Stafford Brief — 2026-09-22
Delta layer over the 12:00 research scan (research-evidence/briefs/2026-09-22.md: 1 staged, 9 skipped). Tuesday — no falsification sweep, no weekly synthesis. Yesterday's run merged clean; no OPS flag.
The scan told Brandon he needs to weaken a claim before publishing it. He published the weakened version four months ago, and the strong version exists only in our own filing cabinet. That is today, and it matters more than it sounds, because the same check turned up a 2012 Harvard paper that settles a question Brandon is about to build an essay on — and settles it against the way the essay is currently planned.
📖 TODAY'S READ
"The Flattened Firm — Not as Advertised" — Julie Wulf, Harvard Business School Working Paper 12-087, April 2012. Read in full; knowledge-base/wulf-flattened-firm-not-as-advertised-2012-04.md.
The point first: the last time companies flattened themselves on a mass scale, somebody measured what actually happened, and the answer was the opposite of the sales pitch. Wulf studied more than 300 large US companies from 1986 to 2006, using each company's own reporting charts — who reported to whom — rather than guessing structure from résumés, which is how most of the research this base holds is done.
The flattening was real. The number of management levels between a division head and the CEO fell by about a quarter. Chief Operating Officers started disappearing. The number of people reporting directly to the CEO went from about four and a half to about seven, and on later data roughly doubled again. If you were checking whether these companies delayered, every number says yes.
What the flattening was sold as doing, it did not do. The pitch — Wulf quotes BCG, which trademarked the word "delayering" in 2005 — was that removing layers pushes decisions down, closer to customers, improving accountability and morale. What she finds instead is that decision-making moved up. CEOs filled their top teams with company-wide functional bosses — CFO, HR chief, CIO, CMO — and division managers' pay fell as those people arrived. Since pay tracks authority, her reading is that the functional bosses took over work the division managers used to do. Her own title says it: Not as Advertised. And in interviews, CEOs were blunt about why they cut the COO — for more control and involvement, not less.
What Brandon should take from it, in one sentence: removing a layer does not remove the work that layer was doing — it moves it, and the layer count, which is the number everyone uses to certify that a flattening succeeded, is the one measurement that cannot show you where it went.
The limits, and they bind. One research programme, not four: this paper summarizes Wulf's own earlier studies on the same dataset, so those studies never count as independent confirmation of each other. It is an unrefereed working paper with no new statistics in it. Decision rights are inferred from pay, which is where a critic will push. And it ends in 2006 — there is no AI in this paper anywhere, and it must never be cited as an AI finding.
🔭 THINKER TO WATCH
Julie Wulf, Harvard — and the reason to watch her is a rare one: she is the only person who systematically checked whether a management fashion delivered what it promised, and then published that it did not, under a title that says so.
Her co-author on the foundational work is Raghuram Rajan — former IMF chief economist, former Governor of the Reserve Bank of India. That matters for Brandon tactically: the flattening literature's original paper has a very well-known name on it, and this base has been citing commentary about flattening for months while holding none of the underlying work. Today's scan flagged exactly that and ranked chasing it first; this is that chase.
How Brandon's position differs, and the concession comes first. Wulf is asking a corporate-governance question — who holds decision rights, and can we see it in pay? She is not asking whether the organization worked afterward, and she has no reason to. What her data support, and her paper has no incentive to say, is that the companies got the structure they asked for and not the behaviour they wanted — and that the gap between the two was invisible to everyone reading the org chart. That is a Five Breakpoints reading of her result, not a governance one, and Brandon can make it without contradicting a word she wrote.
🏢 CASE IN THE WILD
General Electric, July 2002 — the cleanest named instance of the mechanism, and it comes with the CEO's stated reason on the record.
Jeff Immelt eliminated the chairman of GE Capital and had the four GE Capital business-unit heads report directly to him. On the org chart this is textbook flattening: a level removed, the CEO's span widened. His own explanation, from GE's press release, is the interesting part: "The reason for doing this is simple—I want more contact with the financial-services teams. With this simplified structure, the leaders of these four businesses will interact directly with me, enabling faster decision making and execution."
Read that again against what flattening is supposed to mean. Nothing in it is about pushing decisions down. It is a CEO saying he removed a layer in order to be more involved. The delayering and the recentralization are the same decision, announced in the same sentence, and only one of them shows up in a structure chart.
The open question this leaves, and it is answerable. GE Capital in 2002 is the most consequential possible test case — within six years it was the part of GE that nearly took the company down. Did the removal of that chairman layer change who was watching the risk? Nothing in Wulf's paper asks, and nothing in this base answers. It is not today's work, but it is the single highest-value historical case available to the delayering argument, and it is sitting in public records.
⚙️ FOUR FORCES CONCLUSION
Purpose — and specifically the case where an organization's stated goal and its leaders' actual goal are different goals.
Today's scan landed on Capability, correctly, and noted it was breaking a three-day run on Commitment. Wulf's evidence lands somewhere neither has been recently. These companies did not fail to communicate their purpose. They communicated one purpose publicly and pursued a different one privately, and the mismatch was never reconciled — the public rationale was delegation and responsiveness; the CEOs' own account, in interviews, is that they wanted more control.
Brandon's framework says vague purpose produces the illusion of alignment. This is a sharper variant worth naming: precise purpose, precisely misstated, produces the same illusion and is much harder to catch — because nothing sounds vague. Everyone can repeat the goal. The division managers Wulf describes were hired and developed to "be the boss," and then found, in her words, "a mismatch between managerial talent and decision rights." They were aligned to the announced purpose. The announced purpose was not the operating one.
What this changes for the paper-2 argument: the diagnostic question for a restructuring is not is the goal clear? It is does the goal the leadership states match the goal its own behaviour is optimizing? — and Wulf shows the second one is readable from pay movements, which is a measurement rather than an interview.
💭 OPEN QUESTION
Wulf could see where the work went because it went to people who had salaries. If the work goes to a system instead, what is the equivalent of the pay record?
This is the question that makes today useful rather than merely historical. Her entire method rests on one thing: when decision rights moved from division managers to functional bosses, the money moved with them, and she could read the migration in compensation data companies are required to publish. The work relocated, and it left a financial footprint.
Now run the AI version. A company removes a management layer and says agents absorbed the coordination. Yesterday's scan recorded Veeam's finding that two-thirds of large European enterprises have autonomous workflows their own IT cannot fully track. If the supervisory work went there, no one's pay moved, no one's title changed, and nothing appears in any disclosure — so the single method that detected displacement in the previous shock returns nothing in this one. A researcher running Wulf's design on a 2026 delayering would find a flattening with no compensating migration and would reasonably conclude the work was genuinely eliminated.
The uncomfortable possibility, stated plainly: "the work was eliminated" and "the work moved somewhere unmeasurable" produce identical data. Brandon's whole position depends on which of those is true, and there is currently no way to tell them apart. What would distinguish them is not a span count or an agent count. It is a failure record — whether the exceptions, escalations and judgment calls the removed layer used to catch are still being caught. Nobody publishes that, which is precisely why it is worth Brandon being the one to ask for it.
🥊 CHALLENGE
The strongest thing found against Brandon today is Wulf's actual result, and it goes against his published Paper 4. Logged in full: knowledge/thesis-challenges.md, "The previous delayering shock strengthened stewardship at the top rather than stranding it" — OPEN.
The objection at full strength. Paper 4 argues that organizations correctly spot the routing work technology can absorb, "then cut the human layer that also carried informal interpretation, escalation and judgment," so the four forces lose their carriers and the breakpoints accelerate. Wulf measured the one delayering wave anybody has measured end to end, and oversight got stronger, not weaker. More functional bosses with company-wide remit. Decision rights moving toward them. CEOs spending more time in internal meetings and less time alone. The layer went and the judgment work was visibly re-carried — by named, paid, chart-visible people. If that is what the last delayering shock did, Brandon owes his readers a reason why this one differs.
And there is a second, nastier limb. This is not a case where the framework's remedy went untried. Re-assigning stewardship when you remove the layer carrying it is exactly what 300 companies did, twenty years before anyone wrote Paper 4 — and the paper describing it is still titled Not as Advertised, with division managers left holding a mismatch between their talent and their authority. Stewardship being re-carried is not sufficient for the transformation to work, and Paper 4's argument does not currently contain that possibility.
What it would take for the objection to win. It wins if an AI-era delayering shows stewardship re-carried by identifiable people with real authority, the way Wulf's firms did — testable now, on the ASML thread's amended criterion and on Meta's Project OT. It loses if AI-era delayerings show the layer removed with no compensating movement in anyone's pay, title or decision rights, the work having gone to a system rather than a person.
Brandon's best answer, which is strong but has to be said out loud rather than assumed: the carrier changed category. Wulf's firms handed stewardship to people — which is why she could measure it. The AI-era claim is that it is handed to something with no headcount line, no manager and no review. That narrows his prediction and makes it falsifiable, which is a better position than the one Paper 4 currently states. It should go into the essay as a concession, early, in his own voice.
📈 PATTERN BUILDING
The displacement construct now has two independent members, and the scan was right to refuse to open it on one.
Today's 12:00 scan named a construct — the supervisory work does not disappear when the span widens; it reappears off the org chart — and deliberately did not open it as a pattern, on the correct grounds that one Scandinavian firm in 2001 is an observation. It set a prospective test on the ASML thread instead. That discipline was right, and it has been rewarded faster than expected: Wulf is a second member, and it arrived from a different country, a different decade and a different research group, on the same class of instrument — company-reported reporting lines rather than résumé guesswork.
The two members disagree about direction, and that disagreement is the finding, not a defect. In the Scandinavian firm the supervisory work reappeared below, as an unofficial tier of assistant managers the company declined to give a number to. In Wulf's 300 firms the decision rights migrated up, into a C-suite that grew functional specialists. What both measure is displacement. Where it lands is contingent — and a construct that predicts "the work moves and the chart won't show you where" is more useful to a practitioner than one that predicts a single direction, because it tells him to go looking rather than to expect a specific answer.
Proposed as a new pattern at 2 members, not escalated — PROPOSED-UPDATE 2 below. Two honest caveats travel with it: the second member was found by someone who already had the construct in hand, which is weaker than the prospective ASML test the scan set up; and Wulf's four underlying studies are one research programme on one dataset, so the pattern has two members, not five.
🧱 ASSET SUGGESTION
Asset 2 (The Flattening Bet) should not ship the re-centring proposed on 09-18, and today's amendment partly retracts it. Full spec: knowledge/asset-suggestions.md, Amendment 2026-09-22.
In two sentences. The 09-18 amendment proposed rebuilding the essay around a fork — either AI makes hierarchies shrink or it makes authority re-sort — and called it a test anyone could run on published data. Wulf ran that test on the previous shock and the answer is both, simultaneously, in the same firms; she refuses the either/or in her own words: "Firms are doing both." This base had already stumbled on the same result and not recognised it — the 09-20 amendment found Meta cutting layers and detaching decision rights from reporting lines in the same quarter.
What replaces the fork is better and simpler. Not which shape the hierarchy takes, but where the decision rights land — with Wulf's own observable attached: watch whether anyone's pay, title or authority moved to absorb what the removed layer was doing. Layer counts tell you nothing. That is data companies already publish, it connects straight to asset 4 (The Proxy-Statement Instrument), and it is falsifiable on proxy-statement timescales instead of in 2028.
The ledger fact Brandon should see, and it is not about this essay. This is the fifth amendment to asset 2 in thirteen days without the draft moving, and it is the first one that argues with an earlier one. The 09-11, 09-17 and 09-18 notes all flagged identification outrunning execution as this system's constraint. A self-retracting amendment stack is what that constraint looks like once it has been left to run. Only shipping resolves it.
🎓 SPARRING NOTE
Cap check: two per week; none issued this week (09-21 onward). Slot clear. This does not supersede the 09-17 note on growth edge #1, which stands unanswered and is a bigger task — this one is deliberately smaller.
The prompt, fifteen minutes, before any further reading: write the single paragraph that concedes Wulf into asset 2. It has to do three things — state her finding accurately (the last delayering wave strengthened oversight at the top), say why it does not defeat the argument (her firms handed the work to people whose pay moved; the AI claim is that it goes to something uncounted), and convert the concession into a test the essay is willing to lose on.
Why this and not more reading. The 09-17 note observes that Brandon's demonstrated strength is the precise, unforced concession made early rather than under pressure — he did it once already, publicly retiring the 70% figure in Paper 3, and it bought more credibility than the claim it gave up. This is the same move, available a second time, on better material, and currently unmade. The Wulf concession is the cheapest resolution of a challenge logged today and the most attackable gap in the draft.
The part to push back on. I think the category distinction — people versus systems as stewardship carriers — carries the concession. I have not tested it. If someone shows that AI-era delayerings do produce compensating pay and title movement, the distinction collapses and the objection wins outright. Brandon should treat my confidence here as a position to attack, not a finding.
SOURCES & THREADS
Threads: 0 due, 0 advanced by Stafford, 0 opened, 0 closed. Nothing was due today — nearest is Agents in production without a governance model, 2026-09-30 — and nothing was brought forward, per the 09-07/09-14 cadence discipline. The scan advanced two threads (Regime Fork; ASML criteria amended additively) and Stafford does not duplicate them. One proposal below adds Wulf to the ASML thread's amended criterion, because Wulf tells that criterion what to look for in the hiring and the Scandinavian case does not.
Source candidates: 1 proposed. Julie Wulf / the Harvard Business School organizational-economics seam, at 1 signal. Probation argument, stated so a future run can check it: this is the second consecutive day a find has come from company-reported hierarchy data rather than résumé inference, which is the genre the scan opened a candidate on yesterday for exactly this reason. A second signal requires a different paper from this programme carrying an organizational outcome — not another summary of the same panel. Deliberately not proposed as a candidate: hbs.edu as a route. It is an access route, not a source, per the 09-18 BNN Bloomberg distinction and the scan's own doku.iab.de call today.
Exploratory query used (logged to knowledge/query-log.md): Rajan Wulf "The Flattening Firm" evidence panel data changing nature corporate hierarchies — the chase the scan ranked first today and explicitly declined to run. The target paper was not the find. Rajan & Wulf (2006) is gated behind JSTOR and MIT Press; the NBER abstract page confirmed the citation facts. The value was two results down — Wulf's own 2012 synthesis, served openly as a full PDF by Harvard. Per the 09-10 lesson (a gated primary is a failed chase, not a failed query — read the whole result set), which is now the second time it has paid.
Method note worth keeping. The automated fetch returned the Harvard PDF as unreadable binary and said so rather than guessing at its contents. Extracting it locally took one command. That is the third time this month a primary source was one step past a failed summarizer, and it is the same lesson as the scan's own note today that only the extracted PDF gave the authors' actual words.
Still unchased, in priority order: Rajan & Wulf (2006) primary — try juliemwulf.com/publications first, per the 09-04 author's-own-site lesson, which has not been applied here yet; Guadalupe, Li & Wulf (2011), which is where the pay result actually lives and which every claim about it currently rests on at second hand; and Garicano (2000), still the oldest open reading task in this system and now cited at second hand in six places.
Routing
OPS — a structural gap between the two repositories, and it cost a slot today. The scan spent its exploratory query on "paper-2 Claim 1's untested liability, which the case file has never logged as a counterargument." That is true of the case file and false of the system: knowledge/thesis-challenges.md has carried that exact objection since 2026-09-18 — Hierarchy is ordered by monitorability, not by routing, CONTESTED, steel-manned, with a stated win condition and a proposed repair. The scan has no read path to that file; it is Stafford-owned and lives in this repo. The scan did good work today — it read a PDF properly, correctly declined a Technology Illusion tag it could have taken cheaply, and amended the ASML thread additively rather than softening it. This is a plumbing gap, not a craft failure. Proposed fix, one line, no new process: add a pointer to research-evidence CLAUDE.md §Standing Sources or to knowledge/paper-2-case-file.md saying that counterarguments against paper claims live in tgifriday7/stafford-research knowledge/thesis-challenges.md and should be checked before an exploratory slot targets a claim's weakness. Brandon's call — it is the scan's file, not ours.
Strategy — the scan's Claim 1 recommendation restates Brandon's own published position, and the edit it asks for belongs to our filing, not to his argument. The scan's Routing says the comparative version of Claim 1 is "the sentence this costs Brandon" and that he "should make that edit before a critic from organizational economics makes it for him." Paper 4, published 2026-05-26, already says it. Its own words: "A significant portion of management work has historically been communication routing" — a portion, not the whole of hierarchy — and, on the routing-implies-shrinkage argument, *"That argument is partially right."* And on the residual: "Those functions require carriers. They may be redesigned. They may be concentrated in fewer people than before. But they do not disappear because routing got cheaper." The absolute version — hierarchy was an information routing protocol, not a purpose/commitment/capability system — exists only in research-evidence/knowledge/paper-2-case-file.md line 15, where it is a research scaffold still marked "Need: historical evidence." So the real risk is the reverse of the one the scan named: a future draft built from the case file's phrasing would regress from what Brandon has already published and hand a critic an inconsistency between two of his own pieces. Recommended: amend the case file's Claim 1 heading to match Paper 4's published wording. Brandon changes nothing. (PROPOSED-UPDATE 1.)
Strategy — and the better news, which the scan framed backwards. Because Paper 4 already carries the careful version, Smeets & Warzynski and Wulf are not costs. They are the first measured support for a mechanism Brandon published as an assertion in May. Paper 4 says stewardship functions "may be concentrated in fewer people than before" but do not disappear; Wulf measured exactly that concentration across 300 firms, and Smeets & Warzynski caught the opposite-direction version inside one firm's payroll. Two independent instruments, from the previous technology shock, both showing the work relocating rather than vanishing. That is worth a paragraph in the essay and a sentence in the case file, stated as what it is: not new argument, but the first evidence that the published argument describes something real.
Governance — unchanged from the scan and not restated. Its board-level question (who performs the checking our removed managers used to do, and can you name them) is the right one and today sharpens only its method: the follow-up question is whether anyone's pay or title moved when that work was reassigned. If the answer is nobody's, the work was not reassigned to a person.
Personal — seventh consecutive brief, and today it changed character. The backlog note has been "identification outruns execution" since 09-16. Today asset 2 produced an amendment that partly retracts an earlier amendment. The 09-21 note set the test: if Friday makes it four consecutive weeks, that stops being a note about the backlog and becomes the finding about the system. Two scans remain, and today is the first evidence that the cost is no longer just delay — it is rework. The 09-18 re-centring was written in good faith on the best evidence then available, and four days later it needs unwinding. That is what an unshipped draft accruing amendments does, and it will keep doing it.
STAGE-CANDIDATE — ready to run in research-evidence
python3 build/add-research-json.py --json '{
"title": "The Flattened Firm - Not as Advertised",
"publisher": "Harvard Business School Working Paper 12-087 (Julie Wulf)",
"url": "https://www.hbs.edu/ris/Publication%20Files/12-087_bc50bde2-3016-457a-9bee-dc988cb1056b.pdf",
"published_date": "2012-04-09",
"credibility": "high",
"instrument": "Panel of firm-reported reporting relationships, job descriptions and compensation for 300+ large US firms, 1986-2006; hierarchy depth measured as average positions between CEO and division managers; decision rights INFERRED FROM RELATIVE PAY, not observed. Supporting CEO interviews (n=7) and CEO time-use survey run 2010-2011. Pre-generative: no AI variable. NOT INDEPENDENT of Rajan & Wulf (2006), Guadalupe & Wulf (2010), Guadalupe/Li/Wulf (2011) - same author, same panel; one instrument, not four. Unrefereed working paper reporting no new estimation.",
"breakpoints": [
{"name": "strategic_disconnection", "rationale": "Delayering was adopted across 300+ firms on the stated rationale of pushing decisions downward to improve responsiveness and accountability, and Wulf finds decision rights moved upward to functional managers instead, leaving division managers with what she calls a mismatch between managerial talent and decision rights."},
{"name": "momentum_mirage", "rationale": "The metric that certifies a delayering succeeded moved as advertised - depth between division heads and CEO fell about 25 percent, from 1.6 in 1986 to 1.2 in 1999 - while the governance change the exercise was undertaken to produce did not occur, so the layer count reported a completed transformation that had not happened."}
],
"forces": [
{"name": "Purpose", "rationale": "The publicly stated objective (delegation and responsiveness) and the objective CEOs give in interviews are different objectives: they report eliminating the COO position in pursuit of more control, more oversight and more involvement in business operations and decisions, not less."}
],
"key_findings": [
"Firms genuinely delayered: positions between division heads and CEO fell about 25 percent, 1.6 (1986) to 1.2 (1999); firms with a COO fell from 55 percent (1986) to 45 percent (1998)",
"CEO span of control rose from 4.5 (1986) to almost 7 (1999), and on a later smaller sample more than doubled from 4.7 to 9.8 over roughly 20 years",
"Division manager pay declines as functional managers join the executive team, driven by product functional managers (marketing, R&D) who perform activities typically the division managers responsibility - read by the author as decision rights moving up",
"Author refuses the either/or: standard classifications of centralization or decentralization are too simple; firms are doing both",
"CEO interviews: they eliminated the COO position in pursuit of more control, more oversight and more involvement in business operations and decisions, not less"
],
"sample_size": 300
}'
Stage with the KB file if preferred: --from-kb against this repo's knowledge-base/wulf-flattened-firm-not-as-advertised-2012-04.md (copy it across first; Stafford has no write access).
PROPOSED-UPDATE 1 — knowledge/paper-2-case-file.md, line 15 (the Claim 1 heading)
Replace:
Claim 1: Traditional org hierarchy was an information routing protocol, not a purpose/commitment/capability system
With:
Claim 1: A significant portion of management work was information routing — and the routing account is the best-supported of the rival explanations without being the whole of it
Wording aligned 2026-09-22 to Paper 4's published text (
stafford-research/papers/paper-4-the-ai-mirror.md, §"Routing and Stewardship Are Not the Same Thing", published 2026-05-26), which already states the portion claim and calls the routing-implies-shrinkage argument "partially right." The absolute phrasing was a 2026-04 research scaffold that this file has now outrun; a draft built from it would regress from what Brandon has published. Evidence for the comparative version: Smeets & Warzynski (2026-09-22 entry, the four-theory test) and Ide & Talamàs (JPE 2025, on scarce expert time). Residual named honestly: the routing account does not explain the sorting, learning and reallocation of people.
PROPOSED-UPDATE 2 — knowledge/emerging-patterns.md, new pattern at 2 members
Delayering Displaces, It Does Not Eliminate — 2 members, not escalated, opened 2026-09-22. The claim: when an organization removes a management layer, the work that layer carried relocates rather than disappearing — and the layer count, which is the measure used to certify the flattening succeeded, is structurally incapable of showing where it went. Member 1 — Smeets & Warzynski (2006), one Scandinavian firm, 1997–2004, firm-reported chain of command. Spans widened at every level, formal job levels never changed, and an informal assistant-manager tier ("job level 0.5") appeared in 2001 "in those departments where managers had the highest span of control." Work relocated downward, off the chart. Member 2 — Wulf (HBS WP 12-087, 2012), 300+ US firms, 1986–2006, firm-reported reporting lines. Depth fell about 25 percent and CEO spans nearly doubled, while functional managers joined top teams and division-manager pay fell as they did. Decision rights relocated upward, into the C-suite. Why it is a pattern despite the directions differing: both measure displacement on employer-reported hierarchy data, independently, twenty years and one continent apart. The direction is contingent; the displacement is not. A construct that says the work moves and the chart will not show you where is more useful to a practitioner than one predicting a single direction. Three caveats carried from the start. (1) The second member was found by a run that already held the construct, which is weaker than the prospective test the 2026-09-22 scan set on the ASML thread; that test remains the stronger evidence when it lands. (2) Wulf is one instrument, not four — a synthesis of the same author's prior results on the same panel (Rajan & Wulf 2006; Guadalupe & Wulf 2010; Guadalupe/Li/Wulf 2011). (3) Both members are pre-generative. This is the previous information-routing shock, on the Bloom/Garicano footing; never cite either as an AI result. What would escalate it: a third member from a different instrument genre, or the ASML criterion (b) hit — re-layering visible in post-restructuring hiring.
PROPOSED-UPDATE 3 — knowledge/story-threads.md, ASML delayering, append to the 2026-09-22 entry
Appended 2026-09-22 (Stafford): criterion (b), added this morning, asks for evidence that supervisory work reappeared in a form ASML's org chart does not show. Wulf (HBS WP 12-087, 2012;
stafford-research/knowledge-base/wulf-flattened-firm-not-as-advertised-2012-04.md) tells that criterion where else to look, and the Scandinavian case does not. Her 300-firm panel found the relocation going upward — company-wide functional roles added to the top team, with division-manager pay falling as they arrived — which she reads as decision rights migrating up. So criterion (b) should watch two directions, not one: new coordination roles below (team lead, coordinator, chief of staff, deputy, programme manager) and new or newly-empowered company-wide functional roles above the units that lost managers. And it gains an observable ASML actually publishes: Wulf detected the migration through relative pay, so any disclosed movement in the pay or remit of executive-level functional roles against the delayered units is evidence on this criterion. Original criterion unchanged; next check unchanged at 2026-11-01.
PROPOSED-UPDATE 4 — knowledge/source-candidates.md, new entry at 1 signal
Julie Wulf / Harvard Business School organizational-economics seam — first seen 2026-09-22, surfaced by the Rajan & Wulf "The Flattening Firm" citation chase that the 2026-09-22 scan ranked first and left unrun. Signal count: 1 (
wulf-flattened-firm-not-as-advertised-2012-04.md). Probation argument: this is the second consecutive day a find arrived from employer-reported hierarchy data rather than résumé inference — the genre opened as a candidate on 2026-09-22 precisely because this base is oversupplied with one structural instrument (the escalated Résumé-Panel Consensus). A second signal requires a different paper from this programme carrying an organizational outcome — another summary of the same 1986–2006 panel does not promote it, and neither does a wage-and-promotion result. Expiry 2026-11-21. Not a candidate: hbs.edu, which is an access route rather than a source, per the BNN Bloomberg distinction (2026-09-18) and the doku.iab.de call (2026-09-22).