Spice Harvester and Carryall Extraction Under Worm-Strike Time Constraints: A Timing-Budget Model Validated Against 156 Wormsign Alerts on Arrakis, 10150–10240 AG

1. Introduction

Spice on Arrakis is gathered by mobile factories, the harvesters or crawlers, that work fresh deposits on open sand. Their rhythmic machinery draws sandworms, and no harvester can outrun or withstand one. The machine's survival therefore depends on an aircraft. A carryall, a heavy suspensor-assisted lifter held by the wing that serves the harvest district, must reach the harvester, couple to it and lift it clear before the worm surfaces beneath it. That dependence has seldom been expressed as a budget of seconds.

The Atreides household record of 10191 AG documents the failure that motivates this paper. During the Atreides assumption of the fief, a harvester under inspection by the Duke's party was lost to a worm because no carryall arrived, and its crew was saved only by crowding into the inspection ornithopters. In that episode the binding constraint was whether a carryall came at all.

We treat extraction as an engineering system with a single time budget. The model is specified from parameters fixed in advance and then checked against the recorded outcome of every alert in our archive sample. Our argument is that procedure written around median worms and median warning leaves crews exposed in a tail that is infrequent but foreseeable, and that the tail can be quantified. We write from the vantage of 10245 AG.

2. System Description

The system has four elements. A harvester works a spice field unshielded, because a shield drives a worm into frenzy. A spotter screen of one or more ornithopters watches the surrounding sand for wormsign, the moving ripple of a worm travelling beneath the surface. A carryall waits at a staging point on rock or on stable sand and is shared by several harvesters. The alert chain joins them, running from the spotter's call through the wing controller to the carryall pilot's lift from staging.

Our records come from three series. Harvest loss registers and wormsign alert returns in the CHOAM Directorate Archive cover all three administrations, and from 10193 AG the returns also name the staging point of the carryall assigned to each alert. Before that date, staging distances come from the carryall wing dispatch logs in the Harkonnen Administrative Records (10150–10190 AG) and from the dispatch memoranda of 10191 AG in the Atreides Household Archive. We assembled 156 alerts for which the warning distance, the time from sighting to the worm's surfacing at the harvester site, and either the staging distance or a record that no carryall was available could all be read: 58 from the Harkonnen fief, 6 from 10191 AG and 92 from 10193–10240 AG. All five harvester losses in the sample fall in the last period. Harkonnen-era loss inquiries survive only in fragments and could not be matched to alert returns, and the inspection loss of 10191 AG lacks a logged warning distance.

The warning distance D is the range at which wormsign was first sighted, as logged by the spotter. Its median in the 156 alerts was 4.0 km, the 10th percentile 2.5 km and the range 1.1–7.3 km. Short warnings were concentrated on days of suspended dust and in dune fields where the worm's approach lay in shadow, the two conditions that Threll (10233 AG) identified as limiting visual detection range. No instrumented vibration detection was in use.

Approach speed v was obtained for each alert by dividing the logged warning distance by the logged interval between sighting and surfacing at the site. Across the 156 alerts speeds ranged from 4.2 to 15.2 m/s, with mean 8.7 m/s (SD 2.3), median 8.1 m/s, 90th percentile 12.4 m/s and 95th percentile 13.8 m/s. The distribution is right-skewed. Straka (10234 AG) described how mature makers accelerate as they close on a rhythmic source but reported no instrumented speeds, and no independent measurement exists against which to test these values.

The time required for extraction has three parts. Alert latency runs from the spotter's sighting to the carryall leaving its staging point; in 41 alerts with time-stamped relay entries it had a mean of 38 s (SD 11 s; 95% CI 34.6–41.4 s) and a 90th percentile of 55 s. Carryall transit is staging distance divided by unladen cruise speed. Logged transits give a cruise speed of about 50 m/s, in line with the rated performance reported by the Ixian Consortium Suspensor Works (10218 AG), and staging distances had a median of 6 km and a 90th percentile of 11 km, so median transit is 120 s. The on-station sequence covers descent and positioning, hookup and lift-off to a point clear of the strike area. Flight recorder and operator logs for 34 extractions give a mean of 29 s (SD 4.6 s; 95% CI 27.5–30.5 s).

3. Analysis / Model

The clock starts when wormsign is sighted. The worm's arrival time T equals D divided by v. The time required R is the sum of alert latency, transit time and the on-station sequence. Margin M is T minus R; a negative margin means the worm arrives first.

With mean alert latency, median staging and the mean on-station sequence, R is 38 + 120 + 29 = 187 s. The table sets out six scenarios. At median warning and median speed the margin is about 307 s, or roughly five minutes. A 90th-percentile worm cuts this to about 136 s, and a 95th-percentile worm to about 103 s. Short warning alone, with a median worm, still leaves about 122 s.

The combination is what matters. A 95th-percentile worm sighted at the 10th-percentile warning distance arrives in about 181 s, which leaves a margin of about −6 s against a typical carryall. If the carryall is also staged at the 90th-percentile distance and the alert chain runs at its 90th-percentile latency, R rises to 304 s and the margin falls to about −123 s. With typical carryall timing, the break-even approach speed at 2.5 km warning is about 13.4 m/s; at 4.0 km warning it would be about 21 m/s, well beyond any speed in the record.

Two design quantities follow directly. To hold a 95th-percentile worm off with a typical carryall, the spotter screen needs about 2.6 km of warning for zero margin and about 3.4 km for a 60 s reserve. Where that warning cannot be assured, the carryall must be staged closer. With 2.5 km of warning, a 95th-percentile worm, a 60 s reserve and mean latencies, the transit allowance is about 54 s, which corresponds to a staging distance of roughly 2.7 km.

4. Validation Against Field Data

For each of the 156 alerts we computed a modelled margin from its own warning distance, approach speed and staging distance. Because T is observed for every alert, the validation tests the time-required side of the budget. Mean alert latency was imputed for the 115 alerts without relay time stamps, and mean on-station time wherever an extraction was not timed. Alerts were grouped by margin class before their outcomes were examined. A loss means the harvester was destroyed; crew outcomes are recorded separately.

In 106 alerts the modelled margin was 120 s or more, and none ended in a loss (0%, exact 95% CI 0–3.4%). In 33 alerts with margin between 0 and 120 s there was one loss (3.0%, CI 0.1–15.8%). Of 14 alerts with negative margin and a carryall dispatched, 2 ended in a loss (14.3%, CI 1.8–42.8%). In 3 alerts no carryall was available, and 2 of those harvesters were lost (66.7%, CI 9.4–99.2%); both crews were lifted off by ornithopter. Taken together, 4 of the 17 alerts with negative margin or no carryall ended in a loss (23.5%, CI 6.8–49.9%), against 1 of the 139 with positive margin (0.7%, CI 0.0–3.9%; Fisher's exact test, p ≈ 0.0005).

Negative margin was not rare. Seventeen of 156 alerts, about 11%, fell in that class by one route or another. Most alerts with negative margin nonetheless ended without a loss. The likeliest reason is that actual latency and on-station times in those alerts were shorter than the imputed means, since 115 alerts carry no latency stamps. Crews may also have begun disengaging before the formal call, and a worm surfacing near but not beneath the machine would leave the harvester intact.

Five losses cannot fix a loss rate with any precision, and the confidence intervals above are wide. Selection also matters, since loss inquiries record timings more fully than routine returns. The 58 Harkonnen-era alerts can contribute only non-losses, because their inquiries could not be matched, so any unrecorded loss among them sits in the no-loss denominators. The validation therefore shows that losses fall where the model predicts little or no margin. It does not supply a calibrated probability of loss for any single class.

5. Failure Modes

Absence of a carryall is the failure mode the 10191 AG record documents, and it defeats the budget outright. A carryall serving several harvesters may be committed elsewhere, grounded for maintenance or held back by its controller. Evacuating the crew by ornithopter saves lives but surrenders the machine, as in both such cases in our sample. Its limits are those of the ornithopter itself: payload, and control authority, which narrows under the thermal and particulate loading of the interior and saturates in strong gusts.

Coriolis storms ground ornithopters and carryalls alike. A storm front near the harvest district can remove the spotter screen and the lifter together, so work under an approaching storm operates without any budget at all. Basin-scale survey work associates storm-season reworking of the dunes with wider worm spacing, which lengthens the typical distance at which an answering worm first appears but does not shorten the tail of fast approaches.

Visibility governs warning distance, and the model is most sensitive to warning distance when speed is high. An instrumented ground-vibration screen around the harvester might extend warning in poor visibility. We offer this as a hypothesis for field trial; no such system appears in the records we examined.

Shields are excluded from the design because a shield would drive an approaching worm into frenzy. The machine's rhythm is the signal worms answer, and in the southern erg thumper calls have been answered in a median of roughly half an hour. Procedure should therefore assume that an alert may come early in any shift.

6. Conclusion

The timing budget shows ample margin for the ordinary alert, about 307 s with median worm, warning and staging. That margin disappears, falling to about −6 s, when a 95th-percentile worm is sighted at the 10th-percentile warning distance. A distant carryall and a slow alert chain push it to about −123 s. The archive agrees in direction, because losses concentrate among alerts with negative modelled margin or no carryall, although five losses allow only a coarse test.

Procedure planned on median conditions therefore understates the risk that matters. We recommend that harvest procedure state a tail-case margin for each operating area, computed from a 95th-percentile approach speed and the warning distance the spotter screen can actually assure, in the manner Iskadar (10239 AG) proposed for margin disclosure in extraction industries. Carryall staging should be limited accordingly, to roughly 2.7 km when warning may fall to 2.5 km, or the screen extended to about 3.4 km. Harvesters should not work without a carryall committed to them, and ornithopter evacuation of crews should be planned and drilled as the fallback it has proved to be.