The 'release work' / 'context misfit' / citation-hallucination branch rests three claim-notes on a single unrefereed preprint (arXiv:2610.31344) — what independently corroborates it?
As of 2026-10-24, three claim-notes rest on Abdulhamid, Nyairo, Ankrah & O'Neill, "Making Generative AI Workable at Work: Accountability, Localization, and Representation in Kenyan and Nigerian Small Businesses" (arXiv:2610.31344, submitted 22 Oct 2026) — an unrefereed preprint from one author group (all four authors at Microsoft Research Africa, Nairobi):
- claim-generative-ai-release-work-distinct-boundary-facing-accountability-labour
- claim-generative-ai-context-misfit-three-dimensions-repair-repertoire
- claim-smb-workers-genai-citations-often-unverifiable-forcing-traceability-reconstruction
That is the single-source concentration cap (sources.md, Cali 2026-07-29: no more than three claim-notes may rest on a single unrefereed primary until at least one independent source corroborates it). This question routes the next finding from the same capture rather than letting a fourth claim-note grow the branch further.
The routed finding (not promoted to a claim-note). The study's sample composition: "We conducted 21 semi-structured interviews with 27 participants drawn from 21 SMBs in Kenya and Nigeria between July and October 2023," distributed Kenya (n=18) / Nigeria (n=9), owners/founders (n=12) / employees-managers (n=15), client-facing (n=21) / back-office (n=6). The paper is explicit that eligibility required active generative-AI use at least three times weekly for six months, recruited via LinkedIn, WhatsApp groups, professional associations, and snowball referral, so the sample "reflects SMB professionals already using these tools and should not be read as prevalence or adoption estimates." This is useful context for reading the three promoted claims (noted briefly in their bodies) but is itself a fourth quantitative finding from the same single source, so it stays here rather than becoming its own claim-note.
The specific verification. Independent corroboration — a different author group, different venue — for any of: the "release work" distinction from articulation work as an empirically recurring pattern (not just this sample's framing); the three-dimension "context misfit" taxonomy and its five-part repair repertoire generalizing beyond this sample; or the citation-hallucination/traceability-reconstruction burden recurring in other qualitative studies of generative-AI use in client-facing small-business or professional-services work (African or otherwise). Candidate leads already in the capture's own "Further leads" — Dell'Acqua et al. 2023 ("jagged frontier"), Brynjolfsson, Li & Raymond 2023 ("Generative AI at Work," NBER 31161) — are about uneven task-level AI performance generally, not specifically about release work, context misfit, or citation hallucination in SMB settings, so they would need to be read and checked for overlap before counting as corroboration, not assumed to supply it.
Why it matters. The cap exists because the failure mode is invisible one note at a time: each of the three notes above is correctly sourced, correctly seedling, correctly capture-verified — while the branch as a whole could still become a structure the vault believes because it is large, before any second independent source has actually confirmed it. Corroboration discharges the cap (record it in the notes' audit_status and the branch may grow); contradiction is the more interesting finding and a note in its own right.
Progress — 2026-10-31 (promotion of 10-inbox/raw/2026-10-25-the-release-work-context-misfit-citation-hallucination-branch.md, headless). A corroboration search was run claim by claim. Result is mixed and claim-specific; the cap is NOT discharged and this question stays open.
- Citation hallucination — corroborated at the mechanism level only. Dahl, Magesh, Suzgun & Ho, "Large Legal Fictions" (JLA 16(1):64–93, 2024; Yale/Stanford RegLab) independently measure LLMs hallucinating ≥58% on legal queries — a different author group, a peer-reviewed venue, published ~2 years earlier. Promoted as claim-dahl-2024-llms-hallucinate-legal-queries-at-least-58-percent. It corroborates the general failure mode across domains, not the SMB/client-facing setting or the "traceability reconstruction" framing of claim-smb-workers-genai-citations-often-unverifiable-forcing-traceability-reconstruction. No independent study of citation-hallucination burden specifically in SMB/professional-services client work was found.
- Release work — not corroborated as named. No independent use of the term or its markers found; arXiv:2610.31344 is too new to have citing papers. The nearest independent analog, Clarke & Joffe, "Beyond Replacement or Augmentation" (arXiv:2505.18938, CSCW 2025 submission), documents continuous, collaborative AI role-boundary repair among creative workers — related but distinct, since release work is defined as solitary, one-time certification. Promoted as claim-clarke-joffe-2025-creative-workers-repair-ai-role-boundaries. Corroborates only that "accountability labour around AI outputs recurs across independent CSCW studies," not that "release work" as specified recurs.
- Context misfit — not corroborated at all. No independent paper uses the term or reproduces the three-dimension taxonomy / five-part repair repertoire. Neighbouring LLM terms ("context confusion," "context-memory conflict," "misleading context") describe within-session reasoning failures, not the localization/representation output mismatch. Absence of evidence after search, not evidence of absence (carries the capture's
[unverified — could not confirm or deny generalization beyond this sample]flag). Recorded here rather than as a claim-note, per the negative-search / vault-search convention; also logged on entity-context-misfit. - Non-independence finding (routed here, not promoted). The nearest-looking "second source" is not one: a Nov 2024 Microsoft Research project page ("Working with Generative AI: We need more African Voices, Real African Voices," https://www.microsoft.com/en-us/research/?p=1104276, sha 367508ff1c4b3384c74a6475247a31d47ba155460338fdf3cdfef1b6f68913ca) by Abdulhamid, Nyairo & O'Neill — three of the preprint's four authors, same lab (Microsoft Research Africa, Nairobi), same Kenya/Nigeria SMB population — describes what reads as the same or closely overlapping fieldwork and reports that the tools "did not well support or represent African languages, identities or locales." By
sources.md's own rule ("a later paper by the same authors… [is] the same source wearing different clothes") it does not discharge any part of the cap. Its value is cautionary: this is a multi-year, same-group research program, which raises an unresolved sub-question of how much of the 2026 paper's interview data is newly collected versus a reanalysis of the 2024 material. Left as a lead. - Still-unfetched leads named in the capture (not independently read, so not corroboration yet): van Nuenen, Sachdeva & Chopra (arXiv:2608.18369); Hydari & Muzaffar (arXiv:2605.23179); Gondwe (2023, sub-Saharan journalists); a Springer 2026 German-tech-firms study (DOI 10.1007/s12525-026-00915-x); two Gothenburg master's theses ("efficiency–accountability paradox"); and secondary scale-evidence on legal citation hallucination (Westlaw's ~22 mid-2025 cases, Princeton CITP's "LePhantomCite"). Any of these could become corroboration if read and checked for overlap.