Stafford Brief — 2026-09-14
⚠️ OPS — two things before the content.
- Stafford last ran on 2026-09-10. No brief exists for 09-11, 09-12 or 09-13. The research-evidence scan ran all three days and this repo's judgment layer did not consume them. Today's brief covers today; the three-day backlog of pattern, challenge and asset implications is not caught up here and is listed as a queued item in Routing.
- This brief is mostly one thing, and it is a correction to today's scan. The scan's own brief is excellent and I am not restating it — read it for Ash, Koppel and the CPOE case. What follows is the part it got wrong and what to do instead.
📖 TODAY'S READ
Don't publish the sentence today's scan told you to publish. It's wrong, and the right version is better.
This morning's scan brief made a recommendation in its Thinker section: Sittig & Singh's eight-dimension model of why health technology fails inside hospitals has no dimension about whether an effort sustains force after launch, and — verbatim — "Brandon should publish that observation."
The observation is true of that one model's eight dimensions. It is false of the field, and a reviewer needs one sentence to say so. There is a second framework in the same literature, from Oxford, called NASSS — it stands for Non-adoption, Abandonment, Scale-up, Spread and Sustainability. It has seven domains. The seventh is, word for word, "embedding and adaptation over time." It asks how much scope there is to adapt a technology and a service together over time, and how resilient the organization is when things go wrong. It was built from 400 hours of watching people work, 165 interviews and 200 documents across more than twenty organizations over three years.
So: publish "they have no dimension for sustained force" and the first person who knows this literature answers it in a line, and the rest of your argument loses standing alongside it.
Here is the version that survives, and it's a stronger claim than the one it replaces. Both of these frameworks — and, by NASSS's own count, 28 others catalogued in this space — organize the sustainment failure around abandonment: the technology gets dropped, or never becomes normal practice. Ash's 176-hospital survey, which today's scan correctly made the centerpiece, measures the exact opposite case: systems fully embedded in daily work, five years of use, universal adoption — and organizational damage that did not decline at all with years of use.
That is not abandonment. It is not failure to scale. It is a deployment that completely succeeded and kept hurting the whole time — and neither framework has a box for it. That's what Momentum Mirage names.
So what this changes: the differentiation paragraph Brandon has never written now writes itself, and it cites the tradition rather than pretending it isn't there. Asset #8 below. Full entry: knowledge-base/greenhalgh-nasss-nonadoption-abandonment-framework-2017-11.md.
🔭 THINKER TO WATCH
Trisha Greenhalgh, Professor of Primary Care Health Sciences, University of Oxford — the competitor today's scan didn't see.
The scan named Sittig, Singh, Ash and Koppel. Greenhalgh is the harder one for Brandon, for three reasons. Her framework is newer (2017, not 2010), so "that literature is old" doesn't work. It is built on ethnography — people watched at work for three years — which is the same kind of evidence Brandon's own Open Question this morning said nobody is gathering for AI. And she is not confined to hospitals: NASSS covers health and care technologies broadly, which blunts the "one industry" differentiation.
Where Brandon genuinely differs, and it's narrow and defensible: NASSS asks will the organization keep this technology? Five Breakpoints asks will the organization keep paying the cost of having kept it? Greenhalgh's domain 7 is about the capacity to adapt. It is not about force decaying while everything looks fine — and Ash's no-attenuation result is the case that separates them.
🏢 CASE IN THE WILD
Nothing to add today. The scan's CPOE case — 176 hospitals, all eight damage categories, Koppel's 22 error pathways from an integration that was assumed and never built — is the case, it's correct, and repeating it here would be filler. Read it there.
⚙️ FOUR FORCES CONCLUSION
Momentum, and today is the first day it has done competitive work rather than descriptive work.
The scan used Ash's no-attenuation finding to make a point about Capability and time. Fine. But the same finding is the only thing in this base that distinguishes the Four Forces from a serious rival framework, and it does it precisely on Momentum.
Brandon defines Momentum as an effort's ability to convert progress into reinforcement — not merely that activity persists. Every competing framework here treats sustained use as the success condition. Ash shows sustained use with no reinforcement arriving for five years: the organization didn't grow out of the problem, it habituated to it. A framework whose failure category is "they stopped using it" cannot describe that. Momentum can.
So what this changes: Momentum stops being the softest of the four Forces to defend and becomes the one that earns the framework its distinctiveness. Any differentiation writing should lead with it.
💭 OPEN QUESTION
If the organizational damage never declines with years of use, what exactly is the business case for enduring it — and has anyone ever been asked to make it?
Ash's hospitals ran a trade: better medication safety, purchased with permanent organizational friction. That may well have been the right trade — CPOE was backed by a real patient-safety case. But nobody wrote it down as a trade. The friction shows up in the literature as an "unintended consequence," which is the language you use for something you didn't decide.
Every agentic deployment now being procured is making the same trade and calling it a transition cost — a temporary tax that adaptation will pay off. Ash's data say that for the previous technology of this kind, it never did. So the question Brandon can put to any client, and which nothing in this base can answer: if the friction is permanent rather than transitional, does the deal still clear? Nobody is being asked, because everybody is modelling it as temporary.
🥊 CHALLENGE
Logged today as the base's largest standing positioning exposure — knowledge/thesis-challenges.md, "The sociotechnical priority challenge," OPEN.
At full strength: Brandon's framework answers why does working technology fail inside an organization? That question has a forty-year research tradition with 28 catalogued frameworks, empirically derived categories from 245 observed cases (Campbell et al. 2006), and continuous use since. A well-read critic can argue Five Breakpoints re-derives sociotechnical systems theory in management language.
That criticism is partly correct, and today's scan said so. What it missed is that the exposure is bigger than one 2010 model, and that the defence is not "we're different" — it's three specific differences, of which only the second is strong: prediction versus taxonomy; the failure construct (abandonment versus embedded-and-still-costly); and cross-industry scope conceded honestly in both directions.
What it would take for the objection to win: someone showing NASSS domain 7 plus the Sittig–Singh organization dimension jointly cover what the Four Forces add. Nobody has tried, because nobody outside this base has put the two literatures side by side. That is a window, not just a risk.
🧱 ASSET SUGGESTION
"It Worked, and It Kept Hurting" — one essay, sources already read.
The argument in two sentences: forty years of research on why good technology fails inside organizations is organized around the technology being dropped; Brandon's framework names the case that tradition has no category for — the deployment that succeeds completely, embeds fully, runs for years, and leaves the damage undiminished throughout. Clinical informatics produced the measurement of exactly that case and had nowhere to file it.
It's the only item in the ledger that acquires standing against a tradition rather than adding evidence inside his own, and it converts today's challenge from a referee's opening paragraph into a published asset. Full spec, including the one blocking dependency: knowledge/asset-suggestions.md #8.
🎓 SPARRING NOTE
(First this week; the 08-31 and 09-01 notes remain unanswered and are superseded, not stacked.)
Growth edge #1 has been about grounding Claim 1 in a literature before a referee does it for you. Today is the same edge pointed at a negative claim, which is the harder case.
The prompt, and it's one paragraph: write the sentence that concedes the sociotechnical tradition and states what's left over. Constraint — you may not use the words "but," "however," or "unlike." Concession first, residue second, and the residue has to be a case (Ash's embedded-and-undiminished hospitals), not an adjective.
Why this difficulty. The three earlier hooks on this edge all asked you to go read something. This one asks you to write the hardest paragraph in the paper, and you already have every source it needs. If the paragraph comes out easily, the differentiation is real. If it doesn't, that's the finding — and better learned now than in review.
Skill-inventory note — fourth instance of one habit, and a new direction for it. 09-02: check a decomposition's axis is the claim's axis. 09-04: check an instrument's unit of analysis is the claim's object. 09-10: check the claim says what its own literature says. Today: a claim about what a literature does not contain must be checked against the literature next to the one you read. The scan read its source correctly. The error was in the neighbourhood it didn't search.
SOURCES & THREADS
Exploratory query (logged, knowledge/query-log.md). Not a case-file gap — verification of a publication recommendation the scan made the same morning, which is Stafford's lane under CLAUDE.md operating rule 2 and hard rule 10. Coverage grep first (sittig, sociotechnical, NASSS, Greenhalgh, non-adoption across all 529 KB files in both repos — zero hits on NASSS/Greenhalgh), then two searches: the Sittig–Singh primary, and the adjacent sustainability literature the criticism would have to survive. Yield: 1 KB file, 1 challenge, 1 asset, 1 recommendation overturned before print.
Threads: none due, none advanced. Nearest is Team-agent interaction at 2026-09-23. The scan opened Does clinical informatics turn its own instrument on AI? today — I'd propose one amendment to it, in Routing.
Candidates: one proposed, in Routing. Greenhalgh's Oxford group (Wherton, Papoutsi, Shaw) is the field's active ethnographic programme and is absent from the field map entirely.
Access, and it's a degradation worth recording. Eleventh consecutive research day requiring a workaround: PubMed cookie-walled again, jmir.org returned empty, Springer redirected to authentication, and the UTHealth self-hosted Sittig PDF yielded no extractable text — because neither PyMuPDF nor pypdf is installed in this container. The 09-04 lesson ("chase the self-hosted PDF, extract it locally") is currently unavailable to Stafford, and that lesson has been load-bearing four times in two weeks. Flagged in Routing as an ops item.
Routing
personal — the item that needs Brandon, and it needs him before he writes
Do not publish "Sittig & Singh has no Momentum dimension." Today's scan recommended it in good faith off a correct read of one paper. NASSS domain 7 is verbatim "embedding and adaptation over time" and refutes it in a line. Publish the residue claim instead — embedded, five years, undiminished harm, which neither framework classifies. Challenge logged; asset #8 specs the essay.
strategy — a decision expiring in 14 days, and only Brandon can make it
The scan reported that the 08-31 agent-intrusion publication window closes ~2026-09-28 and that today was "the last scan on which it is honestly actionable" — sixteenth consecutive scan unwritten. The scan can only keep reporting that. This is a decision, not a finding: write it in the next fortnight or record it as lost. The 08-23 Accenture window (~2026-10-05, thirteenth scan unwritten) is behind it. The pipeline has now produced two of its strongest durable assets in four days, neither of which can have a window, while two dated windows expire unwritten — that is a capacity allocation question, and it is his.
governance — queued, not done
Three days of scan output (09-11, 09-12, 09-13) have not been through this repo's judgment layer. Known from today's reading: the 09-11 Bloom/Garicano/Sadun/Van Reenen entry and the 09-13 clinical-override literature both landed without a Stafford pass. Next run should clear the backlog before taking a new exploratory slot.
ops — a tooling regression that costs yield
pip install pymupdf (or pypdf) in the Stafford container. Four of the last ten high-yield chases went through local PDF extraction of a self-hosted paper; that route is dead right now and today's Sittig full text is the first confirmed casualty.
STAGE-CANDIDATE — for the next research-evidence scan run
python3 build/add-research-json.py --json '{
"title": "Beyond Adoption: A New Framework for Theorizing and Evaluating Nonadoption, Abandonment, and Challenges to the Scale-Up, Spread, and Sustainability of Health and Care Technologies",
"publisher": "Journal of Medical Internet Research",
"url": "https://www.jmir.org/2017/11/e367/",
"published_date": "2017-11-01",
"credibility": "high",
"instrument": "Data source: primary qualitative fieldwork — six technology-supported programmes across 20+ organizations, plus a hermeneutic systematic review of 28 prior frameworks. Inference method: longitudinal ethnography and action research, non-statistical; framework construction, no effect estimates. Observation window: up to 3 years per programme, published 2017. Not an AI instrument and not a quantitative one — enters on the previous-shock footing and may never be cited as measuring an effect size.",
"breakpoints": [
{"name": "technology_illusion", "rationale": "Programmes characterized by complexity across multiple NASSS domains rarely, if ever, became mainstreamed regardless of the technology'\''s own merit — deploying a working technology onto unresolved organizational conditions does not produce adoption."},
{"name": "momentum_mirage", "rationale": "Domain 7, verbatim '\''embedding and adaptation over time,'\'' is the sociotechnical tradition'\''s sustainment construct, and it organizes that failure around abandonment rather than around persistent effort producing no movement — which is the distinction Momentum Mirage names."}
],
"forces": [
{"name": "Momentum", "rationale": "Domain 7 asks directly how much scope exists for adapting and coevolving the technology and the service over time, and how resilient the organization is to unforeseen eventualities."},
{"name": "Capability", "rationale": "The organization domain and question 7B both assess organizational capacity to absorb and keep absorbing a change, scored qualitatively rather than on an instrument."}
],
"key_findings": [
"Seven domains, the seventh verbatim: embedding and adaptation over time",
"Hermeneutic systematic review identified 28 previous technology-implementation frameworks",
"Programmes complex across multiple domains rarely, if ever, became mainstreamed",
"Built on 400+ hours of ethnographic observation, 165 semistructured interviews, 200 documents",
"Six programmes studied for up to 3 years across more than 20 organizations"
],
"sample_size": 6
}'
PROPOSED-UPDATE — research-evidence knowledge/field-map.md
Trisha Greenhalgh (University of Oxford, Primary Care Health Sciences) — with Joy Wherton, Chrysanthi Papoutsi and Sara Shaw. Author of the NASSS framework (JMIR 2017;19(11):e367). The most direct competing framework in this base and newer than Sittig & Singh. Seven domains, the seventh "embedding and adaptation over time"; built on longitudinal ethnography (400+ hours, 165 interviews, 20+ organizations), and its own review counted 28 prior frameworks in this space. Brandon's differentiation is narrow and must be stated precisely: NASSS organizes sustainment failure around abandonment; Momentum Mirage names embedded, persistent effort producing no movement — the case Ash's own 176-hospital survey measured and this tradition cannot classify. Do not claim the tradition lacks a sustainment dimension. See
stafford-researchknowledge/thesis-challenges.md— the sociotechnical priority challenge.
PROPOSED-UPDATE — research-evidence knowledge/source-candidates.md
The Oxford NASSS programme (Greenhalgh / Wherton / Papoutsi / Shaw; JMIR, BMC Medicine, Journal of Medical Internet Research). First seen 2026-09-14, surfaced by the Stafford verification of the 09-14 scan's Sittig & Singh recommendation. Signal count 1. Promotion condition: a second find in which this group applies NASSS or its successor to an AI or agentic deployment with a disclosed sample — which would simultaneously satisfy criterion (a) of the clinical-informatics thread opened today. Not a second signal: any further paper about the 2017 framework, or any of the 28 catalogued predecessors — the tradition is one territory, not a supply of independent sources.
PROPOSED-UPDATE — research-evidence knowledge/story-threads.md, thread "Does clinical informatics turn its own unintended-consequences instrument on AI?"
Amend criterion (a), 2026-09-14 (Stafford). Add NASSS (Greenhalgh et al., JMIR 2017;19(11):e367) alongside the Campbell/Ash taxonomy and the Sittig–Singh model as a qualifying instrument: any peer-reviewed study applying NASSS to an AI deployment with a disclosed sample meets (a). Rationale: NASSS is the tradition's newer and broader framework and its domain 7 is the sustainment construct this thread's Momentum interest actually turns on; a thread watching whether this field turns its instruments on AI should not watch only the two oldest ones.
PROPOSED-UPDATE — research-evidence knowledge/emerging-patterns.md, pattern "The Previous-Shock Archive"
Caution 3, added 2026-09-14 (Stafford). The previous-shock literatures come with their own frameworks, not only their own findings, and those frameworks are competitors. Pattern members have so far been mined for evidence and not read for priority: Greenhalgh's 2017 review counted 28 catalogued frameworks answering the question Five Breakpoints answers, and none was in this base until today. Standing rule for this pattern: when a member arrives from a new literature, record what that literature's own organizing framework is before recording what its data show — the first is the positioning risk and the second is the yield, and this pattern has been collecting only the second.