Stafford

← 2026-10-03 · all briefs · 2026-10-05 →

Stafford Brief — 2026-10-04

Sunday. Short version: the scan had a very good morning, and the two things it could not do both turned out to cut against us. The strongest finding today is evidence that the framework's remedy may not do the job the framework implies it does — and separately, the consulting product this morning's brief proposed has existed as a free federal standard since 2003.

The 12:00 scan's brief is strong and this one does not repeat it. Government audit as an instrument genre is a real acquisition, the GAO and Uber entries are well made, and the Unpriced Delayering pattern was opened with more discipline than most. Three deltas, two of them uncomfortable:

  1. The exploratory slot went after the scan's own highest-value unread item and came back with something better and worse than expected. A 2026 doctoral dissertation tested whether doing change well protects people from the damage of continuous change. It does not. It helps, by a constant, and the slope is unchanged. A second instrument already in our base says the same thing in software engineering.
  2. The scan's headline market conclusion is false, and its own knowledge-base file says so. This morning's brief concluded that nobody has a format for pricing the capacity a restructuring spends, and that this absence is a consulting product. The format is published, free, and audited. The real finding underneath is better than the one it replaced.
  3. The asset the scan registered this morning is a section of the essay Brandon is already committed to writing, and it contains that essay's best missing paragraph.

OPS: merged clean from main (base 1514b78); no story thread was due today, confirming the scan's count. research-evidence: pushed to claude/modest-faraday-hmvi7n (commit 9ee478b, scripts/verify.sh passes locally) — evidence layer only (one KB file, one staged entry 2026-10-04-university-of-a-navigating-turbulence-in-the-public). Not pushed to its main; promote-claude-branches folds it in once scripts/verify.sh passes, or opens a PR if it does not. No shared meta file, research-index.json, dashboard, brief or CLAUDE.md touched there — those are the four PROPOSED-UPDATE blocks at the end.

📖 TODAY'S READ

Running a change well makes people better off. It does not make the change hurt them less. Those are different claims, and the framework has been making the second one by implication.

The document: Danika Pieters's doctoral dissertation, Navigating Turbulence in the Public Sector, defended at the University of Antwerp on 29 June 2026 — supervised by Koen Verhoest, Jan Wynen and Bjorn Kleizen, the group that owns the change-complexity construct the scan flagged this morning as the thing paper-2's emptiest cell has been missing for eight scans. Full text, free, from the university's repository. File: knowledge-base/pieters-antwerp-change-intensity-change-management-non-moderating-2026-06.md.

What it measures. One large Belgian public welfare agency, around 2,900 staff, surveyed twice a year apart (spring 2021 and spring 2022), plus eight focus groups. The central variable is change intensity — how many distinct changes a person has absorbed, and how fast. Not a mood or an attitude: a count. Seven change types (workplace, job function, job content, team composition, department structure, IT, other), asked of each employee over a three-year window. Seventy-one percent of staff had been through four to six different types of change in three years.

The result that matters, in the author's words: change management "shows a direct, rather than moderating, effect — suggesting that while it is positively associated with outcomes, it does not fundamentally alter the relationship between change intensity and psychological contract fulfillment." And across all three studies: "it does not moderate the relationship between change intensity and these outcomes."

In plain terms. Communicate well, involve people, support the implementation, show up as a visible leader — and your people are better off than they would otherwise have been, by roughly a fixed amount. But each additional overlapping change costs them just as much as it would have anyway. Good change management raises the floor; it does not reduce the damage per unit of change. That is the opposite of what a readiness-and-remedy framework implies when it tells a client to do the work before the transformation.

And it does not stand alone. Faros AI's 2026 telemetry study of 22,000 developers found that strong pre-existing engineering foundations give "no protection against this downstream deterioration regardless of baseline maturity." Two instruments, two countries, two literatures, two completely different ways of measuring — survey panel in a Belgian welfare agency, commit-level telemetry in software firms. Same shape: the prescribed good condition has a level effect and not a protective one.

The supporting half, which is real and arrives in exactly the cell we needed. The scan asked this morning, as the day's open question, what unit you would even price a structural change's capacity cost in. Public administration has been answering that for fifteen years and has a published instrument for it: count the distinct, concurrent and successive changes per person over a window, add a validated four-item pace scale, and manage "the pace, sequencing, and overlap of change trajectories" — which is the dissertation's closing recommendation, almost word for word the question the scan said nobody has a format for. Nobody has yet put an AI variable on the other side of that instrument. That is a much cheaper route into the capacity question than inventing a span-and-queue measure.

What Brandon should take from it: the remedy claim needs re-wording before a reviewer who knows this literature prices it — and the re-wording makes the framework sharper, not weaker. See the CHALLENGE section.

The limits, and the first one bounds everything above. No AI variable, and the window is pre-generative — surveys ran in 2021 and 2022, with change history reaching back to roughly 2018. This enters on the same previous-shock footing as the hospital-systems literature and may not be cited as evidence about AI. One site, one country, one sector. The restructuring "did not result in job losses" — so this is the mild end of the change range, not the delayering end. Every outcome is something an employee said about themselves, not an administrative record. And the psychological-contract measure is a single survey question.

🔭 THINKER TO WATCH

The Antwerp public-governance group — Koen Verhoest, Jan Wynen, Bjorn Kleizen (University of Antwerp, GOVTRUST) — because they have been measuring Brandon's dose variable for fifteen years and nobody in the AI conversation has noticed.

Not a single thinker and that is the point: this is a research programme with a decade of output on what happens to people inside organizations that keep reorganizing. A partial list from the dissertation's own front matter: repeated structural reforms in the Norwegian public sector producing defensive silence (employees withholding useful input for fear of conflict); change complexity reducing role clarity in the Australian Public Service; civil servants with heavy change histories becoming less proactive about new changes than colleagues with lighter ones; high change diversity raising intention to leave and presenteeism.

Why they matter now. Brandon's thesis says AI-era failure is a Five Breakpoints problem running at machine speed. "At machine speed" is a claim about dose — more change, arriving faster, overlapping more. This group has spent fifteen years building and validating the instrument that measures exactly that dose, in a sector where reorganizations are frequent, documented and politically driven. They are not AI researchers and have no apparent interest in becoming them. They are the people who already built the measuring stick Brandon's period clause needs.

Brandon's differentiation, and it is narrow and honest. They measure what cumulative change does to employees — strain, insecurity, broken expectations, silence. Brandon claims what it does to the organization's ability to execute — purpose, incentives, structure, momentum. Theirs is an occupational-health outcome; his is a transformation outcome. Neither has the other's dependent variable. That is a genuine complementarity rather than a competition, and it is the basis for the one study that would settle growth edge #2: their dose variable, his breakpoint score, and an outcome that is neither a feeling nor a commit count.

The caution. Their standing is in public administration journals, not in the rooms Brandon sells into, and every outcome they measure is self-reported. Citing them buys rigour on the dose and nothing on the outcome.

🏢 CASE IN THE WILD

A Belgian welfare agency of about 2,900 people did everything the change-management playbook asks, over four to six overlapping changes per person in three years, and its employees still could not say what their jobs were.

This is not a failure story and that is what makes it useful. The restructuring was incremental process improvement aimed at better service delivery. Nobody lost a job. Change management was measured and was positively associated with outcomes. And the dissertation's third study still found role clarity eroding "during implementation and ongoing fine-tuning, as roles and responsibilities failed to stabilize" — not because anyone failed to explain the change, but because the next adjustment arrived before the last one settled.

The mechanism, and it is Strategic Disconnection in a form Paper 1 does not currently contain. Paper 1's Strategic Disconnection is vague purpose producing the illusion of alignment — a definitional failure at the top. Here the purpose was fine and the definition never got the chance to hold still. Roles were clear each time they were announced and unclear by the time the following change landed. The disconnection is produced by the sequencing, not by the wording. That is a second route into the same breakpoint, and it is the route a continuous-deployment, continuously-reorganizing AI-era organization is most likely to take.

The second mechanism, measured, and it is Momentum Mirage with an unusual signature. Employees with heavy change histories rated the next change worse for themselves while still rating it fine for the organization (personal benefit clearly negative and statistically solid; the organizational-benefit model showed nothing at all). Technical detail: change complexity on personal-benefit appraisal −0.452, SE 0.17, p = 0.009; organizational-benefit model F(1,660) = 0.25, p = 0.6162. Perceived pace reproduces it: −0.638, SE 0.237, p = 0.007 against 0.03, p = 0.912. Translation: the same people, on the same day, still believe the organization is going somewhere and no longer believe it is going anywhere good for them. That gap is exactly what a leader reading an engagement survey would mistake for buy-in.

The open question this leaves, and it is answerable from this dataset by someone with access. The effect held after matching on how much employees trusted change management. So which arrives first — the loss of personal optimism, or the loss of role clarity? If optimism goes first, the sequencing problem is a morale problem and communication is the lever. If role clarity goes first, it is a structural problem and no amount of communication reaches it. Paper 1's prescription depends on the answer and the framework does not currently say which it believes.

⚙️ FOUR FORCES CONCLUSION

Commitment, and for once the evidence is about whether the Force can be repaired rather than whether it is broken.

The dissertation's Commitment measure is unusually direct — employees agreeing or not with "I feel my employer has fulfilled the promises made during the change processes of recent years." That is follow-through in the Four Forces sense, asked of the people on the receiving end of it. It falls as change intensity rises, and the qualitative work attributes the fall to overlapping trajectories that undermine "continuity, follow-through, and alignment with earlier expectations."

Here is what today adds, and it is a problem rather than a confirmation. The framework treats the Four Forces as the things you work on so the breakpoints do less damage. Today's evidence tests the two Forces that are easiest to buy as a programme — good communication and visible leadership (Momentum, part of Commitment), and practice maturity (Capability, via Faros) — and finds that both raise the level and neither flattens the slope. The two Forces that nobody tested are the two that require changing what the organization actually rewards and what it defines as the outcome: Incentive design and Purpose.

For paper-2 that is a sharpening, not a retreat. The defensible claim is narrower and more falsifiable than the one the corpus currently implies: the parts of the remedy you can buy as a programme improve the experience without protecting against the damage; the parts that are protective, if any are, are the ones that require redefining outcomes and decision rights — and nobody has measured those as moderators. That version survives contact with this literature. The current version does not.

Purpose is untouched today in the dose sense — role clarity erodes, but nothing here measures organizational intent. Capability behaves as a stock being spent: "continuous adaptation depletes resources", and the depletion is not offset by running the change well.

🥊 CHALLENGE

"Your remedy has a level effect, not a protective one." Logged in full in knowledge/thesis-challenges.md (2026-10-04), with the measured basis, the two instruments, and the three-part response.

At full strength. Every remedy framework makes an interaction claim whether it states one or not. Five Breakpoints says: here are the five ways transformation fails, here is the diagnostic, do the work and you fail less. That is a claim that the remedy flattens the slope from change pressure to bad outcome. Two instruments now test the nearest available proxies for "did the work" as moderators — change-management quality (Pieters, n=661/699, two waves, matched) and engineering practice maturity (Faros, 22,000 developers, telemetry) — and both find a level effect and no protection. If that generalizes, the framework sells a level improvement while implying a protective one, and the protective one is what a client is buying when they pay for readiness work before a transformation.

What it would take for the objection to win. A well-designed study that measures Purpose and Incentive conditions before a named change, against a real transformation outcome, and still finds no interaction. At that point Five Breakpoints is a diagnostic and not a remedy, and should be sold as one.

And the second charge, which is about this system rather than about Brandon, and is the more important of the two. We already held the Faros result and we read it as support. The knowledge-base file calls the no-protection finding "the sharpest available refutation of the belief that a capable, well-run organization can absorb a step change in tooling without redesigning how work flows." That reading swaps the moderator that was measured (engineering maturity) for one that was not (redesign), and keeps the conclusion. It may be correct. It is not what the instrument tested. It is also the second instance in two days of a reading that happens to favour the thesis — yesterday's gated-metric objection logged the first, against Stafford's own analysis. Two in two days is a practice issue, not an incident.

One piece of bookkeeping worth noting, because it was pre-registered. Yesterday's gated-metric challenge demanded "at least one application that costs Brandon something" and named the Faros-versus-DORA conflict as the first owed test. Today supplies a third instrument on that question, from outside software entirely, and it lands against us. The obligation is met visibly rather than dropped quietly.

📈 PATTERN BUILDING

Proposed as a new pattern at exactly 2 members — The Non-Moderating Remedy — and proposed rather than written, because the pattern file belongs to the scan. Full text in Routing below.

The construct: when the condition a framework prescribes is measured as a moderator of harm rather than as a correlate of good outcomes, it comes back direct but not protective — associated with better outcomes at every level of pressure, and not with a shallower slope.

Why these two are a pattern and not a coincidence: they have nothing in common except the question. Different continents, sectors, substrates (self-reported perception against commit telemetry), prescribed conditions (change-management practice against engineering maturity), and periods (pre-generative against generative). Nothing shared on the right-hand side, so the 2026-08-24 shared-instrument rule does not collapse them — which is the opposite of the GAO pair the scan correctly counted once this morning.

Boundaries set on opening. Not an AI pattern — Pieters has no AI variable and Faros has no organizational variable, so this is a pattern about remedy claims, not about AI. Not The Résumé-Panel Consensus (that is about measurement monoculture; neither member is résumé-inferred). Not Confidence Without Substrate. Promotion to 3: any third instrument testing a prescribed organizational condition as an interaction term against a harm outcome, in any field. Kill test, and it is cheap and real: one study that tests a prescribed condition as a moderator and finds a significant protective interaction — at which point this is publication bias in how interactions get reported, not a property of remedies.

🧱 ASSET SUGGESTION

No new asset, and that is the recommendation. Asset 2 (The Flattening Bet) is committed and being written; the ledger stays frozen per 2026-10-02. Today's two asset-shaped items both belong inside it. Full ledger note in knowledge/asset-suggestions.md.

The scan's "Three routes to a flatter chart" is asset 2's opening, not a sibling. The spine is good — Uber chose the flat chart and stated a mechanism, FEMA executed one with its strategic plan withdrawn first, the SEC arrived at one because of who took a buyout; only one of the three was designed and the market reads all delayering as design. Ship it separately and the committed essay loses its best entrance a month early.

What it adds that the draft lacks, and it is the strongest item in two weeks. Assets 1 and 2 both argue that the résumé-based studies cannot support the flattening consensus they are cited for. That argument has been methodological and therefore hard work for a reader. The SEC gives it a story: an organization whose spans widened because of who accepted a buyout, not because anyone designed anything. A study that watches job titles cannot tell those apart, because it never sees why anyone left. Fold into drafts/the-flattening-bet.md as the taxonomy opening plus one paragraph in the instrument-audit section.

🎓 SPARRING NOTE

Issued — tenth ever, on growth edge #2. Cap check per the 09-28 rule: Monday was 09-28, one note since (10-03, edge #1), cap is two, slot clear. This does not supersede the live 09-17 note — different edge, and answerable in one sitting.

Five Breakpoints implies an interaction: do the work and the failure modes do less damage. Two instruments have now tested the nearest proxies for "did the work" as moderators — change-management quality, and engineering practice maturity — and both found a level effect and no protection.

Write two sentences answering: which Forces does your framework claim are protective, and which does it only claim are good? Then state the one measurement that would distinguish them.

Why this one. Edge #2 has read for five weeks as the worthiest and least urgent item in the development file — a study you would run if you had a cohort. Today it became the thing standing between the framework and a live objection, and the answer does not need a cohort. It needs Brandon to say which half of his own prescription he is willing to bet is protective. The honest answer narrows the framework and makes it falsifiable, which is the opposite of a concession.

Parked for the edge's deliverable: the dissertation hands edge #2 the dose variable it has lacked — distinct change types per person over three years, 7-item, validated, public-sector-tested, with a 4-item pace scale alongside. An ex-ante breakpoint score needs something on the other axis. This is a published candidate nobody has put an AI variable against.

SOURCES & THREADS

Threads: 0 due, 0 opened, 0 advanced, 0 closed. No thread's next-check date fell today; the scan's count is confirmed independently — nearest open are the middle-manager load-bearing problem (2026-10-09) and the verification-cost thread (2026-10-11), then Uber (2026-11-08), which the scan opened this morning. Logged because the check happened. One note for the Uber thread rather than an advance: its pre-registered expected-failure mode — Uber will report headcount and not ownership clarity — now has a mechanism from today's read rather than just a prediction. Nobody measures the dose, so nobody can report the slope.

Candidates: 2 proposed, 0 promoted, 0 expired. Both in Routing below — the Antwerp group as a research body (cadence irregular, so no expiry clock under the 2026-10-02 rule) and Review of Public Personnel Administration as an outlet (cadence quarterly, 7-month clock). Neither is promoted on first contact; both serve probation.

One chase target closed as still-open, and it matters that it is recorded as unread. *Verlinden et al. (2023a), "Blurred Lines", remains unread* — SAGE returns 403. The Antwerp repository link that searches surface for it is a different document (today's dissertation). No claim in this base may rest on Verlinden 2023a, and this morning's description of it as "the construct paper-2's emptiest cell has been missing" is still a judgement from its title, not from its text. It stays an open chase target.

Exploratory slot (Sunday — no falsification sweep): the scan's own unread item, run on the method yesterday handed forward, and it worked twice. Yesterday's slot concluded that "the binding constraint today was reading, not search" and instructed the next slot to check the owed-reading list before composing a query. Done: the target was the scan's own highest-value unread chase. Base checked against it first per the 09-26 correction — zero KB files in either repo and zero index entries on Verlinden, change complexity or role clarity as a construct. The intended document is still 403; the substitute is better and newer. Logged in knowledge/query-log.md. Method handed forward, and it is sharper than yesterday's: today's second finding came from reading the scan's own knowledge-base file against the scan's own brief and finding them in contradiction. The next slot should do that comparison before composing any query at all — it cost one grep and produced the day's second result.

Thinker sites: not checked. The scan checked them at 12:00 and reported the AI Daily Brief null (three consecutive days, longest gap since promotion). Not repeated here — that is the scan's job and re-running it would be exactly the duplication CLAUDE.md rule 1 forbids.

Routing

strategy — the remedy wording is the highest-value correction available, and it is one paragraph. The framework implies a protective effect across all four Forces. The evidence supports a protective claim for none of them yet, and two have now been tested and failed. The fix is to state the narrower claim deliberately, before a reviewer who knows the change-management literature states it for us: the parts of the remedy that can be bought as a programme improve the experience without flattening the damage curve; whether redefining outcomes and decision rights does more is the open question, and nobody has measured it. That version is stronger because it is falsifiable and because it names the study that would settle it. Do not let this surface first in a review.

strategy — kill the capacity-assessment product as this morning's brief pitched it; keep the finding, which is better. The scan concluded that no standard artifact exists for pricing the capacity a restructuring spends, and that the absence is a consulting product. The artifact exists. GAO-04-39, Human Capital: Key Principles for Effective Strategic Workforce Planning (December 2003), principle 3: "develop strategies that are tailored to address gaps in number, deployment, and alignment of human capital approaches" — gaps in number against programmatic results is capacity against obligation. The worked template is public too: the Guide to Workforce Planning on OPM's own site runs Supply Analysis → Demand Analysis → Gap Analysis with named workload steps and reporting templates. And the scan's own KB file records that GAO assessed FEMA "against OPM's Workforce Planning Guide and GAO's talent-management guidance" — a finding phrased as deviation from a standard presupposes the standard. Pitching a 23-year-old federal standard as a new deliverable is the fastest way to be told so in a room. The finding underneath is Brandon's own territory and is better: the artifact is published, free and audited by a body with subpoena power, and two agencies still did not produce it while making irreversible structural decisions. That is not a Capability gap — it is a Commitment failure, a formally owed step skipped under political pressure with no consequence inside the decision window. What is sellable is the gate, not the worksheet — the thing that forces the analysis to exist before the decision is irreversible.

governance — Stafford's answer to the scan's GAO credibility question, since it asked for Brandon's call: amend the list, and the reason is a rule rather than a preference. The scan filed both GAO reports moderate because the enumerated high-credibility list does not name GAO, and routed the amendment to Brandon rather than acting on it. That discipline was right and the answer should be yes. The current list's membership test is effectively brand recognition: it admits every large consultancy and several vendors, and excludes a legislative audit agency that publishes its data systems, names the standard it audits against, labels its own samples non-generalizable, and has no product to sell. Today produced the same mismatch a second time — the Antwerp dissertation is filed moderate because the list names Harvard, Stanford and Oxford but not Antwerp, which grades European public-administration research by university fame. Proposed amendment, narrow and rule-shaped rather than a list of favourites: add (a) national audit institutions and inspectors general, and (b) doctoral dissertations and peer-reviewed journal articles from accredited universities regardless of whether the university is named. Both have a disinterestedness property the consultancies on the list do not. Brandon's call; not acted on in either repo.

personal — the sparring note above, and one observation about how this system is reading its own evidence. Two days running, a reading that favoured the thesis turned out to have substituted an unmeasured quantity for a measured one — yesterday's gated-metric rule, today's Faros reading. Both were caught by the same move: asking what the instrument actually measured, rather than what the conclusion needed it to measure. That is now a standing check rather than a lucky catch.

PROPOSED-UPDATE 1 — knowledge/emerging-patterns.md (research-evidence), new pattern entry

NEW 2026-10-04 (Stafford): The Non-Moderating Remedy — opened at 2 members, with the no-shared-instrument check applied on opening.

The construct: when a prescribed organizational condition is tested as a moderator of harm rather than as a correlate of good outcomes, it comes back direct but not protective — better outcomes at every level of pressure, and no shallower slope.

  • Member 1 — Pieters (2026), University of Antwerp doctoral dissertation, defended 2026-06-29; two survey waves 2021–2022 plus eight focus groups in one ~2,900-employee Belgian public welfare organisation, n=661 (Article 1, entropy-balanced quasi-experiment) / 699 (Article 2, linked). Change-management quality tested as a moderator of change intensity's effect on psychological contract fulfilment: "Change management shows a direct, rather than moderating, effect — suggesting that while it is positively associated with outcomes, it does not fundamentally alter the relationship between change intensity and psychological contract fulfillment." Synthesis across all three articles: "it does not moderate the relationship between change intensity and these outcomes." File: knowledge-base/pieters-antwerp-change-intensity-change-management-non-moderating-2026-06.md
  • Member 2 — Faros AI (2026-04), 22,000 developers, within-organization telemetry: strong pre-existing DORA-style engineering foundations provide "no protection against this downstream deterioration regardless of baseline maturity." File: knowledge-base/faros-ai-engineering-report-acceleration-whiplash-2026-04.md

These are two members, not one. Nothing is shared on the right-hand side: different continents, sectors, substrates (self-reported perception against commit telemetry), prescribed conditions (change-management practice against engineering maturity), and periods (pre-generative against generative). The 2026-08-24 shared-instrument rule does not collapse them — contrast the GAO pair opened the same day, which correctly counts once.

Boundaries set on opening. Not an AI pattern — Pieters has no AI variable, Faros has no organizational variable; this is a pattern about remedy claims and may not be cited as evidence about AI deployment. Not The Résumé-Panel Consensus (measurement monoculture; neither member is résumé-inferred). Not Confidence Without Substrate. Relationship to The Unpriced Delayering (opened the same day): adjacent but distinct — that pattern is about a cost nobody computes, this one is about a benefit that is computed and comes back smaller than claimed.

Promotion to 3: any third instrument testing a prescribed organizational condition as an interaction term against a harm outcome, in any field. Kill test, and it is cheap: one study testing a prescribed condition as a moderator and reporting a significant protective interaction — at which point this is publication bias in how interaction terms get reported, not a property of remedies.

Caution for anyone citing this pattern against the framework. Neither member measured the framework's own remedy. Pieters measured communication, participation, implementation support and visible leadership; Faros measured engineering practice maturity. Nobody has tested Purpose (outcome definition, decision rights) or Incentive design as moderators. The pattern bounds the claim; it does not refute it.

PROPOSED-UPDATE 2 — knowledge/paper-2-case-file.md (research-evidence), Claim 5 and the remedy-wording problem

2026-10-04 (Stafford): Claim 5's prerequisite has a measured objection, and it is about the shape of the claim rather than its direction.

Source: Pieters (2026), University of Antwerp dissertation — knowledge-base/pieters-antwerp-change-intensity-change-management-non-moderating-2026-06.md.

Claim 5 holds that the prerequisite is organizational readiness, not coordination technology. The scan's 2026-10-04 entry gave the claim an audited negative pole (FEMA: structural change executed with the purpose document withdrawn and no capacity analysis). This adds the other kind of evidence Claim 5 has been missing — an instrument that tested readiness-type conditions as a moderator — and it comes back against the interaction. Change-management quality has a direct effect and no moderating effect on the relationship between cumulative change intensity and harm, and the effect on appraisal holds after matching on trust in change management.

What this changes in the claim's wording, which is the actionable part. Claim 5 currently implies that readiness work is protective. The evidence supports level claims for the two conditions that have been measured and no protective claim for any condition yet. Proposed restatement: the readiness conditions that can be purchased as a programme (communication, participation, implementation support, practice maturity) improve outcomes without flattening the damage curve; whether the conditions that require redefining outcomes and decision rights are protective is untested by anybody. That version is falsifiable and names its own test.

Evidence-gap ranking consequence: the top gap on Claim 5 is no longer "contrast cases and moderators" in general. It is specifically an instrument that tests Purpose or Incentive conditions as an interaction term against a transformation outcome. Nothing in the base does this.

Limits that must travel with any use: pre-generative window (2021–2022), no AI variable, single Belgian public-sector site, a restructuring with no job losses, all outcomes self-reported, psychological contract fulfilment measured on a single item.

PROPOSED-UPDATE 3 — knowledge/field-map.md (research-evidence), new entry

Antwerp public-governance group — Koen Verhoest, Jan Wynen, Bjorn Kleizen (University of Antwerp, GOVTRUST) · added 2026-10-04 · Tier 2 (adjacent territory, instrument-relevant rather than positioning-relevant).

What they do: fifteen years of empirical work on the micro-level effects of cumulative organizational change in the public sector — repeated structural reforms producing defensive silence (Norway); change complexity reducing role clarity (Australian Public Service Employee Census); heavy change history reducing proactive behaviour (Belgium); change diversity raising turnover intention and presenteeism. Most recent output: Danika Pieters's dissertation, defended 2026-06-29.

Why relevant now: they own the validated instrument for change intensity — the dose variable Brandon's "at machine speed" clause asserts and the corpus has never measured. Nobody in this group works on AI.

Brandon's differentiation, and it is a complementarity rather than a rivalry: their dependent variables are occupational-health outcomes (strain, insecurity, silence, intent to leave); his is transformation outcome. Neither side has the other's outcome variable, which is precisely why their dose plus his breakpoint score is the design growth edge #2 has been looking for.

Caution: public-administration standing, not practitioner standing; every outcome self-reported. Citing them buys rigour on the dose and nothing on the outcome.

PROPOSED-UPDATE 4 — knowledge/source-candidates.md (research-evidence), two candidates

University of Antwerp / GOVTRUST (Faculty of Social Sciences, Department of Political Science) · first seen 2026-10-04 · surfaced by: Stafford's exploratory slot, Pieters dissertation read in full from the institutional repository · signal count 1 · cadence irregular (a research body publishing when it publishes, so no expiry clock under the 2026-10-02 rule). Promotion condition, deliberately narrow: a second find from this group on a different construct or dataset — not another output from the same dissertation, which would be one signal observed twice. The specific signal wanted: any output from this group putting an AI or automation variable against their change-intensity instrument. That does not exist as far as today's search reaches, and it would be the first thing in this base to measure Brandon's dose variable in the generative period.

Review of Public Personnel Administration (SAGE) · first seen 2026-10-04 · surfaced by: the same read; carrier of Verlinden et al. 2023a and of two further outputs from this group · signal count 1 · cadence quarterly (7-month clock, to 2027-05-04). Logged with a working note: SAGE returned 403 on direct fetch today, so this outlet is currently unreadable by this pipeline and its value is as a citation-chase target reachable through institutional repositories, not as a monitorable feed. Promotion requires a find whose text was actually obtained.

Filed by Stafford (judgment layer), 2026-10-04. Delta layer only — the 12:00 scan's brief at research-evidence/briefs/2026-10-04.md is not restated here.