# Results

Generated by `python -m experiments.analysis` from `results/*.json` (20 seeds per condition). All comparisons are paired by seed and use the two-sided Wilcoxon signed-rank test; differences with p < 0.05 are reported as significant, with no correction for multiple comparisons.

## Scale study (Table 4, Figure 3)

T = 800 steps. Paired Wilcoxon against D-MORPH.

| Scale | Method | Deliveries (mean ± SD) | % of Full Graph | Active links (% of pairs) | D-MORPH minus method [95% CI], p |
|---|---|---|---|---|---|
| Tiny (8) | Full Graph | 55.4 ± 8.6 | 100.0% | 28.0 (100%) | +1.40 [-1.37, +4.17], p = 0.4742 |
| Tiny (8) | D-MORPH | 56.8 ± 5.0 | 102.5% | 6.6 (23%) |  |
| Tiny (8) | MORPH (centralised) | 53.3 ± 9.6 | 96.2% | 21.5 (77%) | +3.50 [-1.90, +8.90], p = 0.2305 |
| Tiny (8) | Proximity-r | 53.5 ± 12.8 | 96.7% | 10.7 (38%) | +3.25 [-3.53, +10.03], p = 0.9105 |
| Tiny (8) | Random-Sparse (fixed) | 16.1 ± 15.7 | 29.0% | 6.0 (21%) | +40.75 [+33.53, +47.97], p = 0.0001 * |
| Tiny (8) | Random-Sparse (per-step) | 35.5 ± 13.6 | 64.0% | 6.0 (21%) | +21.35 [+15.64, +27.06], p = 0.0001 * |
| Tiny (8) | TSG | 53.3 ± 8.8 | 96.2% | 2.6 (9%) | +3.50 [-1.41, +8.41], p = 0.0655 |
| Small (12) | Full Graph | 64.6 ± 6.5 | 100.0% | 66.0 (100%) | +1.40 [-3.54, +6.34], p = 0.1778 |
| Small (12) | D-MORPH | 66.0 ± 8.0 | 102.2% | 11.4 (17%) |  |
| Small (12) | MORPH (centralised) | 64.0 ± 5.0 | 99.1% | 22.8 (34%) | +1.95 [-2.18, +6.08], p = 0.0638 |
| Small (12) | Proximity-r | 66.8 ± 6.3 | 103.3% | 15.1 (23%) | -0.75 [-4.33, +2.83], p = 0.8243 |
| Small (12) | Random-Sparse (fixed) | 18.8 ± 12.9 | 29.0% | 12.0 (18%) | +47.25 [+40.28, +54.22], p = 0.0001 * |
| Small (12) | Random-Sparse (per-step) | 33.1 ± 14.8 | 51.2% | 12.0 (18%) | +32.90 [+24.50, +41.30], p = 0.0001 * |
| Small (12) | TSG | 64.9 ± 3.7 | 100.5% | 3.7 (6%) | +1.10 [-2.91, +5.11], p = 0.1300 |
| Medium (18) | Full Graph | 84.2 ± 7.5 | 100.0% | 153.0 (100%) | +6.50 [+3.20, +9.80], p = 0.0006 * |
| Medium (18) | D-MORPH | 90.7 ± 2.5 | 107.7% | 17.6 (12%) |  |
| Medium (18) | MORPH (centralised) | 88.3 ± 2.1 | 104.9% | 20.3 (13%) | +2.35 [+1.18, +3.52], p = 0.0010 * |
| Medium (18) | Proximity-r | 85.4 ± 12.8 | 101.4% | 24.0 (16%) | +5.30 [-1.24, +11.84], p = 0.2444 |
| Medium (18) | Random-Sparse (fixed) | 12.4 ± 11.4 | 14.8% | 18.0 (12%) | +78.25 [+72.63, +83.87], p = 0.0001 * |
| Medium (18) | Random-Sparse (per-step) | 47.1 ± 17.4 | 55.9% | 18.0 (12%) | +43.60 [+35.17, +52.03], p = 0.0001 * |
| Medium (18) | TSG | 86.2 ± 5.5 | 102.4% | 5.7 (4%) | +4.45 [+1.48, +7.42], p = 0.0031 * |
| Large (24) | Full Graph | 99.2 ± 3.4 | 100.0% | 276.0 (100%) | +5.15 [+2.32, +7.98], p = 0.0028 * |
| Large (24) | D-MORPH | 104.3 ± 6.4 | 105.2% | 23.5 (9%) |  |
| Large (24) | MORPH (centralised) | 101.3 ± 6.5 | 102.2% | 28.7 (10%) | +3.00 [-1.54, +7.54], p = 0.0823 |
| Large (24) | Proximity-r | 104.2 ± 7.5 | 105.1% | 30.8 (11%) | +0.05 [-4.97, +5.07], p = 0.4682 |
| Large (24) | Random-Sparse (fixed) | 24.4 ± 12.6 | 24.7% | 24.0 (9%) | +79.85 [+71.99, +87.71], p = 0.0001 * |
| Large (24) | Random-Sparse (per-step) | 70.8 ± 14.4 | 71.4% | 24.0 (9%) | +33.55 [+25.85, +41.25], p = 0.0001 * |
| Large (24) | TSG | 100.3 ± 3.7 | 101.2% | 7.9 (3%) | +3.95 [+0.88, +7.02], p = 0.0031 * |

\* p < 0.05.

Partner chosen differs from the nearest available robot (share of requests): tiny 1.20%, small 0.44%, medium 0.34%, large 0.32% (D-MORPH).

### Diagnostics (Table 4, remaining columns; mean ± SD over seeds)

Refused: share of partner requests refused; idle: share of steps without a task; task time: steps from assignment to delivery and to return; queue wait: steps from a request entering the queue to its assignment (requests assigned within the episode only); clashes and stucks: the benchmark counters per episode.

| Scale | Method | Refused (%) | Idle AGV (%) | Idle Picker (%) | To delivery | To return | Queue wait | Clashes | Stucks |
|---|---|---|---|---|---|---|---|---|---|
| Tiny | Full Graph | 0 ± 0 | 1.3 ± 0.2 | 6.0 ± 1.5 | 42.7 ± 2.2 | 67.8 ± 3.0 | 21.7 ± 1.9 | 88.2 ± 80.2 | 5.7 ± 12.3 |
| Tiny | D-MORPH | 72 ± 3 | 2.1 ± 0.2 | 9.6 ± 1.3 | 31.0 ± 1.0 | 67.5 ± 3.6 | 94.5 ± 11.8 | 85.7 ± 89.7 | 4.8 ± 12.8 |
| Tiny | MORPH (centralised) | 70 ± 3 | 1.3 ± 0.2 | 11.0 ± 2.4 | 41.0 ± 4.3 | 70.9 ± 8.1 | 25.5 ± 3.1 | 141.5 ± 177.8 | 10.4 ± 22.1 |
| Tiny | Proximity-r | 53 ± 5 | 1.3 ± 0.3 | 7.9 ± 2.0 | 39.1 ± 5.3 | 68.2 ± 8.3 | 24.5 ± 3.2 | 158.8 ± 245.5 | 13.1 ± 27.6 |
| Tiny | Random-Sparse (fixed) | 40 ± 23 | 72.9 ± 22.1 | 77.5 ± 20.1 | 26.5 ± 4.1 | 44.4 ± 12.4 | 94.5 ± 60.7 | 174.7 ± 393.8 | 12.2 ± 29.3 |
| Tiny | Random-Sparse (per-step) | 40 ± 6 | 16.9 ± 7.4 | 13.7 ± 5.4 | 33.3 ± 2.9 | 77.9 ± 9.1 | 90.0 ± 27.9 | 121.8 ± 204.4 | 13.3 ± 27.4 |
| Tiny | TSG | 75 ± 3 | 1.3 ± 0.2 | 12.6 ± 2.5 | 41.4 ± 3.9 | 71.5 ± 10.3 | 25.7 ± 5.6 | 122.0 ± 175.2 | 8.8 ± 23.9 |
| Small | Full Graph | 0 ± 0 | 1.0 ± 0.1 | 4.1 ± 0.5 | 64.7 ± 7.1 | 95.8 ± 8.3 | 25.4 ± 2.4 | 104.3 ± 21.8 | 3.8 ± 2.5 |
| Small | D-MORPH | 72 ± 3 | 2.1 ± 0.2 | 5.7 ± 0.9 | 40.5 ± 1.2 | 93.6 ± 7.7 | 86.6 ± 13.8 | 193.4 ± 269.4 | 10.4 ± 24.2 |
| Small | MORPH (centralised) | 72 ± 2 | 0.9 ± 0.1 | 7.1 ± 0.9 | 57.7 ± 3.5 | 96.3 ± 4.6 | 25.3 ± 4.4 | 159.1 ± 244.3 | 6.4 ± 15.4 |
| Small | Proximity-r | 61 ± 3 | 1.0 ± 0.1 | 5.9 ± 1.0 | 54.8 ± 3.6 | 91.5 ± 5.0 | 28.4 ± 4.5 | 198.2 ± 257.8 | 9.7 ± 17.3 |
| Small | Random-Sparse (fixed) | 41 ± 20 | 72.1 ± 14.4 | 69.7 ± 19.6 | 34.0 ± 7.0 | 59.5 ± 15.4 | 137.2 ± 69.6 | 228.9 ± 454.1 | 20.6 ± 42.6 |
| Small | Random-Sparse (per-step) | 42 ± 8 | 21.8 ± 9.9 | 8.9 ± 3.9 | 44.8 ± 4.4 | 110.4 ± 9.7 | 85.4 ± 31.4 | 47.9 ± 22.6 | 5.8 ± 9.7 |
| Small | TSG | 74 ± 2 | 1.0 ± 0.1 | 7.8 ± 0.9 | 57.4 ± 2.8 | 94.8 ± 2.9 | 27.5 ± 4.7 | 148.6 ± 165.1 | 5.6 ± 10.5 |
| Medium | Full Graph | 0 ± 0 | 0.8 ± 0.1 | 3.5 ± 0.3 | 73.6 ± 4.1 | 107.6 ± 4.5 | 25.9 ± 3.1 | 187.2 ± 133.1 | 8.9 ± 19.2 |
| Medium | D-MORPH | 72 ± 2 | 2.2 ± 0.2 | 4.6 ± 0.2 | 44.4 ± 1.1 | 104.3 ± 2.6 | 69.2 ± 11.4 | 142.9 ± 12.5 | 1.4 ± 1.5 |
| Medium | MORPH (centralised) | 74 ± 2 | 0.9 ± 0.0 | 5.5 ± 0.3 | 64.5 ± 1.8 | 105.1 ± 2.5 | 25.8 ± 3.6 | 138.2 ± 13.8 | 3.4 ± 1.7 |
| Medium | Proximity-r | 64 ± 2 | 0.8 ± 0.1 | 4.4 ± 0.8 | 64.3 ± 3.9 | 104.6 ± 7.0 | 26.3 ± 4.7 | 234.8 ± 304.9 | 14.8 ± 26.6 |
| Medium | Random-Sparse (fixed) | 45 ± 21 | 82.3 ± 12.6 | 82.0 ± 15.5 | 37.1 ± 6.5 | 61.5 ± 13.7 | 63.9 ± 54.9 | 281.1 ± 704.6 | 23.4 ± 50.0 |
| Medium | Random-Sparse (per-step) | 42 ± 4 | 19.9 ± 6.1 | 8.8 ± 3.3 | 49.8 ± 4.9 | 128.4 ± 21.7 | 94.2 ± 36.9 | 112.5 ± 210.6 | 9.7 ± 14.2 |
| Medium | TSG | 75 ± 2 | 0.8 ± 0.1 | 5.6 ± 0.5 | 65.4 ± 2.8 | 106.8 ± 3.9 | 25.6 ± 4.5 | 236.3 ± 291.7 | 8.6 ± 18.5 |
| Large | Full Graph | 0 ± 0 | 0.7 ± 0.0 | 3.1 ± 0.1 | 84.1 ± 3.0 | 123.7 ± 3.7 | 26.9 ± 4.2 | 213.6 ± 22.3 | 14.1 ± 3.0 |
| Large | D-MORPH | 73 ± 2 | 2.1 ± 0.3 | 3.8 ± 0.4 | 53.6 ± 2.0 | 120.2 ± 5.2 | 65.3 ± 7.0 | 259.8 ± 168.0 | 13.3 ± 23.9 |
| Large | MORPH (centralised) | 75 ± 1 | 0.7 ± 0.0 | 4.6 ± 0.4 | 75.7 ± 4.9 | 122.0 ± 5.8 | 25.9 ± 4.2 | 206.2 ± 30.9 | 9.7 ± 6.7 |
| Large | Proximity-r | 68 ± 2 | 0.8 ± 0.1 | 4.2 ± 0.5 | 72.2 ± 3.6 | 116.5 ± 4.3 | 27.0 ± 4.5 | 260.4 ± 148.6 | 19.2 ± 22.6 |
| Large | Random-Sparse (fixed) | 46 ± 11 | 76.4 ± 11.1 | 75.6 ± 12.8 | 42.4 ± 5.9 | 72.9 ± 11.4 | 150.8 ± 68.4 | 243.9 ± 539.9 | 20.6 ± 41.9 |
| Large | Random-Sparse (per-step) | 41 ± 4 | 15.2 ± 3.5 | 10.8 ± 2.5 | 59.4 ± 3.3 | 144.5 ± 9.2 | 113.8 ± 25.1 | 161.8 ± 142.2 | 16.1 ± 20.7 |
| Large | TSG | 75 ± 1 | 0.7 ± 0.0 | 4.7 ± 0.4 | 76.5 ± 2.8 | 122.5 ± 3.6 | 26.9 ± 4.4 | 206.4 ± 24.9 | 9.8 ± 5.3 |

Random-Sparse failed requests per episode (no link / same-role partner / busy partner) and stalled episodes (no delivery in the last 200 steps):

| Scale | Variant | No link | Same role | Busy | Stalled |
|---|---|---|---|---|---|
| Tiny | Random-Sparse (fixed) | 2835 | 9178 | 3201 | 12/20 |
| Tiny | Random-Sparse (per-step) | 490 | 1461 | 934 | 6/20 |
| Small | Random-Sparse (fixed) | 1720 | 11244 | 2455 | 10/20 |
| Small | Random-Sparse (per-step) | 308 | 1803 | 974 | 11/20 |
| Medium | Random-Sparse (fixed) | 1443 | 12475 | 1644 | 17/20 |
| Medium | Random-Sparse (per-step) | 576 | 2796 | 1645 | 6/20 |
| Large | Random-Sparse (fixed) | 4530 | 23781 | 2925 | 6/20 |
| Large | Random-Sparse (per-step) | 1848 | 8224 | 5068 | 1/20 |

## Proximity-r sweep (Figure 4)

Highest-throughput radius per scale: tiny r = 11 (57.0 deliveries, 23.7 links), small r = 10 (69.3 deliveries, 36.3 links), medium r = 10 (94.8 deliveries, 60.9 links), large r = 12 (108.9 deliveries, 105.4 links).

## Rule ablations (Figure 5; deliveries in docs/tables/table_ablation_deliveries.tex)

Full D-MORPH against D-MORPH with one rule removed, same seeds. Each ablation has one primary measure fixed before the runs (docs/ablation_hypotheses.md): No completion credit (R1), No homeostasis (R2), No predictive formation (R3) and Frozen structure (R3, no link formation or pruning after step 400) are pre-specified; No R4 (BCM decay and neuromodulation both off) was added afterwards; the two R4 component tests (No BCM threshold, No neuromodulation) are pre-specified but are not reported in the manuscript. Setting: scale study (T = 800) or location shift (T = 2400).

| Ablation | Setting | Scale | Primary measure | Full | Ablation | Full minus ablation, p | Deliveries full / ablation (p) |
|---|---|---|---|---|---|---|---|
| No completion credit (R1) | scale | Tiny | preference of delivering pairs minus co-assigned-only pairs | 0.02293 | -0.03049 | +0.0534, p < 0.0001 * | 56.8 / 56.1 (p = 0.570) |
| No completion credit (R1) | scale | Small | preference of delivering pairs minus co-assigned-only pairs | 0.007879 | -0.04403 | +0.0519, p < 0.0001 * | 66.0 / 66.9 (p = 0.824) |
| No completion credit (R1) | scale | Medium | preference of delivering pairs minus co-assigned-only pairs | 0.004421 | -0.04763 | +0.0521, p < 0.0001 * | 90.7 / 90.2 (p = 0.457) |
| No completion credit (R1) | scale | Large | preference of delivering pairs minus co-assigned-only pairs | -0.001726 | -0.05258 | +0.0509, p < 0.0001 * | 104.3 / 105.3 (p = 0.727) |
| No homeostasis (R2) | scale | Tiny | mean |links - target| per robot | 0.6 | 0.7039 | -0.104, p = 0.0094 * | 56.8 / 54.1 (p = 0.311) |
| No homeostasis (R2) | scale | Small | mean |links - target| per robot | 0.5988 | 0.675 | -0.0762, p < 0.0001 * | 66.0 / 66.3 (p = 0.511) |
| No homeostasis (R2) | scale | Medium | mean |links - target| per robot | 1.097 | 1.208 | -0.111, p < 0.0001 * | 90.7 / 90.8 (p = 0.461) |
| No homeostasis (R2) | scale | Large | mean |links - target| per robot | 1.945 | 2.118 | -0.173, p < 0.0001 * | 104.3 / 104.3 (p = 1.000) |
| No predictive formation (R3) | scale | Tiny | active directed links at step 50 | 12.9 | 12.8 | +0.1, p = 0.3173 | 56.8 / 56.7 (p = 0.684) |
| No predictive formation (R3) | scale | Small | active directed links at step 50 | 22.05 | 20.75 | +1.3, p = 0.0008 * | 66.0 / 65.7 (p = 0.180) |
| No predictive formation (R3) | scale | Medium | active directed links at step 50 | 35.05 | 30.2 | +4.85, p < 0.0001 * | 90.7 / 90.5 (p = 0.344) |
| No predictive formation (R3) | scale | Large | active directed links at step 50 | 46.75 | 39.05 | +7.7, p < 0.0001 * | 104.3 / 104.0 (p = 0.720) |
| No R4 | scale | Tiny | mean active links (additional ablation) | 6.566 | 6.615 | -0.0493, p = 0.1769 | 56.8 / 54.9 (p = 0.213) |
| No R4 | scale | Small | mean active links (additional ablation) | 11.35 | 11.56 | -0.214, p = 0.1327 | 66.0 / 66.5 (p = 0.897) |
| No R4 | scale | Medium | mean active links (additional ablation) | 17.63 | 17.98 | -0.348, p = 0.0136 * | 90.7 / 90.0 (p = 0.292) |
| No R4 | scale | Large | mean active links (additional ablation) | 23.51 | 24.46 | -0.95, p < 0.0001 * | 104.3 / 105.5 (p = 0.815) |
| Frozen structure (R3) | shift | Medium | co-assigned pairs linked after the first shift (%) | 100 | 33.05 | +66.9, p < 0.0001 * | 257.6 / 244.2 (p = 0.022) |
| Frozen structure (R3) | shift | Large | co-assigned pairs linked after the first shift (%) | 100 | 26.09 | +73.9, p < 0.0001 * | 297.4 / 301.6 (p = 0.881) |
| No BCM threshold (R4a) | shift | Medium | links active 25 steps before a shift still active 50 steps after it (%); not in the manuscript | 44.43 | 46.05 | -1.61, p = 0.1893 | 257.6 / 258.9 (p = 0.887) |
| No BCM threshold (R4a) | shift | Large | links active 25 steps before a shift still active 50 steps after it (%); not in the manuscript | 42.62 | 45.35 | -2.73, p = 0.2455 | 297.4 / 294.5 (p = 0.668) |
| No neuromodulation (R4b) | shift | Medium | mean recovery time (steps); not in the manuscript | 21.24 | 12.95 | +8.29, p = 0.0741 | 257.6 / 250.0 (p = 0.344) |
| No neuromodulation (R4b) | shift | Large | mean recovery time (steps); not in the manuscript | 18.3 | 19.99 | -1.69, p = 0.2310 | 297.4 / 294.2 (p = 0.740) |

## Package location shift (Table 5, Figure 6)

Demand alternates between the upper and lower half every 400 steps (share of new requests in the active half: 1.000). Δ̄ = ((P5 − P3) + (P6 − P4)) / 2 (negative: fewer deliveries on the later visit to the same region); retention: first 100 steps after a shift relative to the last 100 of the previous same-half phase; r̄: mean recovery time (MORPH's definition). D-MORPH minus each baseline, paired by seed.

| Scale | Method | P1–P6 | Δ̄ | Retention | r̄ (steps) | Stalled | Δ̄: D-MORPH minus method [95% CI], p | r̄: D-MORPH minus method, p |
|---|---|---|---|---|---|---|---|---|
| Medium | D-MORPH | 45.7 44.5 41.5 42.4 40.6 42.9 | -0.12 | 0.970 | 21.2 | 0/20 |  |  |
| Medium | Full Graph | 43.2 43.2 41.1 42.1 39.5 37.2 | -3.27 | 0.962 | 25.4 | 0/20 | +3.15 [+0.58, +5.72], p = 0.0327 * | -4.2, p = 0.8960 |
| Medium | MORPH (centralised) | 42.8 41.5 37.4 38.0 33.8 30.2 | -5.62 | 1.016 | 47.1 | 0/20 | +5.50 [+2.19, +8.81], p = 0.0072 * | -25.8, p = 0.0153 * |
| Medium | Proximity-r | 44.5 45.5 41.0 37.6 34.8 32.5 | -5.70 | 0.877 | 65.1 | 2/20 | +5.58 [+1.89, +9.26], p = 0.0070 * | -43.9, p = 0.2611 |
| Medium | Random-Sparse (fixed) | 9.4 5.0 0.8 0.0 0.0 0.0 | -0.38 | 0.037 | 380.9 | 20/20 | +0.25 [-0.72, +1.22], p = 0.6337 | -359.7, p < 0.0001 * |
| Medium | Random-Sparse (per-step) | 32.5 22.7 19.1 16.6 14.8 14.6 | -3.20 | 0.608 | 167.2 | 11/20 | +3.08 [-1.04, +7.19], p = 0.3835 | -145.9, p = 0.0023 * |
| Medium | TSG | 42.7 42.6 38.6 38.0 35.0 34.4 | -3.70 | 0.958 | 28.7 | 1/20 | +3.58 [+0.58, +6.57], p = 0.0206 * | -7.5, p = 0.4440 |
| Large | D-MORPH | 52.3 52.0 48.5 49.5 47.0 48.1 | -1.40 | 0.958 | 18.3 | 0/20 |  |  |
| Large | Full Graph | 49.0 50.8 44.0 45.9 41.2 42.9 | -2.88 | 0.913 | 61.0 | 2/20 | +1.48 [-1.18, +4.13], p = 0.3487 | -42.7, p = 0.1590 |
| Large | MORPH (centralised) | 51.5 53.1 48.6 49.4 46.2 48.4 | -1.65 | 0.989 | 14.2 | 0/20 | +0.25 [-2.00, +2.50], p = 0.9518 | +4.2, p = 0.7841 |
| Large | Proximity-r | 52.7 53.5 48.8 49.0 44.1 44.9 | -4.38 | 0.938 | 35.6 | 0/20 | +2.98 [-0.62, +6.57], p = 0.0591 | -17.3, p = 0.0484 * |
| Large | Random-Sparse (fixed) | 13.5 8.4 4.5 3.4 2.5 1.6 | -1.88 | 0.350 | 302.7 | 16/20 | +0.47 [-2.27, +3.22], p = 0.8404 | -284.4, p = 0.0001 * |
| Large | Random-Sparse (per-step) | 38.3 28.9 22.4 22.6 20.6 17.4 | -3.50 | 0.761 | 145.7 | 9/20 | +2.10 [-2.08, +6.28], p = 0.3545 | -127.4, p = 0.0034 * |
| Large | TSG | 50.5 51.0 46.0 46.6 44.0 44.0 | -2.27 | 0.948 | 33.1 | 0/20 | +0.88 [-2.42, +4.17], p = 0.5194 | -14.8, p = 0.0534 |

\* p < 0.05.

## Emergent allocation structure (Figure 7)

Medium, seed 42, T = 1500: 166 deliveries. After step 200, 16.9 AGV–Picker, 0.7 AGV–AGV and 0.0 Picker–Picker links on average; mean preference on links 0.059. 655 links pruned (median lifetime 21 steps; 50% pruned at 21–22 steps, 35% lived longer than 100 steps); 43 active at the end. Partner differs from the nearest robot in 0.5% of requests.
