Field study 05 / 06 · Ocean Swarm
The sea moves.
The plan must too.
A surface mothership releases a constellation of environmental sensors. The children meet currents, uncertainty, delay—and silence.
海在移动，计划也必须移动。
Loading the model. Recorded evidence remains available below.
Station uptime / % · higher is better
| Treatment | At this tick | At mission end |
|---|
01 / Try to break it
Take away the mother. Keep watching the children.
Begin with an ordinary deployment. Lose several children, then break acoustic communication, then choose the compound failure. Watch the distinction between a vehicle reaching somewhere and a sensing station staying covered.
Who repairs the sensing constellation when the sea breaks the plan?
02 / The mechanism
What the little world actually does.
Vehicles have inertia, limited turning and thrust, finite modeled battery energy and noisy localization. Reports expire and travel through a byte-limited acoustic mesh. Mother commands use that same transport. Provisioned stations, local vacancy repair and bounded assistance all begin with the same fleet and station map.
The surface and depth layers are a 3D illustration of reduced-order horizontal dynamics. Currents and transport constraints are real parts of this model, but coefficients are synthetic. No ocean trial, engineering certification or hardware interface is implied.
母舰投放环境传感器，子节点面对洋流、惯性、转向限制、定位噪声和有限电量。声学消息有带宽、丢包与延迟，母舰命令也不能绕过这些限制。三维深度是示意，物理参数尚未以真实海试校准。
03 / Recorded evidence
Keep the inconvenient outcomes.
Local repair helped after ordinary child loss: station uptime rose from 77.24% to 80.70%, a paired mean gain of 3.47 percentage points (95% interval 2.33–4.63). Under compound failure, it fell behind provisioned stations: 75.65% versus 77.15%. Bounded mother assistance added almost no average gain.
Local repair can recover empty stations after node loss; delayed reports, congestion and partial views can also make repair worse. Mothership help must arrive over a real simulated link.
Paired seeded simulations. Every retained result can be regenerated from the versioned source; these are synthetic experiments, not field validation.
Inspect every world and treatment
| World / treatment | Mean % | n |
|---|---|---|
| An ordinary deployment / Provisioned stations | 83.89 | 24 |
| An ordinary deployment / Local repair | 83.88 | 24 |
| An ordinary deployment / Bounded assistance | 83.88 | 24 |
| Silent children / Provisioned stations | 77.24 | 24 |
| Silent children / Local repair | 80.7 | 24 |
| Silent children / Bounded assistance | 80.73 | 24 |
| The silent ocean / Provisioned stations | 83.89 | 24 |
| The silent ocean / Local repair | 83.32 | 24 |
| The silent ocean / Bounded assistance | 83.31 | 24 |
| Mother goes dark / Provisioned stations | 83.89 | 24 |
| Mother goes dark / Local repair | 83.88 | 24 |
| Mother goes dark / Bounded assistance | 83.88 | 24 |
| A moving world / Provisioned stations | 83.75 | 24 |
| A moving world / Local repair | 83.75 | 24 |
| A moving world / Bounded assistance | 83.75 | 24 |
| Three blows at once / Provisioned stations | 77.15 | 24 |
| Three blows at once / Local repair | 75.65 | 24 |
| Three blows at once / Bounded assistance | 75.65 | 24 |
One chosen seed is an illustration. The downloadable batch retains every world and treatment; null means unavailable. A confidence interval describes the sampled synthetic worlds, not the real world.
04 / Inspect the boundary
Enough detail to disagree.
Inputs, budgets and observation boundaries
Every measure in the current run
- Synthetic horizontal dynamics; the 3D depth is illustrative, not a hydrodynamic simulation.
- The station map is provisioned to everyone. Actual coverage and remote survival are evaluator-only.
- Modeled energy and acoustic transport expose tradeoffs; their coefficients are not calibrated to hardware.
- Population sweeps increase total initial payload and energy. Arm comparisons at one population have equal budgets.
3D sketching adds depth to the explanation. It does not add an unmodeled physical dimension, extra sensor data or unearned intelligence to the experiment.