Egg or Ore: Two Field Screens for Horta Egg Clusters Among Silicon Nodules in 142 Survey Chambers of the Janus VI Pergium Workings, 2268–2275

1. Introduction

Janus VI is a pergium mining colony. In 2267 its miners lost a number of their colleagues to a creature that moved through solid rock and burned its victims with a corrosive agent. The creature proved to be a Horta, a silicon-based being, the sole surviving adult of its species, guarding the eggs that were its next generation. The miners had been destroying those eggs without knowing it, because the eggs resembled the silicon nodules that the workings contain in quantity (Federation Mining Operations Records, 2267). A Vulcan mind-meld carried out by a visiting Starfleet science officer allowed the Horta to be understood, and it etched its meaning into the rock for the miners to read. The parties agreed terms: the Horta would tunnel for the miners, and the miners would protect the young. Most of the eggs hatched in 2267, so the clusters studied here are unhatched remnants and later formations, not the original clutch.

The arrangement leaves a practical problem. The workings hold many chambers in which clusters of silicon-rich spheres lie in the host rock, and some are mineral nodules and some are eggs. Cutting or blasting near an egg cluster destroys it, and the loss of a clutch is a serious loss for the Horta. Survey teams therefore need a screen that separates the two before they work a chamber. The Administration's standing procedure of 2269 formalised a practice in place since 2268, that chambers with suspected biogenic clusters be left untouched until they were classed (Federation Mining Operations Administration, 2269).

We evaluate two screens that survey teams used in the years after the accommodation. An earlier study in this journal reviewed the non-carbon life screen used in pre-terraforming surveys and could not estimate its sensitivity, because no independent method existed to check the worlds that the screen cleared. The Janus VI chambers were all followed up, and this allows us to do here what that study could not. We write in 2376 from the archived survey records.

2. Methods

The Federation Mining Operations Records hold the survey logs of the Janus VI workings. We took every chamber surveyed between 2268 and 2275 in which the survey team logged a cluster of objects resembling silicon nodules (Federation Mining Operations Records, 2278). That gave 142 chambers. No chamber was worked, cut or blasted until the cluster had been classed, as the standing procedure required, so the classification did not depend on a disturbance that could have destroyed an egg cluster.

Survey teams used two screens. The first read one signature, a core-temperature gradient: a cluster was called an egg cluster if the temperature at its core was higher than that of the surrounding rock by more than a set margin, in the manner later described for biogenic silicate clusters (Kirkpatrick-Halloran, 2274). The second screen took three readings, the gradient, faint periodic motion inside the spheres over a ten-minute observation, and a low-level energy reading, and called an egg cluster if at least two were positive. Survey teams logged both screens for each chamber before any classification was made, and the teams did not know the reference result.

A reference standard, independent of both screens, was applied in two parts. A chamber was classed as holding an egg cluster if eggs hatched in it within 24 months of the survey, or if the Horta, consulted through a communication method that the field office developed, indicated the cluster as a clutch. A chamber was classed as holding mineral nodules if neither happened and a sample assay taken after the 24 months was consistent with mineral concretion, using the morphology criteria of Aldercott-Ng (2271); the assay was taken from a sample that did not disturb the cluster as a whole. We report sensitivity, specificity and the positive and negative predictive values of each screen with score intervals. The two screens were applied to the same chambers, so we compared them with the exact form of the paired test for discordant results (Delacroix-Brandt, 2373), and we give paired differences with Wald intervals, separately among the egg clusters and among the nodule chambers. Predictive values depend on how common egg clusters are in this sample, which we state.

3. Results

Of the 142 chambers, 47 (33.1%; 95% CI 25.9 to 41.2) held egg clusters and 95 held mineral nodules. The single-reading screen called 48 chambers egg clusters (33.8%) and the three-reading screen called 63 (44.4%).

The three-reading screen detected 44 of the 47 clusters, a sensitivity of 93.6% (95% CI 82.8 to 97.8). The single-reading screen detected 36, a sensitivity of 76.6% (95% CI 62.8 to 86.4). The difference was 17.0 percentage points (95% CI 3.4 to 30.6). Ten clusters were found by the three-reading screen and missed by the single reading, and two were found by the single reading and missed by the three-reading screen; the exact paired test gave p = .039.

Among the 95 mineral nodule chambers, the single-reading screen called 83 correctly, a specificity of 87.4% (95% CI 79.2 to 92.6), and the three-reading screen called 76 correctly, a specificity of 80.0% (95% CI 70.9 to 86.8). The specificity of the three-reading screen was 7.4 points lower (95% CI −15.2 to 0.5). Eleven chambers were called correctly by the single reading and wrongly by the three-reading screen, and four were called the other way; the exact paired test gave p = .12.

The single-reading screen therefore missed 11 clusters and raised 12 false alarms, 23 errors in 142 chambers (16.2%). The three-reading screen missed 3 clusters and raised 19 false alarms, 22 errors (15.5%). Positive predictive value was 75.0% (95% CI 61.2 to 85.1) for the single reading and 69.8% (95% CI 57.6 to 79.8) for the three-reading screen. Negative predictive value was 88.3% (95% CI 80.2 to 93.3) and 96.2% (95% CI 89.4 to 98.7).

4. Discussion

Egg clusters were found more often by the three-reading screen than by the single reading, and the gain is supported, though the lower limit of the interval is about 3 points. The loss of specificity is smaller in the estimate than the gain in sensitivity, and it is not distinguishable from no loss. Total errors were almost equal, so on counts alone neither screen is better.

Counts do not settle the choice, because the two kinds of error do not cost the same. A missed cluster leads to a chamber being worked as if it held nodules, and a cluster worked in that way is destroyed. That is the event of 2267, on a smaller scale. A false alarm holds a chamber back until the cluster is classed by the slower reference route, which delays the work; we did not measure the delay or its cost in pergium. The first cost cannot be recovered, and the second can. If the Administration weights a missed clutch even modestly above a false alarm, the three-reading screen is the better procedure. We take no position on the weight, which is for the colony and the Horta to judge together.

The design answers a question that the pre-terraforming screen could not. Because every chamber here was left untouched until it was classed, a cluster that a screen cleared was still followed up, and a missed clutch could be counted. Screens elsewhere that clear a site and move on cannot report their misses, and a reported sensitivity from such a screen should be treated with caution.

We also note what the screens do not do. Both rest on physical signatures that we have no reason to think are unique to eggs, and the Horta's own account of their eggs, obtained through the field office's communication method and set out in the observations after the accommodation (Ashworth-Lindeman, 2272), should be preferred to any instrument wherever it can be had.

5. Limitations

The study is of one site and a single set of workings, and the eight years of records span changes in personnel and instruments. Survey teams may have learned to use the screens better over time, and we did not model that. The reference standard is imperfect in one direction. A cluster classed as mineral nodules because nothing hatched within 24 months could be a clutch that did not develop, and such a clutch would be counted as a false alarm when a screen called it an egg cluster. The Horta's indication depends on the field office's communication method, and the accuracy of that method was not measured. Chambers on the same tunnel route are not independent, since a Horta's indication for one may bear on its neighbours, and we did not model agreement by route, so the intervals may be somewhat narrow. Only three clusters were missed by the three-reading screen, so the estimate of its miss rate is imprecise, with an upper limit near 17%. Predictive values would differ in workings where egg clusters were rarer or commoner, and the chambers in our sample were selected because a team had logged a nodule-like cluster, which is a more restricted group than every chamber in the workings. The screens have not been tested outside Janus VI.