Uncited Press › Star Trek Open the interactive journal →
Trek · Military Science & Strategy

Why a Cube Survives: Structural Redundancy and the Convergence of Load Paths in Borg Cubes, from Tactical Scans and Damage Records, 2365–2373

Dr. Dorrak1
1 Utopia Planitia Fleet Yards, Structural Engineering Office
Received 18 Aug 2026 · Revised 2 Oct 2026 · Accepted 24 Oct 2026

Abstract

Federation tactical doctrine holds that a Borg cube, having no bridge, engineering section or other critical system, has no place at which damage is decisive. We show that this is a belief about control which has been applied wrongly to load. Using passive structural scans and fleet damage logs from the engagements between 2365 and 2373, we divide a cube into 216 cells and model its power and structural load as a flow network, with the ratio of primary-conduit to mesh capacity fitted from the earlier scans (800, 95% CI 617 to 1037). The model ranks each cell by the share of demand lost when it is removed. Of 268 recorded damage events, 251 (93.7%) were regenerated within one volley interval, yet regeneration failed in 3.0% of events on cells of negligible consequence, in 20.0% on cells of intermediate consequence and in 4 of 5 on cells of high consequence. The junction on which fire was concentrated in 2373 ranks first of 216 cells, with a loss of 15.5% of delivered power, although it ranks 206.5 by emission and 20.5 by closeness to the cube's centre. A cube survives most fire because most of its structure carries little load, and the junctions that carry the load can be located from the passive scans of an engagement's first exchanges.

1. Introduction

A common misconception holds that a structure without a command centre has no place at which damage is decisive. Federation gunnery doctrine has repeated this belief since the first encounters (Corvane, 2369), and the engagement record is offered in support, because the cubes repaired 251 of 268 recorded damage events before the fleet's next volley arrived. The belief is wrong. This paper locates the point at which it fails and explains why the failure stayed hidden in the very records that were taken to confirm it.

We write in 2374 and use only records closed by the end of 2373. They are the passive structural scans and fleet damage logs of the engagements with Borg cubes, which fall into three groups, held by the Starfleet Tactical Engagement Registry (Starfleet Tactical Analysis Division, 2374). The first is the encounter of 2365. The second comprises the engagement of 2366 and, in 2367, the engagement at Wolf 359 and the approach to Earth, for which the surviving logs are few. The third is the engagement in Sector 001 in 2373. The cube destroyed in 2367 was lost after a command entered the Collective through the link of its captured officer and placed the drones in regeneration, which tested control and not structure, and we exclude that loss from every count below.

The doctrine confuses control with load. It is correct that a cube lacks a bridge, an engineering section and living quarters in the Federation sense. However, a vessel without a place where decisions are taken is not a vessel without places where power and stress must pass. Power travels from generation to use, structural load travels from its source to where the frame bears it, and every such path has junctions. A junction that cannot be bypassed is a critical point whatever intelligence directs the vessel, and the doctrine has never asked whether a cube has one.

Exhaustion fire, which assumes that damage will in time outrun repair, follows from the doctrine (Tavenor, 2371). The engagement of 2373 contradicts it, and the literature has treated the contradiction as an exception. The fleet fired for a long period at sections that regenerated within the volley interval. It then fired at a coordinate with no tactical signature, and the cube was destroyed. The accounts credit the judgment of the officer who gave the order. We do not examine how the order was reached, and show instead that the scans taken before the change of target already single out the coordinate, so that the coordinate could have been chosen from the record alone, without any knowledge of the Collective from inside it.

The corrected claim is narrow, and it can be tested. A cube has no point at which damage to control is decisive, but it has a junction at which damage to load is decisive, and that junction can be found in the scans before it is fired on.

2. System Description

The scans resolve each cube into a lattice of 216 cells, six along each edge, so that a cell is one sixth of a cube edge in every direction. This is the finest division that passive sensors support at tactical range (Marrowby, 2367). The registry localises each impact to a single cell, with an uncertainty of one cell which Section 5 examines. A damage event, our unit of analysis, is the set of impacts in one cell during one scan frame, and the registry holds 268 of them: 9 from 2365, 221 from 2373 and 38 from the engagements between them (Table 2).

Two structures are visible in every scan. The first is a fine mesh of secondary conduits and framing, joining each cell to its six neighbours, that carries little. The second is a set of primary conduits and structural spines, with emission strengths far above the mesh, which run along the axes of the cube. In the frames of 2373 recorded before the change of target, six primary arms meet at one junction, in cell (3, 2, 4) counted from a corner of the cube. The junction lies 1.7 cell edges, or 0.28 of a cube edge, from the geometric centre. It is a passive crossing that emits no more than its neighbours. The scans place each primary spine in the corridor of a primary conduit, and one network therefore describes both power and structural load.

Generation is distributed, so every cell generates and consumes power, and total generation exceeds total demand by 5 percent. The surplus is uneven, and we divide the cube into six sectors by the face nearest each cell, resolving ties on the diagonals in the order x, y, z, which gives sectors of 56, 56, 32, 32, 20 and 20 cells. Local generation, set from the emission levels of the approach scan, ranges from 0.29 to 1.89 of a cell's own demand across the sectors. Three sectors export power and three import it, and the importers together lack 42.6 units, a unit being the demand of one cell. Each sector delivers and receives through a collection point on its primary arm, which is joined to the cells of the sector by feeders of capacity 1.5 units. The mesh also crosses the sector boundaries. Its links, however, are narrow.

Because the mechanism of regeneration is not visible to passive scans, we do not model it and record only the outcome. Following Hallister (2373), a cell counts as regenerated when its scan signature returns to the value before the hit within one volley interval of the attacking force, which has a median of 41 seconds in the registry.

3. Analysis / Model

We represent the network as a flow problem, in the manner that Brennick (2371) used for segmented space frames. The nodes are the 216 cells. Each cell is a source of its local generation and a sink of one unit of demand. Mesh links join neighbouring cells with capacity m, feeders join each collection point to the cells of its sector, and each primary arm is a chain of cells along its corridor, from the collection point to the junction, whose links have capacity T, a division of capacity between primary and secondary conduits that follows the practice for distributed-generation hulls (Ardell, 2370). Delivered power F is the maximum flow from sources to sinks, computed by an augmenting-path algorithm. In the intact network F equals 216, and every demand is met.

The convergence index of a cell is I = 1 − F′/F, where F′ is the maximum delivered power after that cell is removed and the flow is rerouted along the best remaining paths. It is the share of the total demand of the cube that can no longer be met when the cell is lost, including the demand of the cell itself. The index measures consequence and carries no information about emission or regeneration.

One parameter governs the result, namely the ratio κ of the capacity of a primary arm to that of one mesh link. We estimated it from 14 paired readings of primary-conduit and mesh emission in the scans of 2365 to 2367, which gave a geometric mean of 800 (95% CI 617 to 1037, from a t interval with 13 degrees of freedom on the logarithmic scale). We set T = 40 and m = 0.05 units, and we vary κ through m with T held at 40. The layout, the position of the junction and the sector generation come from the 2373 frames recorded before the change of target. Therefore nothing from the fatal sequence enters the model.

Cells are sorted by the model into three tiers according to index. Tier A (index below 0.01) holds 200 cells, each with an index of 0.005, which is the loss of the cell's own demand and nothing more. Tier B (0.01 to 0.05) holds 11 cells, with indices of 0.018 to 0.049. Tier C (0.05 or above) holds 5 cells, namely the junction and four cells on the primary arms, with indices of 0.052 to 0.155. Tiers B and C together are the 16 cells of the primary arms, the junction included.

Among the 216 cells the junction ranks first, with an index of 0.155. Its loss removes 33.5 of the 216 units of delivered power. Of that loss, 32.5 units are the transfer that the importing sectors need across their boundaries, or 76 percent of the 42.6 units they lack, since the mesh restores only the remaining 10.1 units. The cell that ranks second has an index of 0.060, and the junction therefore stands at 2.6 times the next cell.

This ranking is structural, since it follows from the network that the scans reveal and no count of damage events enters it. The data decide whether that network is the cube's, and Section 4 tests this in three ways.

4. Validation Against Field Data

The first test asks whether regeneration fails more often as the index rises. Of the 268 events, 251 were regenerated within a volley interval (93.7%, 95% CI 90.1 to 96.0, score interval). Table 1 gives the result by tier, fixed from the model before any outcome was tallied. Regeneration succeeded in 226 of 233 tier A events (97.0%, 95% CI 93.9 to 98.5), in 24 of 30 tier B events (80.0%, 62.7 to 90.5) and in 1 of 5 tier C events (20.0%, 3.6 to 62.4). Failure was 6.7 times as frequent in tier B as in tier A (95% CI 2.4 to 18.5, log method). We compared the tiers pair by pair with an exact conditional test on the two-by-two table of outcomes, and we multiplied each p value by three to allow for the three comparisons. Tiers A and B differ (adjusted p = 0.004), and so do tiers A and C (adjusted p < 0.001). Tiers B and C are not separated at the 5 percent level (adjusted p = 0.051), because tier C holds only five events, one for each of its cells.

Fire did not seek the cells of high consequence. Tier A holds 200 of the 216 cells (92.6%) and received 233 of the 268 events (86.9%). Tier B holds 11 cells (5.1%) and received 30 events (11.2%), about twice its share. Tier C, where loss is of high consequence, holds 5 cells (2.3%) and received only 4 of the 267 events recorded before the change of target in 2373 (1.5%). Before that change, 192 of the 220 events of 2373 (87.3%) fell on tier A cells.

Four tier C events other than the junction were hits on primary-arm cells, each with an index of at most 0.060, and the cube survived all four. One was regenerated within the interval and three were not, and the scans, moreover, show the cube operating on rerouted flow.

The second test compares predictors, and its force is limited. The model, built from frames recorded before the change of target, ranks the junction first among 216 cells. The fleet then fired at that cell, and the cube broke up shortly afterwards. A cell drawn at random would rank first with probability 1 in 216, or 0.46 percent, but this is not a significance level, because a junction at which six primary arms meet ranks first in any flow model of this kind and the model was specified with the target known. What the test decides is that the scans contain the arms and their junction, and that the two predictors available to a gunner do not find it. The two predictors are taken from the same frames. Ordering cells by emission, as fire directed at the most active sections would, places the junction at a mid-rank of 206.5, because 196 cells emit more and 20, including the junction, tie with it. Ordering cells by closeness to the geometric centre places it at a mid-rank of 20.5, because 8 cells lie closer and 24, including the junction, are equally close.

Behaviour after the loss is the third test. In the frames following the junction event, regeneration signatures disappear from the three importing sectors within two frames.

Dependence on κ appears only at the lower end. We re-ran the model with T = 40, feeders of 1.5 units and m = 40/κ, and recomputed every index. The junction ranks first for every ratio from 100 to 6400, and its index rises from 0.037 to 0.170 over that range. At κ = 50 the mesh carries enough that no cell exceeds an index of 0.05 and the junction falls to rank 28, which is the cube the doctrine describes. The scans, however, do not describe it, because the fitted interval of 617 to 1037 lies far above 100.

5. Failure Modes

Localisation of an impact is uncertain by one cell. A hit recorded next to the junction could have fallen on it, and the reverse. The neighbours of the junction on the primary arms have indices of at most 0.060, and a one-cell error in 2373 would therefore have named a cell of far smaller consequence. The registry alone cannot exclude such an error. The third test can, however, because the loss of a neighbour on one arm would cut the transfer of one sector at most, whereas regeneration ceased in all three importing sectors, as the loss of the junction requires.

Only one engagement supplies the test, and one success cannot estimate a rate of success, and we therefore claim none. It does separate the predictors, because the model ranked the cell first and the two baselines did not.

Layout comes from the cube of 2373. The earlier scans resolve the primary arms too coarsely to locate a junction, and we used them only to estimate κ. We assume that the earlier cubes shared the layout. If they did not, the tiers assigned to their 47 events are unreliable. However, the pattern survives their removal, since the tally restricted to 2373 gives 186 regenerated of 192 tier A events (96.9%) and 19 of 24 tier B events (79.2%), against 97.0 and 80.0 percent over all three groups.

A single network stands for both power and structure, since the scans place spines and conduits in the same corridors. A structural network of its own could contain further junctions along the spines, which the model would not find.

The registry holds no cube scanned after 2373, and the findings therefore describe the cubes scanned and nothing later.

6. Conclusion

The corrected claim, therefore, is the one stated at the outset. The cube of 2373 was decentralised in control and centralised in load, and its power and structural paths met at a junction that the mesh could not bypass. A model built from frames recorded before the change of target in 2373 ranks that junction first of 216 cells, although the same cell ranks 206.5 by emission and 20.5 by closeness to the centre.

It is tempting to take the destruction of 2373 for one fortunate volley, and equally tempting to take it for proof that every cube carries the same weak point. The first fails because the tiers grade the whole record and not one event, since regeneration failed in 3.0 percent of events on cells of negligible consequence, in 20.0 percent on cells of intermediate consequence and in 4 of 5 on cells of high consequence. The second goes beyond the evidence, because the layout is measured on one cube and the earlier scans are too coarse to show a junction.

The records that the doctrine read wrongly are the engagement logs themselves. The encounters of 2365 to 2367 were filed as proof that no point of a cube matters, and the engagement of 2373 was filed as an exception. All three groups record the same fact, that fire on cells of low consequence is repaired and fire on a cell of high consequence is not. The tier A and tier B events in the registry are therefore measurements of regeneration and not tests of vulnerability.

Load, and not command, therefore decides where a cube can be destroyed. Exhaustion fire spent its volleys on cells whose loss the mesh makes good, and the cube repaired them. A gunnery plan that ranks cells by convergence index from the scans of an engagement's first exchanges, and which fires on the highest-ranked cell before any other, is the plan that the record of 2373 supports.

Borg cubestructural redundancyload pathsnetwork flow modelregenerationpassive structural scanningSector 001 engagement

References

  1. Starfleet Tactical Analysis Division (2374). Passive structural scans and damage event records, Borg cube engagements 2365–2373. Starfleet Tactical Engagement Registry, series B-7.
  2. Corvane, R. (2369). Tactical assessment of Borg vessels after the first encounters, the absence of critical systems and its consequences for fleet gunnery. Starfleet Tactical Analysis Reports, TAR-69-14.
  3. Brennick, S. (2371). Load-path convergence in segmented space frames under distributed damage. Utopia Planitia Fleet Yards Technical Notes, 15(2), 88–109.
  4. Ardell, P. (2370). Primary and secondary conduit capacity in distributed-generation hulls. Starfleet Corps of Engineers Reports, SCE-70-112.
  5. Tavenor, M. (2371). Gunnery against self-repairing targets, exhaustion fire and its limits. Journal of Starfleet Tactical Studies, 10(2), 112–131.
  6. Hallister, L. (2373). Regeneration signatures of Borg hull sections in the encounters of 2365 and 2367. Starfleet Corps of Engineers Reports, SCE-73-231.
  7. Marrowby, E. (2367). Structural resolution limits of passive scanning at tactical range. Subspace Engineering Quarterly, 13(2), 140–158.
Read this article inside the full journal experience — browse by faculty, search across universes, and filter by discipline.
Open in Uncited Press →