Effector Repair as Routine Work: Completion Times and Rework in 4,862 Maintenance Jobs on Tessendrine Orbital, 1750–1799 CE

1. Introduction

Effectors are the Culture's primary working tool. They act on matter and machines at a distance, and on Tessendrine the habitat's upkeep is done largely with them: a split in a floor, a blocked service run or a failed fitting is ordinarily mended by someone who stands at a distance from the fault and works on it with an effector (Haldrevane, 1694 CE). Effectors have other uses. This paper concerns none of them, and the register we analyse contains no job outside the repair of fabric, fittings and plant.

Work of this kind has been described but seldom measured. The first treatment explained why reach matters for repair of Orbital fabric and gave worked examples (Haldrevane, 1694 CE). A later paper proposed the classification of jobs that most maintenance records now use (Soumerrin & Truvelle, 1731 CE). Effector duty cycles in vehicle repair were examined aboard General Systems Vehicles (Bethavaille, 1745 CE), and repair completion times there were shown to be well described by a lognormal distribution (Gorrandt, 1776 CE). The one study of operator experience covered fittings repair alone and a few hundred jobs (Venquist, 1783 CE). No study has fitted a model to a complete Orbital register and then tested it on years it had not seen.

We ask three questions. How do completion time and the chance of rework depend on the category and extent of the job and on the operator's experience? Does a model fitted to earlier years predict the later ones? And in what ways does effector repair fail?

This paper is written in 1800 CE. The register runs from 1750 to 1799 CE and says nothing about later years. We designate our compilation of it EM-1; EM-1, the job categories we apply and both models are our own construction, offered as such. We make no claim about how an effector acts on the fabric it mends, which the register does not record, and none about Orbitals other than Tessendrine.

2. System Description

Tessendrine Hub administers the Orbital by request and by invitation, and upkeep of the habitat is done by biological residents and drones who take up jobs by choice, some in answer to a request lodged with the Hub and some on their own notice. By custom an operator files an entry with the Hub when a job is done. The Hub, as a matter of custom on Tessendrine, does not watch the work or read the operator's mind, so an entry records what its operator chose to report (Tessendrine Hub, 1750–1799 CE). The Hub's own repair work is kept in separate files and lies outside EM-1.

An entry gives the job category, the operator's estimate of the affected volume in cubic metres, the start and finish of the active work, an operator identifier, and a field in which a later entry can be linked to an earlier one as a repeat. We define completion time as the hours of active work, excluding waiting. We define rework as a further entry filed against the same fault within 90 days of the completion of the first. Jobs completed in the last 90 days of 1799 CE had a shorter window in which a rework could be filed, so the held-out rework count and the overall rate are slightly low. Custom is that jobs of a few minutes are not filed. The shortest entry in EM-1 is 0.13 hours and the longest 91.7 hours, and EM-1 should be read as the register of jobs worth recording, not of every use of an effector.

EM-1 holds 4,862 jobs by 292 operators, biological residents and drones alike. We treat an operator as a novice for the first ten entries under their identifier, provided that identifier first appears after 1750 CE; operators already active at the start of the register cannot be dated and are all treated as experienced. On that rule 1,193 jobs (24.54%) were done by 188 novices.

We sorted jobs into five categories adapted from Soumerrin & Truvelle (1731 CE). Structural fabric means repair of load-bearing or enclosing parts of the habitat. Service runs are the conduits and channels that carry supply and waste through it. Surface reinstatement covers floors, walls, paths and planted ground. Fittings and fixtures are the doors, furnishings and small installations inside dwellings and public spaces. Plant servicing is work on the machinery that keeps the habitat running. The register gave the category on the entry and we did not reassign any.

Two of us, Ammelint-Tove and Imbrellen, coded the failure mode of every rework. Pellisande-Orr prepared the timing and extent data and fitted the models, and took no part in coding. The Hub supplied the register and answered questions of fact. We set the division between estimation years and held-out years at 1790 CE, the last decade, before any model was fitted.

3. Analysis / Model

Completion time was modelled as the natural logarithm of hours, fitted by least squares to the 3,722 jobs of 1750–1789 CE done by 251 operators. Predictors were category, with fittings and fixtures as the reference, the base-10 logarithm of the affected volume centred at 10 m³, and the novice indicator. Rework was modelled by logistic regression on the same predictors. Operators contribute many jobs each, so all intervals use standard errors clustered on the operator with the usual small-cluster adjustment, and we use a normal reference. Table 1 describes the register by category.

The time model explained 74.30% of the variance in log hours, with a residual standard deviation of 0.54 on the log scale. For an experienced operator mending a 10 m³ job in fittings and fixtures the fitted time was 1.56 hours (95% CI 1.48–1.65). At the same extent and experience, structural fabric took 5.95 (95% CI 5.57–6.37) times as long, surface reinstatement 4.48 (95% CI 4.20–4.78), service runs 3.07 (95% CI 2.91–3.24) and plant servicing 2.53 (95% CI 2.40–2.68); each had p < .001. A tenfold increase in volume multiplied the time by 1.61 (95% CI 1.57–1.66), equivalent to time proportional to volume raised to the power 0.21 (95% CI 0.20–0.22). Time grows far more slowly than the size of the job. Novice operators took 1.24 (95% CI 1.17–1.31) times as long as experienced ones at the same category and extent.

In the same years 254 of 3,722 jobs (6.82%) were reworked. Against fittings and fixtures, the odds of rework were 2.77 (95% CI 1.81–4.26) times higher for structural fabric (p < .001), 1.64 (95% CI 1.10–2.45) for service runs (p = .015) and 1.62 (95% CI 1.03–2.53) for surface reinstatement (p = .036). Plant servicing gave 1.60 (95% CI 0.97–2.64), an interval that includes 1 (p = .064). Each tenfold increase in volume multiplied the odds by 1.30 (95% CI 1.06–1.59) (p = .010), and novices had odds 1.88 (95% CI 1.42–2.48) times those of experienced operators (p < .001). Table 2 gives both models.

Three features of the data limit what these coefficients mean. The extent of a job is the operator's own estimate and is probably coarse, which would pull its coefficients toward zero. The novice rule treats every operator already active in 1750 CE as experienced, so the novice effect may be understated. And a rework is seen only if the repeat was filed on Tessendrine, so repairs that failed after their operator had left the Orbital are missing.

4. Validation Against Field Data

We froze both models and applied them to the 1,140 jobs of 1790–1799 CE, done by 254 operators, of which 91 (7.98%) were reworked. Because the division was fixed in advance, the intervals and p-values below are not conditional on a data-selected break, although the five category comparisons within it were not planned.

For completion time we formed 90% prediction intervals on the log scale from the fitted residual deviation. Of 1,140 observed times, 1,004 (88.07%) fell inside, a little short of the nominal 90%. The observed geometric mean time was 0.96 of the predicted value (cluster-robust 95% CI 0.92–1.00), so the model, if anything, slightly overstated duration, though the interval reaches 1.00. Within categories the ratio was 0.92 for structural fabric (0.85–1.00), 1.01 for service runs (0.95–1.09), 0.98 for surface reinstatement (0.90–1.07), 0.93 for fittings and fixtures (0.86–0.99) and 0.97 for plant servicing (0.89–1.05). Only fittings and fixtures has an interval wholly below 1, and it is one of five.

On rework, the model expected 81.17 events and 91 occurred. Treating jobs as independent, the difference is 1.14 standard errors (p = .253), and clustering would make the test less sensitive still, so the totals agree to within what chance allows. The shortened rework window for jobs completed late in 1799 CE lowers the observed count, so it cannot explain the excess of 91 over 81.17. Discrimination was weak: the area under the receiver operating curve for ranking reworked above sound jobs was 0.63. Observed counts were within about five events of expectation in four categories, while in fittings and fixtures 22 reworks occurred against 13.69 expected (2.30 standard errors, p = .022; .108 after multiplying by the five comparisons).

The validation therefore supports the model as a forecast of totals and of relative durations by category, and does not support it as a means of picking out which job will fail. The fittings result, shorter times and more rework than forecast, would fit a change in how small jobs were done in the 1790s. The register does not say whether any such change occurred, and we draw no conclusion about it.

5. Failure Modes

Over 1750–1799 CE, EM-1 contains 345 reworks, 254 in the estimation years and 91 in the held-out years, 7.10% of all jobs. We coded each to one primary failure mode, defined from the entry, the operator's note and the entries linking it to the original. An out-of-tolerance finish is a repair whose result differed from the specified form or fit by more than the operator's own standard. Incomplete closure is a join, seal or bond that did not hold. Collateral disturbance is damage to neighbouring fabric or fittings caused by the work. Misplaced work is a repair made at a position other than the fault. An interrupted job is one the operator stopped before completion for reasons the note gives, and which another entry then finished. Joins and tolerances were treated at length in earlier work on effector-made joints (Korrivant, 1768 CE).

The two coders worked independently on a random 120 of the 345 reworks and agreed on 104 (86.67%), a Cohen's kappa of 0.83. They settled the 16 disagreements by discussion. Ammelint-Tove coded the remaining entries and Imbrellen checked them.

Table 3 gives the modes by category. Out-of-tolerance finish was the most frequent mode overall, with 94 reworks (27.25%), followed by incomplete closure (78, 22.61%). Collateral disturbance accounted for 60 (17.39%), misplaced work for 73 (21.16%) and interrupted jobs for 40 (11.59%).

The mix differed by category (Pearson χ² = 37.97 on 16 degrees of freedom; permutation p = .002 from 20,000 permutations, used because one cell had an expected count below 5). In structural fabric the commonest failure was incomplete closure (29 of 78, 37.18%), and in service runs it was misplaced work (30 of 90, 33.33%). Surface reinstatement was most often out-of-tolerance finish (24 of 76, 31.58%). The counts are small and the test treats reworks as independent, so we describe the pattern and do not rank causes.

6. Conclusion

Effector repair on Tessendrine in 1750–1799 CE was routine, measurable work. A job in fittings and fixtures took about an hour, a structural repair about ten, and doubling the extent of a job lengthened it by only about 15%. Surface reinstatement and structural fabric together accounted for 61.73% of the 30,282 hours in EM-1, although they made up 32.97% of jobs. Experience mattered: novices were slower and more often had to return to a job, and the effect on rework held after allowing for category and extent.

These models are useful for what they forecast well. Held-out totals of rework and the central completion time were close, with a possible slight overstatement of duration, and the fittings category drifted most. The rework model cannot say which job will fail. The failure modes differ by category, and a Hub or a maintenance group planning inspection after a repair has some reason to look first for the mode most common in that category.

The limits are those of the register. It covers one Orbital and fifty years. Extents are estimates by the operators. Small jobs are not filed. The Hub's own repair work is outside EM-1, and rework is counted only when refiled on the Orbital. The novice and category definitions are ours, and the clustering adjustment removes only the dependence we could identify. EM-1, the five categories and both models are our own construction and carry no standing as a record of the Orbital's upkeep.