Dependence of Battle Droid Formations on a Control Link: An Empirical Study of Captured and Republic Engagement Records, 32–21 BBY

1. Introduction

At the Battle of Naboo in 32 BBY the Trade Federation's droid army stopped functioning when the Droid Control Ship that served it was destroyed. The episode is among the best attested in the primary record, and it has supported a widely repeated assessment of the droid army as an instrument with a single point of failure. A military-historical assessment by one of us accepts the Naboo evidence but treats it as the outcome of a single-vessel deployment and cautions against generalising from it (Dreth, 23 BBY). Engineering case studies of remotely commanded ground units describe abrupt failure on loss of command (Tavrenne, 25 BBY), but they do not address droid armies in the field. Anyone who has read only the Naboo material would expect a droid formation deprived of its control ship to cease acting altogether.

The record of the Clone Wars, 22 BBY onward, does not sit easily beside that expectation. Separatist droid forces are attested fighting extended ground campaigns across many worlds, and the accounts of those campaigns do not, for the most part, place a control ship in the engagement. Two bodies of primary evidence therefore appear to disagree. One shows a total, immediate loss of function on severing the control link; the other shows sustained, apparently self-directed operation with no such link in evidence. We treat this as a problem of the record and not as a puzzle to be settled by preference for the more famous source. One of us has argued, before the war, that Separatist-aligned droid armies would be reorganised away from central command after Naboo (Calloran, 26 BBY). That argument and the caution about Naboo are both plausible, and neither has been tested against engagement-level data.

Our question is therefore a narrow one. How does the coordination of a droid formation behave in the engagements where a control link is present, where it is lost, and where none appears in the record at all? The comparison of the third condition with the first two is the point of the study, since it is the only condition in which the two bodies of evidence can be set against each other directly.

The vantage of this paper is 20 BBY. The war is in progress, the captured and Republic records are incomplete, and the material examined runs from 32 BBY to 21 BBY. Conclusions are offered as provisional for that reason.

2. Methods

Sources. We drew on two archives. The first is a set of Trade Federation droid operations logs seized by Republic forces, held in the captured series (Trade Federation Naboo Occupation Command, 32 BBY; Trade Federation Droid Operations Staff, 21 BBY). The second is the after-action material of the Grand Army of the Republic for ground engagements between 22 and 21 BBY (Grand Army of the Republic, 21 BBY). The captured logs contributed 31 engagements, the Republic summaries 22, and 11 engagements were documented in both, giving 42 distinct engagements. One of these is the Naboo engagement of 32 BBY. The 11 Republic-only engagements have no captured log, so their link status could not be documented and they were coded absent by construction; they supply 22 of the 40 absent records, and the other 18 come from 8 engagements with a captured log. Where the two archives described the same engagement differently, we coded from the captured log for the state of the control link and from the Republic summary for the observed behaviour of the formation, and we recorded the discrepancy, following the source-critical cautions set out for seized machine records (Halkeer, 24 BBY).

Unit of analysis. The unit is the formation-engagement record: one formation, as listed in an order of battle, in one engagement. The 42 engagements yielded 96 such records. Formations were coded into three types, the tactical category following an earlier survey of the Separatist order of battle (Ossuli, 22 BBY): line infantry formations, formations led by tactical or command droids, and heavy assets, meaning armoured and walking platforms. Cross-classification used the type stated in the order of battle. The coding categories throughout this paper, and the thresholds defined below, are our own constructs. They were not used by the archives and are not terms of the record.

Control link. Link status was taken from the captured logs where one existed. A record was coded present if the log documented control-link acknowledgements or a control vessel in the order of battle for the whole engagement, severed if the log documented the loss of that link or vessel during the engagement, and absent if neither a link entry nor a control vessel appeared at any point. Absent does not mean that no link existed. It means that the record supplies no evidence of one, and the whole design depends on keeping those two readings apart.

Outcome. Coordination was coded as maintained when no loss of coordinated movement or fire was documented, degraded when a loss lasting more than ten standard minutes was documented, either by an entry stating its duration or by a loss recorded across at least two consecutive status entries, while the formation continued to act, and collapsed when coordinated activity ceased across the formation for the remainder of the engagement. The ten-minute figure is an operational choice of ours, made because shorter interruptions are indistinguishable from ordinary logging gaps. Status entries were not made at a uniform interval, so durations are approximate. Two coders scored a random sample of 30 records; they agreed on 26 (86.7%), with Cohen's κ = 0.77 (approximate 95% CI 0.55–0.98). The coders worked independently of each other, but neither was blind to link status when scoring outcome. Disagreements were resolved by discussion and the remaining records were coded by the first author.

Analysis. The outcome is categorical, so we report counts and proportions with Wilson 95% confidence intervals. The association between link status and the three-level outcome was tested by Pearson's χ² with a Monte Carlo permutation p-value, because four of the nine expected counts fell below five. Risk ratios for disruption (degraded or collapsed) are given with log-scale 95% intervals and Fisher exact p-values. To adjust for formation type we fitted a logistic regression of disruption on link status (reference: present) and formation type (reference: line infantry), and assessed each factor by likelihood-ratio test. As a check on source coverage, absent records with a captured log were also compared with present records. Time-to-event methods were considered and set aside, since the logs give the state of a formation at intervals and not the moment of each loss. Records from one engagement are not independent, and the analysis treats them as though they were; we return to this in Section 5.

3. Results

Of the 96 records, 40 were coded link present, 16 severed and 40 absent. Formation types numbered 50 line infantry, 28 tactical or command-led and 18 heavy assets. Coordination was maintained in 66 records, degraded in 19 and collapsed in 11. The distribution of outcomes differed sharply by link status (Table 1; χ²(4) = 45.0, permutation p < .001).

Severance was the condition in which coordination failed. Of 16 severed records, 14 (87.5%; 95% CI 64.0–96.5) showed disruption, against 6 of 40 (15.0%; 7.1–29.1) where the link was present, a risk ratio of 5.8 (95% CI 2.7–12.5; Fisher p < .001). Collapse was the modal outcome after severance, in 9 of 16 records (56.3%; 33.2–76.9), and occurred in only 2 of the remaining 80 (2.5%), a risk ratio of 22.5 (5.4–94.5). The interval for that ratio is wide, as the count of collapses outside the severed group is very small.

The severed group is not homogeneous. Four of its 16 records come from Naboo, and all four collapsed, matching the primary account. The other 12 severed records, from five wartime engagements, gave a milder picture: 10 disrupted (83.3%; 55.2–95.3) but only 5 collapsed (41.7%; 19.3–68.0), with the remaining 5 degraded and 2 maintained. Even without Naboo, disruption after severance remained far above that with the link present (risk ratio 5.6; 95% CI 2.5–12.1; p < .001).

Absence of a link entry did not behave like severance. Ten of 40 absent records (25.0%; 14.2–40.2) showed disruption, and only one collapsed (2.5%; 0.4–12.9), the same proportion as in the present group. The risk ratio for disruption, absent against present, was 1.7 (0.67–4.15; p = .40). The data therefore neither show that formations without a documented link were as steady as linked ones, nor that they were worse; the estimate leans towards somewhat more disruption, with an interval that includes no difference at all. Restricting the absent group to the 18 records with a captured log, disruption occurred in 4 (22.2%; 9.0–45.2), against 6 of 40 present (risk ratio 1.5; 0.48–4.62; Fisher p = .48); among the 22 Republic-only records it occurred in 6 (27.3%; 13.2–48.2).

Formation type explained little (Table 2). In the adjusted model, severed status carried an odds ratio for disruption of 50.0 (95% CI 8.1–308.6; p < .001) relative to present, and absent status an odds ratio of 1.90 (0.61–5.98; p = .27). Command-led formations had an odds ratio of 0.34 (0.08–1.41; p = .14) and heavy assets 1.35 (0.37–4.99; p = .65), both against line infantry. The likelihood-ratio test was strong for link status, χ²(2) = 29.8, p < .001, and weak for formation type, χ²(2) = 3.4, p = .18. The odds ratio for severance is imprecise, since the severed cells contain almost no undisrupted records, and should be read as indicating a very large effect and not as a measured magnitude.

One descriptive pattern deserves mention but not a test, because the cells are small. Among absent-link records, disruption occurred in 1 of 11 command-led formations (9.1%) and in 7 of 22 line infantry formations (31.8%).

4. Discussion

The analysis supports two statements, and they answer the question more precisely than either body of primary evidence does alone. Where a control link existed and was lost, formations very often lost coordination, and just over half of the severed records show a complete collapse. Where no link is documented, formations were not obviously different from linked ones. The two bodies of evidence conflict only if dependence on a control link is assumed to be a fixed property of every droid formation. Our data are compatible with a narrower reading, in which dependence is a property of formations that were actually operating under a link.

Two hypotheses could account for the absent-link records, and this study cannot choose between them. The first is that those formations operated under a link the records do not show, for instance one carried by equipment not mentioned in orders of battle, in which case their steadiness reflects an intact link. The second is that they operated autonomously or under local direction, and that formations of the wartime armies differed in organisation from the Naboo invasion force. Disruption was low among command-led formations lacking a link, 1 of 11, but it was equally low among command-led formations with the link present, 1 of 12 (8.3%), so the pattern does not discriminate between the hypotheses, and the adjusted estimate for command-led formations crosses 1. Some accounts of later campaigns place a Separatist flagship or central facility among the objectives of Republic operations, and the records examined here do not establish whether the forces involved depended on such a vessel or facility. Neither hypothesis is a claim about how droids are built. Nothing in these records establishes the internal architecture of a droid or its command equipment, and we make no such claim.

The wartime severed records deserve separate attention. Their collapse rate, 5 of 12, was lower than that at Naboo, where all four records collapsed, though the Naboo figure rests on a single engagement. If wartime formations were more resilient to severance than the invasion force, that would fit the pre-war argument for reorganisation away from central command (Calloran, 26 BBY); if it merely reflects differences in how the two archives record degraded coordination, it would not. The current data do not separate the two. Nor can they tell us whether the Naboo collapse was typical of pre-war practice, as only one pre-war engagement survives in the archives.

Republic planners might be tempted to draw an operational conclusion, namely that severing a link is worth attempting. The evidence supports that conclusion only where a link is known to be present. Attacking a link that a formation does not use would gain nothing, and the absent group accounts for 40 of the 92 wartime records, but the records cannot show which absent-link formations, if any, fell in that class. We would counsel that link status be established before any action is planned around it, and that captured logs be examined for how it is recorded.

5. Limitations

The sample is small and shaped by capture. Records enter the captured series only when Republic forces seized a log, and the Republic summaries cover only engagements the Grand Army chose to document; neither is a random draw from all engagements. The 96 records come from 42 engagements, and formations within one engagement share terrain, opposition and command decisions. Treating them as independent makes the confidence intervals and p-values optimistic, and the true uncertainty is larger than reported. We did not attempt a cluster-adjusted analysis, since 5 severed wartime engagements cannot support one.

Our coding scheme is a construct of the authors. Link status was inferred from what logs happened to mention, and the absent category is a residue that mixes true autonomy with undocumented links. The absent category also partly reflects source coverage, since 22 of its 40 records come from engagements documented only by Republic summaries. The ten-minute threshold for degradation was chosen on practical grounds and was not varied, and irregular log intervals make measured durations approximate, so we cannot say how sensitive the degraded category is to it. Inter-coder agreement was good but not high, and the interval for κ is broad because only 30 records were double coded. Pre-war coverage rests on one engagement.

Finally, the outcome is observational. The logs record what a formation did after a link was lost; they do not show why. We cannot exclude that severance coincided with other events, such as the destruction of a formation's commander, that would disrupt coordination independently. Because the war is ongoing at the vantage of this paper, further captured logs are likely to alter these estimates, and the present figures should be treated as a first reading of an incomplete record.