explanation
The ZALA item is a manufacturer and belligerent claim. An independent analysis of the broader ecosystem adds context but does not validate the exact control mode of the named engagement. The conclusion therefore remains alleged or disputed.
Disputed or alleged. The ZALA item is a manufacturer and belligerent claim. An independent analysis of the broader ecosystem adds context but does not validate the exact control mode of the named engagement. The conclusion therefore remains alleged or disputed.
Unknowns: The selected mode, target-selection locus, human authorization, logs, configuration, and independent incident record are not established by these sources.
Does not establish: Marketing or belligerent language about intelligent recognition does not by itself prove strict human-out-of-the-loop lethal use in a particular incident.
Evidence card 6 · Public-source claim
The U.S. Air Force says that a human retains authority over weapons-release decisions throughout the Collaborative Combat Aircraft development and testing activity described in its February 2026 update.
Lineage reveal
Multiple documents do not become independent merely because they are formatted differently. Follow the material proposition back to its origin.
usaf-cca-control
Classification explanation
The Air Force is the relevant program operator and explicitly states the authority arrangement for the described development and testing context. That makes it an official operator statement rather than independent validation of every future configuration.
Government or operator statement. The Air Force is the relevant program operator and explicitly states the authority arrangement for the described development and testing context. That makes it an official operator statement rather than independent validation of every future configuration.
Unknowns: The final operational interface, lost-link behavior, intervention windows, and whether later fielded configurations use identical rules remain publicly incomplete.
Does not establish: It does not prove that all collaborative aircraft, all future CCA increments, or every autonomous aircraft worldwide retain the same release arrangement.
Evidence card 7 · Public-source claim
The reviewed public record specifies that every IBCS track-correlation function uses machine learning and that the system autonomously authorizes firing in every configuration.
Lineage reveal
Multiple documents do not become independent merely because they are formatted differently. Follow the material proposition back to its origin.
army-ibcs
Classification explanation
The Army source supports integrated sensors, command-and-control nodes, and effectors. It does not establish universal machine-learning use or autonomous firing authority across configurations, so those details must remain publicly unspecified.
Publicly unspecified. The Army source supports integrated sensors, command-and-control nodes, and effectors. It does not establish universal machine-learning use or autonomous firing authority across configurations, so those details must remain publicly unspecified.
Unknowns: Configuration-specific fusion algorithms, model use, operator permissions, automated engagement settings, and deployment doctrine remain incomplete in the reviewed public record.
Does not establish: An integrated sensor-to-effector architecture does not by itself prove that AI performs every function or that the system independently authorizes force.
Evidence card 8 · Public-source claim
DoD Directive 3000.09 was issued on January 25, 2023 and publicly requires autonomous and semi-autonomous weapon systems to be designed so commanders and operators can exercise appropriate levels of human judgment over the use of force.
Lineage reveal
Multiple documents do not become independent merely because they are formatted differently. Follow the material proposition back to its origin.
dod-3000
Classification explanation
The issuance date and quoted policy requirement are directly inspectable in an official public directive. The correct classification is narrow: it verifies the policy text, not successful implementation by every system.
Verified public fact. The issuance date and quoted policy requirement are directly inspectable in an official public directive. The correct classification is narrow: it verifies the policy text, not successful implementation by every system.
Unknowns: System-specific compliance evidence, test results, interfaces, operator workload, and the practical quality of human judgment may be classified or otherwise unavailable.
Does not establish: It does not establish that any named system is lawful, reliable, safe, meaningfully supervised, or compliant merely because the policy exists.
Evidence card 9 · Public-source claim
Rafael describes Iron Dome as using radar, battle management, launchers, interceptors, selective defense, and simultaneous handling of multiple threats.
Lineage reveal
Multiple documents do not become independent merely because they are formatted differently. Follow the material proposition back to its origin.
rafael-iron-dome
Classification explanation
The claim is traceable to the system manufacturer. It is appropriate to say “Rafael describes” these functions, while keeping performance, universal operating mode, and implementation details separate from independently corroborated fact.
Manufacturer-described. The claim is traceable to the system manufacturer. It is appropriate to say “Rafael describes” these functions, while keeping performance, universal operating mode, and implementation details separate from independently corroborated fact.
Unknowns: Public material does not establish every engagement threshold, launcher-assignment rule, authorization setting, fallback mode, or software technique used in every deployment.
Does not establish: It does not independently verify a success rate, prove one universal human-authorization arrangement, or establish that the core operational logic depends on machine learning.
Classification report
Evidence quality depends on source class, independence, bounded support, and visible unknowns.
Why this matters
Source classification is a control against false equivalence. It prevents manufacturer marketing, official self-description, analysis, allegation, and genuine public unknowns from collapsing into one authoritative-looking answer.
Automation
A system can complete a fast deterministic sequence without a learned model. The site uses “AI” only when a reliable public source establishes it.
Compare the termsAuthority
Detection, fusion, ranking, navigation, and recognition can be automated while a human retains authority over an external effect.
Open the Human Control LabUnknown
Classified or undisclosed authority must remain publicly unspecified rather than being guessed in either direction.
Read the evidence method