DEMO 4 / EVIDENCE IN MOTION
Published static preview
Workflow assurance remains unestablished

Is the evidence enough
to support an inspection?

Three demonstrations, one guided civilian inspection story. Explore collection, analyst review and assurance as distinct chapters. Pause at every handoff to see what is known, what is proposed, and why a release can be withheld.

Illustrative walkthrough · no live execution. Retained observations and the separate historical analyst run are not connected by a verified handoff.

DECISION POINT / COLLECTIONILLUSTRATION ONLY
Illustration, not an observed site
7,512Retained rowsHistoricalNot live collectionUncertainTiming & accuracy
PARALLEL ACTORS / EXPLANATORY SEQUENCE
Retained observation · separate analyst history · public reference
No recorded joining run connects these sources.

Actor
What happens · consequence
What we checked · evidence
Why it matters · limit
Inspect this freeze frame
Synchronized explanation, not an execution log. The permission-boundary illustration follows the synthetic reference style; it is not evidence of a retained assurance run.
Playback changes the explanation only. It runs no inspection, model, challenge or command. Space: play/pause · Left/Right: step · Home: restart (when this walkthrough has focus).
Inspect the ten assurance questions
Questions to ask, not a coverage score or certification.
01 / Where did it come from?
Each source keeps its own identity. Retained observations, a separate analyst run and public references are not one joined record.
02 / Are these the same bytes?
Recorded digests identify retained files. A matching digest does not prove that the original observation was true.
03 / Was collection trustworthy?
Capture authenticity is not established. This walkthrough cannot repair that missing proof.
04 / Is it recent enough?
The observations are historical; stale timing and accuracy uncertainty remain relevant.
05 / Does it describe the same place?
The public reference has no location or time join to the observation. The diagram is illustrative only.
06 / Was this handoff authorized?
The retained analyst execution has no demonstrated admission link to these observations.
07 / What does the analyst actually support?
The three branch explanations are prepared examples, not new model results.
08 / What happens if evidence disagrees?
An explicitly injected label conflict withholds the simulated release instead of hiding disagreement.
09 / Can the decision safely refuse?
Insufficient scene association also withholds release. A reported experimental continuation lacks a verified real-refusal run.
10 / What has really been demonstrated?
A separate synthetic consumer path reported acceptance, with unresolved assurance basis and incomplete qualification. No combined live workflow is established.
Inspect retained source records, limitations and historical reports
DEMO 4 / EVIDENCE → DECISION
Know what the evidence supports.
RELEASE HELD · NO LIVE MISSION



Real inputs. Honest boundaries.
A retained DP observation, a real historical Hermes response, and public land-cover references are available. Their existence does not prove they belong to one mission.
CONCEPTUAL MISSION MAP · NO PRIVATE GEOGRAPHYDPHVHELD?ObservationHermesVLM taskingChallenge / gateProposed waypointRETAINED REALDP LINK MISSINGNOT CONNECTEDNO CURRENT VERDICTSIMULATED ONLY
  1. DP observationRETAINED REAL
  2. Hermes handoffDP LINK MISSING
  3. VLM taskingNOT CONNECTED
  4. Challenge / gateHELD · NO CURRENT VERDICT
  5. Proposed waypointSIMULATED ONLY
Dashed edges are proposed handoffs, not traffic or a flight path. This diagram uses no observed coordinates or geographic transform.
RETAINED_REAL_REPLAY
DP observations
7,512
CSV rows, not distinct devices. Stale data and accuracy uncertainty retained.
RETAINED_REAL_REPLAY
Hermes execution
1
Historical real execution; unrelated synthetic inventory input.
PUBLIC REFERENCE
EuroSAT records
3
AnnualCrop, Forest, River. No location/time/mission identity.