Uncited Press › Star Trek Open the interactive journal →
Trek · Artificial Intelligence & Distributed Cognition

How Far Does the Collective Reach? Latency, Range and Capacity of the Borg Link from Intercepted Traffic and Severed Drones, 2365–2375

Dr. Anuradha Venkataraman1
1 Daystrom Institute, Subspace Theory Division
Received 21 Aug 2026 · Revised 5 Oct 2026 · Accepted 24 Oct 2026

Abstract

Ten years after the first Federation contact with the Borg, the speed, reach and carrying capacity of the link that joins the drones of the Collective have never been bounded. We bound them from two Federation record sets: 118 sessions of intercepted carrier from nine engagements between 2365 and 2373, and 50 drones in 20 episodes of separation from the Collective. For 99 timed pairs of events, the lag between an adaptation at one drone and its appearance at another had a median of 0.43 seconds and showed no trend with separation over 2.5 to 21 light-seconds (slope −0.003 seconds per light-second, 95% CI −0.013 to 0.007). Among drones with a known source, 1 of 24 within 6.5 light-seconds of a relay had lost shared shield adaptation after an hour, against 10 of 16 beyond 14 light-seconds. The radius at which half of drones lose it was 24.5 light-seconds (95% CI 11 to 53). The Shannon ceiling of one cube's carrier was 8.30 gigabits per second (95% CI 7.85 to 8.78), or 0.52 to 1.38 megabits per second per drone. Among drones with no source, nine lost lockstep movement after a median of 8 seconds and shared adaptation after 20 minutes, and individual behaviour began in 8 of 10 after a median of 30 hours. The link is fast, short in reach and narrow, and the Collective's reach beyond a vessel is a matter of relay spacing.

1. Introduction

In 2365, in a section of the Enterprise-D, a Borg drone that had come aboard from the cube was brought down by a single phaser discharge. A second drone beamed in a moment later and went about its work beside the body. When the officer fired again, the beam was turned aside by a shield that the first drone had not carried.

Ten years have passed since that section was cleared, and we still cannot say how the second drone knew. Starfleet has met the Collective in nine engagements since 2365, and in each of them a Federation vessel recorded its subspace traffic. But nobody has bounded the speed, the reach or the carrying capacity of the link that must have done the telling. Without those bounds a boarding party cannot judge how fast an adaptation spreads, how far a drone can stray from its vessel, or what a drone that loses contact is still able to do.

The basic idea here is simple: a link of finite speed, range and capacity must leave marks on both sides of its failure. Traffic recorded while the link works should show how long an event takes to cross it and how much it can carry. Drones severed from it should shed what they shared in an order set by how much of the link remains, and the distance at which they begin to shed it should mark its reach.

Five questions follow from this idea. How long does the link take to carry an event from one drone to another? Does that delay grow with distance? How far from a relay can a drone operate before the link lets go? How much can the link carry? And in what order do drones lose their shared abilities once it lets go? We answer them in that order. Earlier work showed that the Collective's carrier can be received passively at useful range (Venkataraman, 2371) and resolved by long-baseline receivers (Marrowby, 2374), but neither study asked what the carrier carries between drones.

We write in 2375 and use Federation records only. By a relay we mean any Borg vessel, or any device aboard one, that emits or re-emits the link carrier, so that the vessel a drone came from is its first relay.

2. Methods

Two record sets bear on the link. The first is the set of subspace intercepts held in the Starfleet Fleet Telemetry Archive, made by Federation vessels in nine engagements with Borg vessels between 2365 and 2373, by the archive's count (Starfleet Fleet Telemetry Archive, 2375). We cut these into 118 sessions, a session being an unbroken recording of at least 60 seconds of one cube's carrier, and all 118 record cube carriers. The second is the Starfleet Tactical Engagement Registry's extract of drones observed after losing contact with the Collective (Starfleet Tactical Engagement Registry, 2375).

A timed event pair joins an adaptation at one drone to the first appearance of the matching adaptation at a second drone that had not been fired on. Adaptations were logged by tricorders and ship sensors that resolve 0.05 seconds (Thornquist, 2372). We found 58 pairs inside a single cube and 41 between a cube and a drone on another Borg vessel in the same engagement, separated by 2.5 to 21 light-seconds, where one light-second is about 300,000 kilometres. Of the nine engagements, three involved more than one Borg vessel at once, and these supplied the 41 cross-vessel pairs. At a drone that had not been fired on, an adaptation was read as a shift in its personal-shield modulation, and times at the far vessel were corrected for the sensor path. The lag of a pair is the delay between the two events. We compared the two sets of pairs with a difference in means and a normal interval, and regressed lag on separation for the 41 cross-vessel pairs by least squares.

For capacity, each session gave the occupied bandwidth of the carrier and the signal-to-noise ratio at the receiving vessel. We applied the Shannon limit to each session and took the geometric mean, since capacities are positive and skewed, with a t interval on the log scale. The result is a ceiling for the carrier of one cube as received. To turn it into a ceiling per drone we divided by the number of drones in a cube, a figure Starfleet does not know. Tactical Analysis scan estimates run from 6,000 to 16,000 drones (Starfleet Tactical Analysis Division, 2368), and we used 11,000 as a central value.

Each severed drone was placed in a band by its distance in light-seconds from the nearest relay, reconstructed from vessel tracks. The near band runs from 0.0004 to 6.5 light-seconds and the mid band from 14 to 190. A third group has no source within any distance the tracks can bound, either because its vessel was destroyed or because the drone was stranded, as the single drone recovered by Federation crew in 2368 was. For that drone only the onset of individual behaviour can be timed, from the crash of its vessel. These distances are radii from a source and not diameters of any region. We coded three abilities: lockstep movement with other drones, shared shield adaptation, and individual behaviour, meaning any act that a Starfleet observer coded as chosen, such as hesitation or refusal of a task. Two analysts coded all 50 drones independently from the Registry files, disagreed on the band of 3 and on the timing of an ability in 5, and settled each by reviewing the vessel tracks and the sensor logs together. Time zero for each episode was the last moment at which a source was known to be transmitting within reach, and where the record gave only a window we took its midpoint.

For drones with a source, the outcome was whether shared adaptation had been lost one hour after separation. We fitted a logistic regression on the base-10 logarithm of distance, with standard errors clustered by episode and t critical values on 14 degrees of freedom, since there are 15 such episodes. The distance at which half of drones have lost shared adaptation follows from the fit, with an interval by the delta method. Because drones in an episode share their fate, we also compared the share of episodes with at least one loss between the bands by Fisher's exact test. For drones with no source we report medians and quartiles of the time to each loss, and give the time to individual behaviour among the drones that showed it.

3. Results

The 118 sessions came from nine engagements. The Registry gave 50 drones. Of these, 24 lay in the near band, 16 in the mid band and 10 had no source. They fell into 20 separate episodes, numbering 9, 6 and 5 in the three groups.

Nine of the 10 with no source came from four episodes and could be timed on all three abilities. The tenth was the drone of 2368, timed on individual behaviour only. We leave the faction that followed Lore out of the counts, since Starfleet did not see its separation.

It is striking how little distance matters to the delay. The 58 pairs inside a cube had a median lag of 0.42 seconds (interquartile range 0.32 to 0.55), and the 41 cross-vessel pairs had a median of 0.43 seconds (0.35 to 0.55). The difference in means was 0.006 seconds (95% CI −0.064 to 0.076). Across the cross-vessel pairs, lag changed by −0.003 seconds per light-second (95% CI −0.013 to 0.007), and over the 18.9 light-seconds spanned that is a change of −0.06 seconds (95% CI −0.24 to 0.12). No lag in the 99 pairs exceeded 0.97 seconds.

Range behaves differently, and the near band shows it: 1 of 24 drones had lost shared adaptation after an hour (4.2%, Wilson interval 0.7 to 20.2%), and in the mid band 10 of 16 had (62.5%, 38.6 to 81.5%). By episode, 1 of 9 near-band episodes showed any loss against 5 of 6 in the mid band (Fisher exact p = .011). The one near-band loss came from the 6.5 light-second episode, the farthest in its band. In the logistic fit, the half-loss radius was 24.5 light-seconds, with a 95% interval of 11 to 53. The fit predicts that under 1% of drones have lost shared adaptation at 1 light-second, 19% at 10, 58% at 30, 81% at 60 and 93% at 120. Table 1 gives the bands.

The carrier was wide and clean, with an occupied bandwidth that had a median of 1.81 gigahertz (interquartile range 1.55 to 2.09) and the signal-to-noise ratio a median of 13.9 decibels (11.9 to 16.1). The Shannon ceiling per session had a geometric mean of 8.30 gigabits per second (95% CI 7.85 to 8.78), with sessions ranging from 3.5 to 17.0. Divided among 11,000 drones this is 0.76 megabits per second per drone. It is 1.38 megabits per second at 6,000 drones and 0.52 at 16,000.

Drones with no source lost their abilities in one order. The 9 that could be timed all lost lockstep movement, after a median of 8 seconds (interquartile range 6 to 12, range 3 to 17), and all lost shared adaptation, after a median of 20 minutes (14 to 28, range 9 to 42). Individual behaviour appeared in 8 of the 10 (80%, Wilson interval 49 to 94%), after a median of 30 hours among those that showed it (22 to 43, range 14.0 to 70), and the other two were still silent when observation ended. The order lockstep, adaptation, individual held in all 7 of the timed drones that showed all three. Table 2 gives the times.

4. Discussion

The link is fast, short in reach and narrow, and each of these is a measured property.

Delay belongs to the link and not to the distance. A median of 0.43 seconds held inside a cube and between vessels 21 light-seconds apart, and the slope with separation is indistinguishable from zero. So an adaptation reaches every drone within reach after about four-tenths of a second, wherever it stands. For a boarding party this fixes a window. Half of drones fired on within about four-tenths of a second of an adaptation have not yet received it, and most fired on within three-tenths have not.

Reach is set by signal and not by delay. Lag stays flat out to 21 light-seconds, but half of drones have lost shared adaptation by about 25 light-seconds. A link that failed because messages arrived late would show lag rising before it failed. This one fails with its delay unchanged, so the limit lies in what arrives and not in when. The figure is a radius from a relay, and it is small. It is about 7.3 million kilometres, about twenty times the distance from Earth to its Moon and millions of times smaller than the span between neighbouring stars.

For the Collective the consequence is large. A drone aboard a Federation vessel within a few light-seconds of its cube, as in any boarding action, lies inside the near band, so a boarding action does not sever drones by distance. They are cut off by destroying the source, or by carrying them beyond about 25 light-seconds from it. And a Collective that holds drones together across interstellar distances must do so through relays, since no single source holds a drone much beyond this radius. How far apart the relays themselves may stand is not settled here: traffic between vessels crossed 21 light-seconds, the widest separation in the record, with its delay unchanged. Its reach beyond one vessel is a matter of relay spacing. Relays are what Starfleet should look for along any route the Borg travel.

Capacity fits the order in which abilities fail. A ceiling near 8 gigabits per second for a whole cube leaves about three-quarters of a megabit per second for each drone. That is room for summaries of state but not for a full sensory feed from every drone, so the link carries summaries of state, and lockstep movement needs the freshest of these and fails within seconds. Shared adaptation needs less and lasts for tens of minutes. Individual behaviour is what remains when the shared state stops arriving, and it begins after a median of thirty hours. The faction that followed Lore cannot test this order, since it was cut off out of Starfleet's sight and its individuality began with a drone returned to it, so we leave it out of the counts (Starfleet Security, 2370).

Severed drones do not simply stop. Half of the drones beyond reach of any relay were behaving as individuals within about thirty hours, and 8 of 10 did before observation ended, so the Collective's hold on a drone is a matter of continuous supply.

5. Limitations

The record is archival and some of its margins are wide. Distances were reconstructed from vessel tracks and are good to perhaps a fifth, and this shifts the half-loss radius by about the same fraction. The mid band holds six episodes, so the interval on that radius is wide, from 11 to 53 light-seconds, and the slope of the fit is not tight at 15 episodes. The share of episodes with a loss does not depend on the fit.

Capacity is a ceiling, and it was measured by Federation receivers whose signal-to-noise ratios depend on where they stood. Per-drone figures scale inversely with the assumed population of a cube, and the radius scales directly with the reconstructed distances. Neither scaling touches the delay, but sessions and pairs cluster within nine engagements, so the intervals on delay and capacity are narrower than they would be for independent draws.

In the section of 2365, on the account of this paper, the second drone was not warned by anything that crossed a room. It was served by a link that moves an event in about four-tenths of a second, over a distance that is nothing against the link's reach. A second shot fired within that interval would probably have met a shield not yet adapted.

We first showed that the link has finite speed, reach and capacity, using recorded traffic and the order of losses in severed drones. We then put numbers on each: a lag of about 0.43 seconds that does not grow with distance, a half-loss radius near 25 light-seconds from a relay, and a carrier of about 8 gigabits per second for a whole cube. As future work, we intend to test the radius against live intercepts from a pair of receivers set 30 light-seconds apart, so that a drone can be followed across the edge of reach.

Borg Collectivesubspace linklatencyBorg relaysevered droneschannel capacitydistributed cognition

References

  1. Starfleet Fleet Telemetry Archive (2375). Subspace intercepts of Borg vessel carriers in nine engagements, 2365–2373. Starfleet Fleet Telemetry Archive, series BI-2.
  2. Starfleet Tactical Engagement Registry (2375). Extract of drones observed after loss of contact with the Collective, 2366–2373. Starfleet Tactical Engagement Registry, series B-4, extract S.
  3. Starfleet Security (2370). Recovery of disconnected Borg drones and their records, 2369–2370. Starfleet Incident Records, SIR-70-014.
  4. Venkataraman, A. (2371). Passive reception of Borg subspace carriers, methods and range. Subspace Engineering Quarterly, 17(1), 133–158.
  5. Marrowby, E. (2374). Noise floors of long-baseline subspace receivers. Subspace Engineering Quarterly, 20(1), 12–29.
  6. Thornquist, B. (2372). Timing resolution of tricorder and ship sensor logs in field use. Starfleet Corps of Engineers Reports, SCE-72-044.
  7. Starfleet Tactical Analysis Division (2368). Population estimates for Borg cubes from sensor scans, 2365–2367. Starfleet Tactical Analysis Reports, TAR-68-07.

Cited By

Read this article inside the full journal experience — browse by faculty, search across universes, and filter by discipline.
Open in Uncited Press →