Persistence and Interference of Sliver-Acquired Skills After Removal: A Randomised Staged-Use Study of 132 Participants, Berne and Chiba, 2041–2043

1. Introduction

Microsofts are skill slivers. A sliver plugs in behind the ear, and its wearer performs a skill, a language, a repair trade or a clerical procedure, that the wearer did not previously possess. The common understanding of the technology is that the skill belongs to the sliver. When the sliver is withdrawn, so is the competence, and the wearer returns to the baseline from which they began. Hiring practices, licensing rules for slivers in regulated trades and the pricing of short-term skill rental are commonly said to rest on this assumption.

The assumption has seldom been examined. Manufacturers' bulletins describe production quality and wearer tolerance (Brennwald, 2041; Aebischer, 2042) and a descriptive framework treats the sliver user's learning curve as continuous with ordinary learning (Egloff, 2040), but no study has measured performance after removal against a pre-use baseline. Human-factors observations of assisted performance in industrial settings report that some users retain part of what they were helped to do (Hollister, 2038), and a modelling proposal predicts a retention curve of a specific shape for assisted skills (Sethuraman, 2041). Neither tested sliver users.

Practical stakes lie in two directions. If skills persist, a worker who rents a sliver for a contract keeps something of value, and a regulator who treats sliver-assisted work as unqualified work may be counting a competence the worker later holds. If skills do not persist, the extra cost of repeated sliver rental is justified. A second question concerns interference. Many wearers move from one sliver to another within weeks, and if a later sliver erodes what an earlier one left behind, rotation schedules matter.

We report a randomised staged use-and-removal study. We ask how much of the performance gain from a sliver remains twelve weeks after removal, and whether wearing two further slivers in the interval changes that retention. We write in 2044 and report participants studied between 2041 and 2043.

2. Methods

Participants and allocation. We enrolled 132 adult volunteers in Berne and Chiba under the research agreement and consent protocol of the Maas Biolabs applied programme (Maas Biolabs Applied Biosciences, 2041). Each volunteer received a neurological examination before enrolment and again after each removal, using the screening protocol of the Chiba clinic (Kanzawa-Lind, 2040). Each was assigned a target skill according to the availability of a matching sliver: language (41 participants), mechanical repair (62) or procedural and clerical work (29). Participants were assigned at random to the single-sliver condition (66) or the sequence condition (66) from one list of 132 places, without stratification by skill or site, so the skill mix differs somewhat between conditions (language 18 single and 23 sequence; mechanical repair 31 single and 31 sequence; procedural and clerical work 17 single and 12 sequence). Allocation was concealed from the staff who scored the batteries.

Procedure. Participants completed a baseline battery with no sliver. They then wore the target-skill sliver for six weeks, with a second battery at the end of that period while the sliver remained in place. The sliver was then removed. A third battery followed one week after removal. The one-week battery was taken before any further sliver was worn. In the single condition, participants then wore no sliver. In the sequence condition, participants wore a sliver for a different skill from week 1 to week 5 after removal and a third sliver, for another skill, from week 5 to week 9, with both removed by week 9. The final battery was taken in both conditions twelve weeks after removal of the target sliver, three weeks after the last sequence-condition sliver was removed. The sequence condition therefore differs from the single condition both in having worn other slivers and in having worn a sliver more recently. We did not describe or examine the slivers' internal construction, which the manufacturer retained.

Outcome. Each battery presented a fixed set of task items in the target skill, scored for errors by assessors blind to condition. Parallel forms of 60, 80 or 100 items were assigned at random to occasions, so the number of items attempted varies and is used as the exposure. The outcome is the error count per battery. Over all 528 batteries, participants attempted 41,940 items and made 9,267 errors.

Analysis. Counts of errors with a known number of opportunities, with heterogeneity beyond the Poisson expectation, call for count regression. We fitted a negative binomial model with a logarithmic items offset (Lindqvist-Ruiz, 2039). The model contained occasion, condition, their interactions and target skill. Because each participant contributed four batteries, standard errors were clustered on the participant and intervals use the t distribution with 131 degrees of freedom. Results are incidence rate ratios (IRR), the ratio of error rates per item between two occasions or conditions. An IRR below 1 means fewer errors. The estimated dispersion parameter was 0.17. For description we also report crude error rates per 100 items, pooled over participants, and a crude share of the sliver-period gain still present twelve weeks after removal, computed as (baseline rate − twelve-week rate) divided by (baseline rate − in-place rate).

3. Results

All 132 participants completed all four batteries. At baseline the crude error rate was 41.0 per 100 items in the single condition and 44.0 in the sequence condition (Table 1), and the model-based ratio of the two was 1.08 (95% CI 0.93–1.26; p = .334), so the conditions started alike. The skill terms were not detectably different from zero (mechanical repair against language IRR 1.04, 95% CI 0.89–1.23; procedural and clerical against language 0.81, 95% CI 0.65–1.02).

With the sliver in place, errors fell sharply. In the single condition the error rate was 0.21 of baseline (95% CI 0.19–0.24), and in the sequence condition 0.20 (95% CI 0.18–0.22). One week after removal, errors had risen but remained far below baseline: 0.27 of baseline in the single condition (95% CI 0.25–0.31). The sequence condition did not differ detectably from the single condition at this point (ratio 1.03, 95% CI 0.89–1.19), as expected, since the one-week battery was taken before any extra sliver was worn.

Twelve weeks after removal, the single condition's error rate was 0.51 of baseline (95% CI 0.47–0.56), and the sequence condition's was 0.72 (95% CI 0.67–0.78). In both conditions errors remained below baseline, and in both they had risen from the one-week level (Table 1); in the single condition the twelve-week rate was 1.87 times the one-week rate (95% CI 1.67–2.09). At twelve weeks, the sequence condition's error rate was 1.52 times that of the single condition (95% CI 1.29–1.79; p < .001).

Expressed as a crude share of the gain, computed from the pooled rates in Table 1, 64% of the improvement achieved with the sliver in place was still present twelve weeks after removal in the single condition, against 35% in the sequence condition.

4. Discussion

Two findings stand out. First, error rates one week and twelve weeks after removal were both below baseline, and in the single condition roughly 64% of the crude gain remained at twelve weeks. This is compatible with persistence of sliver-acquired skill, but we had no group tested on the same batteries without ever wearing a sliver, so some of the improvement over baseline may reflect familiarity with the items and format and not the sliver. Second, the persistence depended on what the wearer did afterwards. Wearing two further slivers was followed by a higher error rate at twelve weeks than wearing none, by a factor of about 1.52. The conditions also differed in how recently a sliver had been worn.

The data do not say why. The most direct reading is interference, in which later slivers displace what the earlier one left behind. Other explanations remain open. Participants who wore three slivers had less time and occasion to use the target skill in daily life, and ordinary rehearsal may sustain retained skill. Persistence itself could reflect learning by the wearer during use, a residual effect of the sliver, practice on the batteries, or a mixture, and our design cannot tell these apart. We did not measure daily use of the target skill, and the study does not isolate mechanism.

Error rates also rose between one week and twelve weeks in both conditions. In the single condition the twelve-week rate was about 1.87 times the one-week rate, so error rates were still rising between the two postremoval batteries. With two postremoval time points we cannot say whether they would settle at a plateau above baseline or approach baseline over a longer period. The twelve-week horizon is the principal reason to treat our retained share as a snapshot and not a curve.

For practice, the findings support two cautious inferences. A regulator who treats sliver-assisted performance as wholly separable from the wearer's own competence is assuming something these data do not support, though the unslivered comparison needed to be sure is missing. An employer who rotates wearers rapidly through different slivers may be reducing what each leaves behind.

5. Limitations

The participants were volunteers who accepted removal of a sliver and repeated testing, and the sample of 132 spans three skills, one of which (procedural and clerical work) is represented by only 29 participants. Retention may differ across skills, and we did not have the numbers to analyse each skill separately. The model includes skill as a main effect only, and does not model the Berne and Chiba sites, so differences between sites are absorbed into the participant clusters. Participants completed all batteries, which probably reflects a stipend schedule tied to attendance and may not describe less closely supervised wearers.

Task batteries are laboratory measures. They test the skill in standard items and not in the settings where wearers use it. The error count does not capture speed, and a wearer who works more slowly after removal could show the same error rate. Assessors were blind to condition but could not be blind to the occasion, since baseline and in-place performance differ visibly.

The first two authors are employed by the Maas Biolabs applied programme, which supplied the slivers under the research agreement. The third author, at the Chiba clinic, ran the neurological examinations and did not take part in assigning conditions. The analysis plan was fixed before the twelve-week batteries were scored, but readers should weigh the commercial interest in the finding that skill persists. We did not follow participants beyond twelve weeks, and the examinations recorded no persistent neurological change after removal.