READY
Waiting for a mission
The panel shows READY and REMOTE START.
Next question or action
An operator-authorised rehearsal can begin. Readiness is a display state, not proof that every subsystem passed a pre-flight check.
EXPLORE / QUADCOPTER SURVEY
Learn to read
the swarm.
An observation becomes a report. A second viewpoint can support it—or leave a question open. Follow that story across the panels.
These are code-rendered documentation fixtures from a scripted survey rehearsal. They explain the display vocabulary; they do not record autonomous flight, live packets or an agent’s private reasoning.
01 / THE MISSION STORY
Prepare. Observe.
Recover.
The objective gives the same wall a different emphasis. Observe foregrounds the survey; Compare reports foregrounds checking and disagreement; Recovery foregrounds returning and availability.
01
Prepare
A shared start and an objective.
-
READYWait for a run
-
ORDERSRead the objective
-
TAKEOFFBegin the rehearsal
-
DEPLOYForm up
02
Observe
Notice, report and seek another viewpoint.
-
SCANLook for a candidate
-
SEENNotice something
-
SHAREOffer a report
-
CHECKSeek corroboration
03
Recover
Returning is a valid outcome, even without a finding.
-
R2BReturn to base
-
LANDFinish the return
-
CHARGEShow replenishment
-
STANDBYWait for another role
Selected milestones, not a complete state machine. A node can wait, lose an observation or return without following every step. The camera, altitude and battery graphics describe a simulated situation.
Open the original timeline diagram ↗02 / QUESTIONS BEHIND THE PICTURE
What would justify the next action?
An explanatory path for the observer. The scripted scheduler does not execute this decision tree.
-
Which objective is active?
Read the briefing. A survey, a comparison and a recovery sequence ask different things of the group.
-
Continue, wait or return?
Completion and availability matter too. Returning without a positive observation can be the appropriate outcome.
-
What did this node observe?
Separate a fresh private observation from a peer’s report, a stale view or missing information.
-
Would another viewpoint help?
SHARE invites comparison. CHECK asks for corroboration. Two repeated reports may still come from one underlying source.
-
Does the evidence agree?
If it conflicts, keep uncertainty visible. DOUBT is a useful prompt to investigate rather than manufacture certainty.
-
What changed next?
Connect the report to the decision and its outcome. That is where an experiment begins.
03 / READ THE BORDER AND THE PEER
A report invites
another viewpoint.
Orange / TX means sending.
Green / RX means receiving. In this
illustrated pair, A1 shows TX B1 and B1 shows
RX A1. The footer identifies the peer; it does not show
the message payload.
0–250 ms
First flash
250–375 ms
Gap · peer label remains
375–625 ms
Second flash
625 ms onward
Normal footer returns
Four still samples of the same double flash: 125, 312.5, 500 and 750 ms. The one-pixel border is on for 250 ms, off for 125 ms, on for 250 ms, then off. These paired indicators are constructed from scenario time and node identities.
Receipt is not agreement.
TX ALL marks a broadcast; each receiver still names
the source. A matching flash does not by itself establish radio
delivery, an acknowledgement or a change of decision.
Keep three different problems separate: LOST means the observation is unavailable; NOLINK represents a stale view; DOUBT presents conflicting claims.
READY, APPROVED, ORDERS and BLANK bypass the general border overlay. The optional ARGB channel grid is another way to make communication visible, described in the swarm build guide.
Open the original communication diagram ↗04 / ONE OBSERVATION, DISTINCT ROLES
Who reports?
Who takes the task?
A separate twelve-second browser fixture illustrates shared ownership. The roles and P1 label are predetermined; the sequence explains a coordination story rather than demonstrating a learned auction or consensus algorithm.
-
ALERT
A source announces beacon P1 and broadcasts to the group.
-
PRIORITY
P1 is displayed first in the task order. No new event is sent.
-
CLAIM / CHECK
Two selected nodes offer to inspect. The fixture shows claims addressed to the source at 4.000 and 4.875 seconds.
-
ASSIGN / YIELD
The source broadcasts an assignment at 6 seconds. One node owns the task; the others yield.
-
SEEN / SCAN
The owner observes while the others return to scanning.
The broadcaster and the observation owner can be different nodes. “I sent the assignment” and “I own this task” describe different roles.
05 / THE PANEL VOCABULARY
Twenty-five states.
Six ways to read them.
Open a group, then a state’s next question. These 64 × 64 firmware frames are enlarged without smoothing. Their descriptions are an observer’s interpretation of a documented fixture.
Prepare · 5 states
APPROVED
Mission acknowledged
MISSION APPROVED blinks on a black panel.
Next question or action
Show the mission briefing. This is not a real aircraft arming signal.
ORDERS
Understand the objective
The briefing shows SURVEY, COMPARE or RECOVER with SHARE DATA and KEEP CLEAR.
Next question or action
Interpret later observations against the selected objective. The illustration uses the survey briefing.
TAKEOFF
Enter the simulated world
The camera and altitude presentation transition into the flight view.
Next question or action
Proceed to the deployment presentation. No motor command is produced by this renderer.
DEPLOY
Form up
DEPLOY and FORM UP accompany the simulated movement.
Next question or action
Begin the observation phase. The moving picture alone does not establish a computed flight plan.
Observe · 5 states
SCAN
Look for an observation
A scanning overlay marks the observation phase.
Next question or action
Keep searching, or enter SEEN when the scenario presents a candidate observation.
LOITER
Continue observing
LOITER is accompanied by a GRID or ORBIT caption.
Next question or action
Continue observing, respond to a review request, or move into the recovery branch when the scenario requires it.
SEEN
A candidate observation
A bracket highlights an observation in the scene.
Next question or action
Share a report or seek a check. A bracket is not a confidence score or proof of correct identification.
LOST
Observation no longer available
LOST is accompanied by SEARCH.
Next question or action
Revisit the observation task or record that the evidence is unavailable. It does not mean that the network link has failed.
HOLD
Wait for coordination
HOLD and HQ WAIT distinguish waiting from active observation.
Next question or action
Continue when the scenario supplies the next instruction. A waiting panel is not evidence that a model is currently thinking.
Communicate · 3 states
SHARE
Offer a report
SHARE normally shows REPORT. During a communication event, the footer can show TX plus the peer label.
Next question or action
A receiver can consider the report alongside its own observation. The current scripted flash does not itself deliver a real payload.
CHECK
Seek corroboration
CHECK and VERIFY identify a review state; an RX footer can temporarily replace VERIFY.
Next question or action
Continue observation if evidence is incomplete; record agreement or disagreement in an instrumented experiment. Do not equate receipt with agreement.
DOUBT
Keep uncertainty visible
DOUBT, a contrasting reticle and the 2 CLAIMS caption show a conflict presentation.
Next question or action
Request another check, keep the observation unresolved, or resume the survey. Two claims can repeat the same source and are not necessarily independent.
Coordinate · 5 states
ALERT
Announce a shared observation
ALERT shows BEACON P1. In the coordination fixture, its source transmits to the other panels.
Next question or action
Move into the fixture's review sequence. P1 is a scripted label, not an inferred urgency or confidence value.
PRIORITY
Display the selected task order
PRIORITY displays P1 FIRST.
Next question or action
The fixture presents candidate volunteers. This display does not implement a learned priority calculation.
CLAIM
Offer to inspect
CLAIM displays CLAIM P1. Two predetermined participants send their claims separately to the source in the preview fixture.
Next question or action
Wait for the assignment presentation. A claim is an offer, not ownership or evidence that the node is best placed.
ASSIGN
Show the observation owner
ASSIGN displays OWNER and a node label.
Next question or action
The chosen observer continues observing. The fixture chooses the owner deterministically; no bidding optimisation is demonstrated.
YIELD
Avoid duplicate ownership
YIELD displays the same OWNER label as the assignment.
Next question or action
Other participants can return to their survey roles. This is a presentation of coordination, not proof of distributed consensus.
Recover · 4 states
R2B
Return to the base
R2B and HOME identify the return presentation.
Next question or action
Proceed toward LAND. The example does not specify a real aircraft's reserve threshold or navigation policy.
LAND
Finish simulated flight
The camera/altitude presentation descends into the home view.
Next question or action
Show CHARGE or STANDBY according to the scripted outcome.
CHARGE
Recharge presentation
CHARGE includes an animated battery indicator.
Next question or action
Return to standby when the sequence advances. The indicator is simulated, not a measured charging current.
STANDBY
Ready for the next role
STANDBY presents the parked state.
Next question or action
Remain available for a subsequent authorised rehearsal.
Exceptions · 3 states
NOLINK
Stale view or missing link
NOLINK freezes the camera presentation and shows an AGE caption.
Next question or action
Treat the represented information as stale. In this scripted mode, AGE is scene time rather than measured packet latency.
EMPTY
Simulated energy exhausted
The battery indicator falls and the view dims.
Next question or action
Record the simulated failure outcome. This is not a physical battery measurement or an instruction for aircraft recovery.
CRASH
Simulated failure outcome
The view rolls toward the ground, then later shows signal loss.
Next question or action
Preserve the event trace for review. The thumbnail shows an early phase of the display sequence, not a photograph or real crash report.
This selected atlas predates the latest rescue-language update. The current public renderer uses EVENT and MARK for two additional states outside this selection. These twenty-five fixtures are preserved as published; they are not a complete catalogue of the current firmware.
06 / FROM A DISPLAY TO A FINDING
Did the message change the outcome?
The wall gives human observers something to notice. To test whether communication helped, the AI experiment team needs a record connecting what a node knew to what it did.
- Observe
- Record the private observation before the report arrived.
- Receive
- Identify the source, payload, message ID and actual receipt time.
- Decide
- Record the action, uncertainty and a concise reason code at decision time.
- Evaluate
- Compare the outcome with separately held ground truth.
- Compare
- Use the same seed and observations with messages withheld or scrambled.
A reviewer’s later explanation is different from a recorded decision. The separate communication-learning lab tests receiver learning with assigned sender vocabularies; it does not establish that the panel renderer learned a language.
07 / SOURCES AND VERSION
Pixels from code.
Context alongside.
Adapted from the project’s
Reading the swarm field guide. Images and diagrams are copied unchanged from public snapshot
2c02336; source hashes identify the earlier renderer
used to generate them.
Version details and reproducible sources
The published manifest’s pixel UI, font, communication fixture, coordination fixture and learning-lab hashes match this snapshot. Its model and renderer hashes differ following the rescue-language update. The illustrations are retained as documentation fixtures, rather than presented as newly rendered firmware output.