The observatory’s evidence infrastructure
A governed path from changing behaviour to reviewable observation.
The FASO System is not FASO itself. It is the technical and human-review instrument through which the planned organisation would observe, preserve, replay, analyse, place and explain evidence. DAMS adds a controlled Version 2 lens for changing agent-bearing systems without giving FASO power to act on them.
The system journey
Eight governed movements from subject identity to external verification.
Select any movement to open its explanatory window. This public architecture view explains purpose, evidence flow and limits without disclosing protected implementation or FASO’s internal analytical method.
01 Subject identityObserved or discovered systemFix the subject, version, scope and observation authority before evidence is interpreted.
- What enters
- A declared or lawfully observed system, its known version or deployment context and the authority under which it may be observed.
- What it establishes
- The strongest supportable subject identity and the distinction between known, declared, observed, provisional and unresolved information.
- What leaves
- A bounded subject record that later stages can reference without silently changing what the subject is.
- What it cannot do
- It cannot infer an undisclosed identity or treat a name, label or apparent similarity as proof of continuity.
02 Controlled evidence intakeEvidence intake and identityReceive evidence under governed conditions and retain where it came from, when it arrived and what it can support.
- What enters
- Authorised declarations, observations, events, reports or other source material together with their timing and source footing.
- What it establishes
- Whether the material is admissible for the next bounded stage and which limitations must travel with it.
- What leaves
- Source-bound evidence with identity, timing, integrity, custody and limitations preserved.
- What it cannot do
- It cannot silently repair missing information, strengthen an uncertain source or convert absence of evidence into evidence of absence.
03 Governed analytical routingSystem recognition and controlled routingRecognise the applicable kind of system and use the accepted route without losing evidence identity or authority.
- What enters
- Admissible evidence, the bounded subject record and the controlled route definitions applicable to the observed system.
- What it establishes
- Which governed analytical path is applicable and which evidence remains unavailable, inapplicable or unresolved.
- What leaves
- A route-bound analytical packet that retains its subject, source and authority references.
- What it cannot do
- It cannot create a new system type for convenience or promote a transient agent event into a new model identity.
04 Bounded comparisonControlled analysisCompare admitted evidence against the correct governed context while preserving uncertainty and competing explanations.
- What enters
- The route-bound packet, applicable historical context, comparison footing and declared limitations.
- What it establishes
- The strongest analytical posture supported by the available evidence, including when the result remains unresolved or insufficient.
- What leaves
- A traceable result with its evidential basis, uncertainty, limitations and replay references intact.
- What it cannot do
- It cannot declare a system safe, infer intent or convert a candidate condition into verified change without the required evidence.
05 Evidence-supported state placementState placement and visual explanationPlace the supported result in a reviewable state and show how the evidence reached that position.
- What enters
- The traceable analytical result, its supported state, its limitations and the evidence needed to explain its placement.
- What it establishes
- A bounded visual and narrative representation of the supported position at the applicable observation point.
- What leaves
- A reviewable state result linked back to its source evidence and forward to human review.
- What it cannot do
- It cannot improve the evidence, inflate certainty or manufacture a visual result for a state the evidence does not support.
06 Knowledge continuity and governed replayKnowledge and replayPreserve admitted history and reconstruct earlier handling under the authority that applied at the time.
- What enters
- Source evidence, governed results, corrections, limitations, time order and the authorities that governed each observation.
- What it establishes
- Whether an earlier result can be reconstructed and compared without silently applying a later rule to an earlier state.
- What leaves
- Append-only institutional memory and replay material that retains predecessor and successor truth.
- What it cannot do
- It cannot overwrite an earlier event, replace source evidence with a summary or retrospectively clean up an unresolved state.
07 Decision gates and auditGate, certainty and auditApply non-bypass gates, express only warranted certainty and retain the complete review path.
- What enters
- The supported state, evidence sufficiency, limitations, replay posture and the authority for any proposed outward use.
- What it establishes
- Whether the record may advance, must be held, needs more evidence or is ineligible for a stronger claim.
- What leaves
- A gated, auditable posture with the reason for advance, hold, deferral or rejection made explicit.
- What it cannot do
- It cannot bypass a missing condition, hide an adverse result or turn an internal system output into an external conclusion.
08 Bounded publication and external challengeHuman review, publication and verificationRequire human judgement before releasing only the safe, supported and appropriately limited part of the record.
- What enters
- A gated publication candidate, its evidence, limitations, redaction needs, authority and review history.
- What it establishes
- What may be explained publicly, what must remain protected and what independent reviewers may examine.
- What leaves
- A bounded outward artefact or an explicit hold, deferral, downgrade or rejection record.
- What it cannot do
- It cannot expose protected implementation, imply endorsement or make operational, safety or causal claims beyond the admitted evidence.
Version 2 · Dynamic Agent Monitoring System
DAMS follows changing agent-bearing systems without becoming part of them.
DAMS is the public-facing name used here for FASO’s controlled dynamic-agent observation work. It is a cross-cutting layer across the existing system—not a ninth model type, not an autonomous actor and not a new source of analytical truth.
Linked identity layers · never collapsed
One changing system can contain several different kinds of identity.
The links describe governed context. They do not merge the identities, prove that an interaction occurred or attribute a result to an unseen internal cause.
01Identity
- What it observes
- The core model, deployment, repeatable agent role, individual agent occurrence and applicable relationship-structure period as distinct layers.
- What it preserves
- Separate identity and lineage even when several agent occurrences share one role or several roles share one core model.
- What it cannot conclude
- A new agent occurrence is not automatically a new model, model version, baseline or relationship structure.
02Lifecycle
- What it observes
- Source-bound creation, activation, modification, delegation, suspension, resumption, completion, termination or reappearance events.
- What it preserves
- Append-only event order, timing, source, parent or supervisor context, corrections and known limitations.
- What it cannot conclude
- A lifecycle event does not by itself prove drift, intent, fault, concealment, danger or developer knowledge.
03Relationships
- What it observes
- The declared or observed structure linking models, agent roles and material agent occurrences during a bounded period.
- What it preserves
- Direction, role, epoch, predecessor structure and unresolved changes without collapsing component-level results.
- What it cannot conclude
- A relationship map establishes comparison structure; it does not by itself prove that an interaction occurred or caused a result.
04Coverage
- What it observes
- Declared, independently observed, bounded sampled or unavailable population footing together with telemetry, exclusions and backlog.
- What it preserves
- The denominator, observation interval, source footing and every condition that limits a completeness or absence claim.
- What it cannot conclude
- Missing telemetry is not proof of absence. Not observed is not the same as did not occur.
What DAMS can show
Change in identity, lifecycle, relationships and the available observation coverage over time.
What remains separate
Each component result, event, relationship, limitation and observation period retains its own evidence footing.
What DAMS cannot do
FASO observes. It does not control, direct, interrupt, repair or otherwise act upon the observed system.
FFD · Formation and Family Detection
Recognising new agents and the relationships between them.
As an AI system delegates work, new agents can appear and related groups can grow. FFD is designed to recognise evidence of that formation and help FASO understand the connections between agents.
What we mean by an agent family
An agent family is a group connected by evidence of a shared origin or recorded relationships. Similar names, similar behaviour or use of the same underlying model are not enough to establish that connection.
Why it matters
A view of one agent may miss the wider pattern of activity. Keeping the group’s history alongside each member’s record gives human reviewers a fuller account of what changed, when it changed and which relationships are supported by evidence.
What the evidence can establish
FFD’s conclusions must stay within the evidence FASO can legitimately observe or receive, with gaps made clear. The appearance of an agent or growth of a family does not, by itself, establish model-derived drift or danger. A finding about one member does not automatically apply to the whole group.
Technology and people
The system forms evidence. People remain responsible for judgement and publication.
FASO is not designed as an autonomous authority. Human review, independent challenge and organisational governance are load-bearing parts of the proposed system.
Technology preserves discipline
Identity, authority, sequence, repeatability, audit and controlled presentation.
Human reviewers interpret
Context, significance, ambiguity, limitations, alternatives and the need for corrective or further work.
Governance controls authority
What may enter, what may be concluded, what may be published and who may challenge it.
Evidence of the system
See what has been built and how it is being tested.
The evidence and validation pages separate build proof, run proof and future independent proof.