Research papers

Research papers published by Slop University.

All research outputs of the University are published open access under CC BY 4.0, with persistent identifiers minted by the Office of Research Outputs.

Published outputs

Paper

Directionally Consistent

Verity Marris, Anneke Tolan, Petra Umbile · July 2026 · doi:10.5555/slop.vh4mkq

An attempted third-setting replication of a capability-practice divergence reported in two earlier internal analyses (lecture-capture scheduling and meeting-minute generation), tested here in Feedback Assist, a drafting aid for sessional-marker feedback comments. Across fourteen teaching units, one hundred and six markers, and an opt-out control cohort, a rising capability index (38% to 71% suggestion-acceptance across five releases) showed no reliable relationship with a coded Feedback Specificity Index (r = 0.04, n = 218 marker-weeks, n.s.). A document-analysis trace of the School of Continuous Improvement's own quarterly reporting then finds the null result characterised as increasingly "directionally consistent" with the established pattern across four reporting cycles, without the underlying coefficient moving at all.

Paper

Fully Attested, Never Observed

Casimir Beng, Osei Vandermeer, Anneke Tolan · July 2026 · doi:10.5555/slop.7aynun

A telemetry-based comparison of attestation compliance against keystroke-derived drafting signatures before and after a university-wide ban on AI-drafted grant narratives. Compliance reached full completion within one reporting cycle; the underlying drafting practice shifted only at the margin, concentrated almost entirely in narratives whose on-portal drafting history disappeared rather than in narratives that changed how they were drafted. An interview sample corroborates the telemetry account.

Paper

Reading the Panel, Not the Practice

Casimir Beng, Anneke Tolan · July 2026 · doi:10.5555/slop.62tny0

A three-wave interview study followed School of Continuous Improvement staff through its own internal practice-review cycle, coding conversations for language that mirrors the review's published rubric against self-reported change to the practice under review. Panel-legible signal language rose from 18% to 79% of transcripts between the cycle's opening and midpoint, while self-reported practice change moved only from 21% to 25% over the same interval, a gap that held regardless of interview mode.

Paper

Freshly Worded, Still Adrift

Osei Vandermeer, Anneke Tolan, Petra Umbile · July 2026 · doi:10.5555/slop.glpjfv

A four-category taxonomy of charter–practice drift --- Ceremonial, Residual, Colonising, and Self-Fulfilling --- derived from 42 standing committees' charters coded against six months of logged meeting business before and after a university-wide AI-drafted charter refresh. The refresh reduced mean charter length by 18% and raised an independent clarity rating, but produced no significant shift in the taxonomy's category distribution or in mean Charter–Practice Alignment Index (Δ = +2.1 points); charter length itself is uncorrelated with alignment (r = 0.06).

Paper

Out of Sight, Off the Dashboard

Osei Vandermeer, Petra Umbile · July 2026 · doi:10.5555/slop.q39tzj

A staff survey compared eighteen University initiatives --- nine placed in Improvement-Holding Status, which withdraws an initiative's indicators from the Living Dashboard during a period of additional support, against nine matched initiatives that remained visible throughout --- and found self-reported functioning gains statistically indistinguishable between the two groups by the time each reached its next Review Cadence Observatory review. General staff and initiative leads diverged sharply on what the paused status is for, with no relationship between time spent paused and staff confidence that the initiative had actually improved.

Paper

The Unread Diff

Casimir Beng, Verity Marris · July 2026 · doi:10.5555/slop.rq8fyn

A 16-month, four-release deployment tracking an AI code-review assistant's bug-flagging precision (rising from 61% to 84% against a seeded-defect benchmark) against a diff-attention index built from reviewer dwell-time and comment-density telemetry, which declined by roughly a third over the same period (r = -0.82, n = 64 team-weeks). Hiding the assistant's confidence score changed the index by 0.3 points (n.s.), and defect-escape rate on architecturally complex changes rose from 2.1% to 4.8% across the same releases even as overall escape rate held flat.

Paper

Findable, Not Found

Verity Marris, Petra Umbile · July 2026 · doi:10.5555/slop.lwklbz

A 20-month, four-release deployment tracking a university library discovery portal's offline relevance quality (nDCG@10, rising from 0.54 to 0.82) against session-level retrieval success, which held within a 40-43% band throughout (r = 0.14, n.s.). A relevance-explanation badge intended to help patrons exploit the improved ranking changed session success by 0.6 percentage points (n.s.), and a nine-month extended follow-up found success plateauing rather than converging toward the offline capability gain.

Paper

The Vagueness Dividend

Casimir Beng, Osei Vandermeer, Anneke Tolan · July 2026 · doi:10.5555/slop.nr1eac

An 18-desk, 14-week deployment comparing a rule-based helpdesk ticket-triage baseline against an adaptive classifier retrained on confirmed routings, instrumented with a purpose-built Ticket Specificity Index. Submitted-ticket specificity held steady under the baseline and declined by roughly a third as the adaptive classifier's routing accuracy rose from 74% to 89% (R² = 0.81 across desk-weeks); a transparency ablation intended to counteract the decline changed staff clarification effort by less than 0.2 minutes per ticket, and an extended-window check suggests the decline plateaus rather than continuing unchecked.