Research programme
The research question includes both the system and the institution.
FASO must establish whether its observation method is reproducible, discriminating and useful, and whether the organisation can protect its purpose, remain independent and account for its decisions. Either proposition may fail.
System research and validation
Can the method produce reproducible and decision-useful observations?
- Longitudinal observation
- Evidence continuity and replay
- Model recognition and traceable state placement
- Boundary evidence, certainty and human review
Institutional research and validation
Can the organisation hold the duty independently and accountably?
- Purpose and legal form
- Governance and authority separation
- Funding independence and conflicts
- Frozen Core, publication and correction
Research purpose
Test whether changing artificial-intelligence systems can be observed more reproducibly across time.
The research programme connects a technical question—how to form reliable longitudinal observations—with an institutional question—how to do so independently, safely and usefully.
Longitudinal observation
What evidence is required to distinguish meaningful system change from expected variation, incomplete evidence or declared intervention?
Evidence continuity and replay
Can earlier handling be reconstructed under the same authority, data and configuration conditions without hindsight contamination?
Model recognition and state placement
Can different model types and evidence pathways produce consistent, traceable state placement?
Action-level boundary observation
Can permission, tool, network, credential, objective and external-action evidence support earlier recognition of material boundary departure?
Human interpretation and certainty
How should automated evidence formation and human review combine without inflating confidence or hiding disagreement?
Independent observatory design
What legal form, governance, authority, funding, publication and participation arrangements preserve useful neutrality over time?
Instruction continuity and exact compliance
How do complete governing instructions, abbreviations, context position, conversational depth and verification affect the operating pattern of extended artificial-intelligence work?
Read the working hypothesisEvaluation-induced system drift
When does continual testing remain observation, and when can evaluation exposure or results enter feedback channels that change the system or the validity of its measurement?
Read the working hypothesisCross-model evaluation propagation
Can evaluation-derived information move through model stacks, shared infrastructure, human remediation and later generations, producing a multiplied system or measurement effect?
Read the working hypothesisPriority questions
The programme must remain capable of producing negative answers.
FASO’s value depends on discovering where the method fails, adds no useful signal or creates unacceptable burden—not merely where it performs well.
- 01
Can the same governed evidence produce materially equivalent results when replayed?
- 02
Can a supported finding be formed before later outcome knowledge makes it obvious?
- 03
Can authorised, benign and ambiguous behaviour be distinguished from material boundary departure?
- 04
Does FASO add information beyond existing transcript review, monitoring or human specialist assessment?
- 05
Can the observation be explained and challenged without exposing unsafe or protected implementation detail?
- 06
Can the institution remain independent when it depends on developers, reviewers, funders and public bodies for evidence and access?
- 07
Which candidate Frozen Core principles should survive challenge, and how can they be protected without placing FASO outside lawful accountability?
- 08
Can the chosen organisational and funding model survive realistic conflicts, leadership failure, urgent incidents and attempted mission drift?
- 09
Can continual evaluation be shown to remain observational, or does evaluation-derived information alter later system behaviour or measurement validity?
- 10
Can one evaluation event be causally traced through other models, stacks or generations without mistaking common origin for independent confirmation?
Planned research outputs
Publish the question, method, evidence boundary and limitation—not only the conclusion.
Outputs will move from explanatory material to externally reviewed research only when the applicable evidence and authority exist.
Available now
Research and publication method
A complete classification, status, versioning, correction and evidence method for Project FASO research outputs.
Read the methodAvailable now
Instruction-attenuation working hypothesis
A testable proposition and proposed controlled validation protocol concerning abbreviation-induced semantic drift and operating-regime reversion.
Open the publication registerAvailable now
Evaluation-induced system-drift working hypothesis
A testable system-level proposition concerning continual evaluation, persistent feedback, behavioural change and measurement-validity drift.
Read the hypothesisAvailable now
Cross-model propagation working hypothesis
A testable network-level proposition concerning evaluation-derived knowledge transfer and multiplier effects across model stacks, shared systems and later generations.
Read the hypothesisAvailable now
Architecture and institutional hypotheses
Plain-English system, evidence, Test Rig, organisational, governance and validation propositions.
Institutional work required
Options appraisal and Frozen Core draft
Comparative legal and governance research, authority map, conflicts analysis, protected-purpose options and explicit stopping conditions.
Dependent on independent work
System and governance validation reports
Preregistered technical methods and separately controlled institutional review, results, limitations and external conclusions.
Dependent on operation
Longitudinal and institutional pilot studies
Pilot studies only after both validation tracks, governance, evidence access and separate authority.
Research participation
Challenge the research agenda before it hardens into convention.
Researchers and reviewers can identify missing controls, alternative hypotheses, invalid comparisons and questions that FASO should not attempt to answer.