Stafford

← 2026-11-01 · all briefs · 2026-11-03 →

Stafford Brief — 2026-11-02 (Monday)

Research-evidence: branch claude/modest-faraday-jeqxe3 pushed (1 KB file, 1 staged entry 2026-11-02-journal-of-econ-why-organizations-fail-models-and-c; scripts/verify.sh passed), confirmed merged to its main as 04d5016 by the promote-claude-branches workflow. Brief emailed to brandon.freitag@gmail.com.

OPS. Clean run. Started from main at 7662d1c; both repos fetched and synced before work began. Yesterday's run merged (0d3677b, and its research-evidence addendum folded into that repo's main as 5863ab1), so there is no OPS flag to carry. Today's 12:00 scan had landed and merged (1415542 → c7ecddc), so this brief works from today's scan, not a stale one. Not Wednesday — no falsification sweep. Not Friday — no weekly synthesis. Research-evidence branch and ship disposition are on the last line.

What this brief is. The scan's brief this morning is the best it has written and I am repeating none of it. I spent the day on the one check it queued and could not afford, and that check produced the most consequential document this base has acquired in nine weeks — against Brandon, from a field nobody was looking in, reached in three calls for free. I also have to take back part of this morning's headline. That is five consecutive days a registered publication sentence has not survived verification, and today's failure is different in kind from the previous four: it is not a misread number, it is a claim the paper does not make and the scan's own entry contradicts two paragraphs later.

📖 TODAY'S READ

Somebody already wrote the Five Breakpoints, ten years ago, in the Journal of Economic Literature, and Brandon has never cited the field it came from.

Luis Garicano and Luis Rayo, "Why Organizations Fail: Models and Cases," Journal of Economic Literature 54(1), March 2016, pages 137–192. I found it in the reference list of the paper the scan ingested this morning, chased it to a free full text, and read it end to end. File: knowledge-base/garicano-rayo-why-organizations-fail-models-cases-2016-03.md. Staged in research-evidence as 2026-11-02-journal-of-econ-why-organizations-fail-models-and-c.

What it is. A survey of why organizations fail, organized under two root causes — people don't want to act in the organization's interest, or they don't know enough to — and then split into five failure families: who gets to decide, short-termism, communication that gets filtered on the way up, resistance from people who'd lose, and talent put in the wrong seats. One simple model per family. One named disaster per family: Daimler–Chrysler, AIG, BP Texas City, the NYSE's failure to go electronic, HP's inability to escape the "HP Way," the Spanish Cajas.

Why this is a bad day and not a good one. The relabeling objection has sat in the challenge ledger since 17 September with a precise win condition written into it: a prior work treating three or more of the five breakpoints as a single interacting system of failure conditions, with a diagnostic attached. The asset-12 entry has been blocked on that search for forty-six days and pre-registered both outcomes in writing — empty means a victory lap, a hit Brandon hasn't cited means "this is not a section, it is a revision."

It is a hit, it is five of five rather than three, and the interaction is claimed out loud: "decentralized authority... aggravates all other incentive and communication problems we have discussed." Brandon's strongest defence — nobody integrated these; they're five separate literatures in four fields across four decades that don't cite each other — is false as written. Economics integrated them in 2016. I checked his whole corpus today: papers/, the book manuscript, five-breakpoints-source.md, searched for Garicano, Holmstrom, Milgrom, Gibbons, "multitasking," "organizational economics." Zero hits. He has never cited the field.

The two defences that survive, survive stronger — and that is the whole of the good news, so take it precisely. There is no diagnostic anywhere in the survey — no instrument, no checklist, no test an organization runs on itself. And the authors don't merely lack one; they list the fact that short checklists work among the things their literature cannot explain (they cite the ICU catheter checklist that cut bloodstream infections 82%). The field that integrated the failure families says in print that building a diagnostic is the thing it can't do. And on speed, they concede it themselves: the models are static.

So what Brandon takes from this, in one sentence: the Antecedents section just got worse and the piece just got better — conceding an antecedent your reader didn't have in mind is a completely different act from conceding the one they did, and "I read outside my field and here's who got there first" is a stronger opening than any of the three claims it replaces.

🔭 THINKER TO WATCH

Luis Garicano — and the point is not that he's an unknown it's safe to concede to. He's the opposite.

LSE professor of economics and strategy when he wrote this, later a Member of the European Parliament, widely published for general audiences, and his co-author Luis Rayo is at Utah and CEPR. This is not obscure prior art. Toyama was obscure enough that conceding him read as scholarship. Garicano is the kind of name a reviewer recognizes, which means the concession has to be volunteered rather than extracted.

And I have to restate the market lane the scan routed this morning, because the cheap check it queued came back and it does not say what either of us expected. The scan's claim was that the AI-safety researchers "name these as failure modes of multi-agent systems and never connect them to the literature that catalogued the same failures in people" — and it flagged, honestly, that it had not read the reference list and the claim would weaken if that tradition were cited.

I read the whole reference list today — all sixty entries, main text and appendices. Result:

So "they cite none of the human organizational literature" is now false and must not be said. The accurate and still-valuable version: they reached for a theory of organizational failure and reached into economics, not into the team literature — so the unclaimed bridge is specifically to team effectiveness, and the bridge they did build is to the tradition that just cost Brandon his integration defence. The lane is narrower than the scan said and the reason it exists is more interesting: two literatures are describing the same failures and neither has the diagnostic.

🏢 CASE IN THE WILD

The scan had no human organizational case today and said so honestly. I have four, they are forty years old, and the mechanism is the one Brandon's Paper 1 is built on.

Garicano & Rayo's Section 5.1 has a formal result — call it the veto result — and it says this: a firm that is already successful will struggle to profit from a new technology, because the people whose skills don't fit the new technology would rather block it and keep extracting what their old skills are worth. Their phrase: "firms with strong goodwill do not adopt radical technological changes."

Then the cases:

That is Paper 1's Incentive Fragmentation case — one executive with veto power stalling a migration — with a model under it, four cases around it, and two remedies the framework doesn't have. The remedies: pay the losers off in advance (they cite Nabors' Eugene Isenberg relinquishing control for a chairmanship, and Eric Schmidt being "kicked upstairs" at Google), or don't let blocking power concentrate in the first place, spreading authority across people with different kinds of expertise.

The open question, and it is a live one for a client room this quarter. Their argument is that the remedy has to be in place before you know which direction the change will come from — their words, "a priori it may be hard to know the direction of change and therefore who the losers will be." Brandon's Tier 3.2 instrument runs the test at engagement entry: name the two or three people whose numbers won't improve if this works, and get a dated written concession. If Garicano & Rayo are right, a test run at engagement entry — on a transformation that has already been scoped, where everyone can see who loses — is catching the problem late. So: is the Commitment test a diagnostic or a governance design? If it's a diagnostic, entry is the right moment. If it's a design, the right moment was three years ago and the honest finding at entry is you concentrated blocking power and it's too late to unconcentrate it cheaply. I think it is both and the framework should say which it is doing.

⚙️ FOUR FORCES CONCLUSION

Commitment gains a formal result and a timing condition, and the timing condition is the part that is new to the framework.

The scan paired two Commitment findings last week and this morning, correctly: the ONS gradient (real firms, follow-through on a declared AI plan at 17% / 31% / 48% across the management-practice spread — real behaviour, no identification) and this morning's agent experiment (incentives manipulated rather than observed, identification but no real organizations). One has reach, the other has control, and they point the same way.

Today adds a third thing neither of them has: a reason the force fails that is about when the design was made rather than how hard anyone tried. Garicano & Rayo's remedies are commitment devices chosen before the direction of change is known — diffuse the authority now, compensate the likely losers now, because later you will know who they are and it will be too late to do either credibly. Commitment in the framework is currently about follow-through under pressure: acting consistently with stated direction when it gets expensive. This says a large part of Commitment is spent before the pressure arrives, in how authority was allocated — and that an organization can be structurally incapable of follow-through regardless of how committed its leaders are.

That is a genuine addition, and it is the one place today's find helps rather than costs. It also gives the force something it lacked: an account of why well-intentioned follow-through fails that isn't about character.

💭 OPEN QUESTION

The scan asked you to decide whether the structural null goes in Paper 4. I've verified it and the answer changed: the question is no longer whether to use it, it's whether there is anything there to use.

See the Challenge and the Routing block below for the verification. Short version: the paper reports no number at all for structure — no coefficient, no interval, no significance test, nowhere in the main text or appendices. The estimates exist only as a picture in Figure 4. The one significance marking in the paper belongs to a different table about which model was used.

So the real question is the one underneath, and I can't answer it for you. The scan's own standard today was that "an unmeasured outcome and a measured null are different findings" — it used that standard to refuse merging its new pattern candidate with an existing one, and it was right to. Apply it one paragraph earlier and the structural null is on the wrong side of its own line.

Here is what I actually want your read on. For about eleven weeks this base has been accumulating a complaint about the AI-and-structure literature: nobody connects a structural change to an outcome. Today it accumulated a second complaint, and the two are in tension. The first says measure the outcome. The second — today's — says when someone reports a null, check whether they measured it or just didn't find it. Both are correct and the second is much more expensive to run, because it means every null this base treats as evidence needs its interval checked, and most of them don't have one. Do you want that standard applied retrospectively? It would be the largest piece of re-verification this system has ever attempted and I think it would find more than one load-bearing null that is really an absence. I am not proposing it. I am telling you the question exists and that today is the second time in a week the answer would have changed a brief.

🥊 CHALLENGE

Two things, and they go in opposite directions. The first is against Brandon. The second is against this morning's brief.

(1) The relabeling objection's strongest limb is gone. Covered above and written up in full at knowledge/thesis-challenges.md, Update 2026-11-02. Status stays CONTESTED — the objection does not win outright, because the diagnosis and speed defences survive and are now evidenced by the opposition's own survey. What would make it win outright is narrower than it has ever been: a framework that integrated these failures and instrumented them. CFIR is the only remaining candidate and has never been checked — the 09-17 ledger named it "the most dangerous of the five" and no slot has been spent on it in six weeks. I recommend the next exploratory slot go there, ahead of any case-file gap. It is the last thing standing between the contribution claim and a clean answer.

And note how the search failed for forty-six days, because it will fail the same way again. The ledger named five candidates to check — Beer & Nohria, Burke–Litwin, McKinsey 7-S, Pettigrew, CFIR. Every one is change management or implementation science. The integration was in economics. The objection was specified as a priority dispute inside Brandon's own field and it is a cross-disciplinary one.

(2) Do not publish "structure was the free variable and it did not pay." The paper does not report a structural null — it reports no structural result at all, and this morning's own entry contradicts the sentence two paragraphs later.

I read arXiv:2604.10290 today, main text and Appendices F–H. Findings:

What survives, and it is publishable — it is just a different sentence. The only organizational variable these authors could put a number on was whether the agents' incentives pointed the same way. They rearranged the org chart ninety different ways and reported no number for it at all. That is true, checkable, and it is a claim about what the field can measure rather than about what organizations are like — which is the safer and more interesting claim, and it doesn't need the simulation caveat to do its work. The sentence it replaces would have been handed back by the first competent reviewer, and the reply would not have been "that's a simulation." It would have been "where's the test?"

Five in six days. 10-28: a self-classified share read as organizational state. 10-30: a composition claim survived where a shape claim didn't. 10-31: a coefficient from an uncentred interaction read as a population average. 11-01: an unadjusted two-group contrast read as a causal effect. 11-02: an unreported null read as a measured zero. The first four were all misread numbers. This one is a number that isn't there, which is a new failure mode and the harder one to catch — yesterday's proposed rule (state the population and state what was held constant) would not have caught it. The rule it needs is one line longer: and if the claim is that something had no effect, say what the interval was. Filed as asset-12 material alongside the other four.

🪟 WINDOW FLAG

None, and today's decline is unusually clear-cut. A 2016 survey article has no clock on it — CLAUDE.md rule 5, an old document is argument value, not a publication window. It is worth more than any window this base has declined in thirty-two days, and it is worth it to master-plan.md Tier 2.1, not to a timed post.

One note on the scan's registered asset, which is a withdrawal of a sentence rather than of the asset. Same shape as yesterday: the asset keeps its slot and its inventory count, and the sentence is replaced. PROPOSED-UPDATE in Routing.

🧱 ASSET SUGGESTION

No new asset. Asset 12 is unblocked and its scope changed, which is the amendment logged today at knowledge/asset-suggestions.md. Hard stop 1 holds: nothing new is numbered.

What changed, in one line each. The forty-six-day block is cleared. The answer was the expensive one, so #12 is a revision rather than a section, exactly as the entry pre-registered. Effort moves from low to moderate. It now discharges three challenges rather than one — relabeling, the McAdam standard, and no theory of who holds the remedy.

And it has a new dependency that is yours alone: I read the February 2015 pre-publication draft, not the published 2016 article. The full text is free at conference.nber.org/conf_papers/f80906/f80906.pdf; LSE holds the record but no full text, and the published version is behind the AEA. Under CLAUDE.md rule 14 nothing I quoted today may go into print until the published text is checked — section numbers and wording may differ, and that rule exists because this base has been burned twice on versions of one working paper. Either get the JEL article, or authorize drafting with every quotation flagged pending verification. Second day running the binding constraint here is a paywall rather than a question, and this time the paywalled document is load-bearing for the gate asset.

🎓 SPARRING NOTE

None issued. The 2026-10-02 decision holds and the coaching loop is still switched off pending the Tier 0.1 paragraph — ten consecutive declined notes is a signal about the channel, not about the material. Today's find is the strongest sparring material the base has ever held for growth edge #1 and it is going into asset 12 instead, where it has a chance of being used.

SOURCES & THREADS

Candidates: one promotion case, one expiry case, both PROPOSED below. The Journal of Economic Literature and organizational economics generally should be in sources.md, not in candidates — one find, but a field-defining review journal rather than a blog, cadence irregular. And the acquisition route the scan named this morning is already discharged: it logged the "managerial-myopia / CEO-career-horizon literature" as a named route it couldn't reach because the journals are gated. Section 3 of the paper in its own day's reference list is that literature, with a model, free, and already read. Short-termism as intertemporal multitasking — an agent with wide discretion and high-powered short-term incentives picking apparently-safe projects that expose the organization to catastrophe, with AIG and BP Texas City as the cases. That is Paper 3's career-calculus mechanism with a formal result behind it, and it arrived at no cost.

The route lesson, and it inverts yesterday's. Yesterday: composition is solved, reachability is the constraint. Today the most consequential document in nine weeks was reached in three calls, free, because it was sitting in the bibliography of the paper the 12:00 scan had already read. The scan spent its exploratory slot web-searching for a literature that was cited in its own ingest. The cheapest acquisition channel in this system is the reference list of what it just ingested, and nothing is systematically reading them. CLAUDE.md §Source Discovery names citation chasing as step 1 and it is being run on outlets rather than on works. PROPOSED below.

Threads: 0 due, 0 checked by me, and one thread's next check is cheaper than it was. Nearest due is Uber at 11-08, correctly timed to Q3 earnings; Regime Fork and middle-manager both 11-09. I did clear the item the scan called the cheapest high-value thing in its queue, so the middle-manager thread's 11-09 check doesn't spend itself on it:

arXiv:2604.09623, "The Hourglass Revolution" (Balaraman & Ganuthula) — read today at abstract level. It is a theoretical framework, not an empirical study. It proposes an hourglass shape emerging as AI absorbs middle-management functions, names three mechanisms (algorithmic coordination, structural fluidity, hybrid agency), and compares developed and emerging markets. No data, no sample, no firm-level layer measurement. For the middle-manager thread, whose criterion is firm-level evidence about layers, this is null — and it is the useful kind of null, because a framework paper arriving on an empirical thread is exactly the substitution that thread exists to prevent. Worth one KB file as a Category 8 design claim if the full text names what it assumes solved; not worth one as evidence. One discrepancy to flag rather than resolve: the arXiv identifier is a April-2026 number and the abstract page reports a March 2026 submission. I did not chase it and nothing depends on it, but anyone citing this should settle the date first.

Exploratory query: one, and it was the integration search — the single named blocker on the gate asset, pre-registered as never run. Logged at knowledge/query-log.md with route and yield. Both logs grepped for garicano, rayo, organizational economics, multitasking before running: zero hits in either, not a repeat. Yield: 1 KB file in both repos, 1 staged entry, 3 challenge updates, 1 asset amendment, and one defence of the thesis lost on the facts. Not a null day and not a comfortable one.

Patterns: no Stafford pattern crossed two sources, so no 📈 section. One annotation proposed against the scan's new candidate — see Routing.

Routing

strategy — the day's finding, and it is the one that changes what Brandon does next. No PROPOSED-UPDATE needed: it lands entirely in Stafford-owned files, which are already written. The integration search that has blocked asset 12 for forty-six days came back with a hit from a field nobody pointed it at. Garicano & Rayo, JEL 2016, five failure families under two root causes, interaction claimed explicitly, zero citations anywhere in Brandon's corpus. The integration defence is gone; the diagnosis and speed defences survive and are now evidenced by the opposition's own survey saying it cannot build a diagnostic and its models are static. Asset 12 is a revision rather than a section and it now discharges three challenges instead of one. Full write-ups: knowledge/thesis-challenges.md (three updates dated today), knowledge/asset-suggestions.md (amendment to #12), knowledge-base/garicano-rayo-why-organizations-fail-models-cases-2016-03.md.

strategy — do not publish this morning's asset sentence. PROPOSED-UPDATE to research-evidence/knowledge/publication-windows.md, anchor: Scan 2026-11-02, "Asset registered" and the inventory line.

CORRECTED 2026-11-02 (Stafford, 14:30 layer). The registered sentence is withdrawn and replaced; the asset is not withdrawn and the inventory count is unchanged at seven.

"Structure was the free variable and it did not pay." The paper reports no structural result to pay or not pay. Verified against arXiv:2604.10290 (main text and Appendices F–H, read 2026-11-02): no coefficient, standard error, confidence interval or p-value is reported for any structure feature — hierarchical, hub-and-spoke, flat or random — anywhere in the paper. Figure 4 presents the organizational-feature estimates graphically only, with no accompanying numeric table. The paper's only significance marking (p < 0.001) attaches to Table 2, which concerns model choice, not structure. The strings not significant, no significant, statistically and power do not occur in the paper, so the authors assert neither a precise zero nor that their design could have detected an effect. The sweep is 90 randomly sampled organizations and §5.1 does not state the rollouts per organization — the 15 and 50 figures belong to §4.1's main experiment, not the sweep.

Additional condition not recorded this morning, and it is internal to the entry rather than to the paper. The same KB entry tags Process Friction on the finding that hierarchical and hub-and-spoke "slightly reduces both the business efficacy and ethics of proposals" and "may introduce inefficiencies in communication, creating organizations that fall below the Pareto frontier." That is a measured structural effect, and it cannot coexist with "structure did not pay." Either structure moved outcomes — in which case the asset sentence is wrong and the Process Friction tag is earned — or it did not, in which case the tag is unearned. The paper's actual structure result is a small, unquantified negative effect of the two most company-shaped topologies, which points the opposite way from the registered sentence and is the more useful finding.

Replacement asset sentence, and it is a claim about the field's measurement rather than about organizations: "The only organizational variable these researchers could put a number on was whether the agents' incentives pointed the same way. They rearranged the org chart ninety different ways and reported no number for it at all."

Conditions that travel with the replacement: it is a statement about what this paper reports, verifiable by anyone who opens it, and it must not be restated as "structure doesn't matter" in either direction — an unreported null is not a measured zero and this correction does not license the opposite claim either. The incentive-alignment half of the paper is unaffected and remains its strongest result. The simulation limits recorded this morning all still bind.

strategy — the pattern candidate opened this morning is standing on an unreported null. PROPOSED-UPDATE to research-evidence/knowledge/emerging-patterns.md, anchor: The Structural Null candidate, 2026-11-02.

Annotation added 2026-11-02 (Stafford). The candidate is defined as "where structure is varied and outcomes are measured, structure does not move them" and has one member, arXiv:2604.10290. That member does not report a measured null — no interval, no test, no significance claim; the estimates exist only as a figure (verification above). The candidate's own refusals apply to its only member. This morning's note refused the obvious second members on exactly the right ground: "an unmeasured outcome and a measured null are different findings — merging them would let this candidate borrow confirmation from an absence of evidence." Applied one paragraph earlier, that standard puts the founding member on the unmeasured side of its own line. Recommend either holding the candidate at zero members until a structural null with a reported interval arrives, or restating it as what it can actually support: "where structure is varied, the result is reported without an interval" — a pattern about the literature's reporting, which this base could populate quickly and which would be a real finding about a field that keeps publishing structural claims it does not test. I prefer the second. The story thread the scan opened today ("Does the multi-agent alignment literature ever find a structure that works?", next check 12-14) is well-founded either way and should keep its pre-registration.

market — restate the lane before Brandon uses it, because the cheap check came back and part of it is now false. PROPOSED-UPDATE to research-evidence/knowledge/field-map.md, anchor: the 2026-11-02 multi-agent alignment experimentalists entry.

Amended 2026-11-02 (Stafford) — the queued reference-list check is discharged. Full reference list of arXiv:2604.10290 read 2026-11-02 (sixty entries, main text and appendices). No team-effectiveness literature is cited: no Hackman, Edmondson, Lencioni, Katzenbach & Smith, Salas, Kozlowski, Marks or Mathieu. However, the paper has a "How Human Organizations Fail" section resting on three organizational-failure works — Garicano & Rayo (JEL 2016), McMillan & Overall (Academy of Management Perspectives 2017), Mellahi & Wilkinson (International Journal of Management Reviews 2004). So the formulation "they cite none of the human organizational literature" is false and must not be used. Accurate restatement: they reached for a theory of organizational failure and reached into the economics of organization rather than into team effectiveness. The differentiation claim survives narrowed — the unclaimed bridge is to the team-effectiveness tradition specifically — and it is now more interesting, because the bridge they did build is to the literature that supplies the strongest version of the relabeling objection against Brandon. Two fields are describing the same organizational failures and neither has a diagnostic.

strategy — the route the scan logged as gated this morning is already discharged, free. PROPOSED-UPDATE to research-evidence/knowledge/source-candidates.md, anchor: the 2026-11-02 managerial-myopia / CEO-career-horizon acquisition route.

Discharged 2026-11-02 (Stafford). The route was logged as a named literature the scan could not reach (SSRN 403; the confirmable papers having investment inputs rather than capability outcomes on the left). The literature's canonical survey treatment was in the reference list of the same day's ingest, is free, and has been read. Garicano & Rayo (JEL 2016) Section 3 models short-termism as intertemporal multitasking: an agent with wide discretion under high-powered short-term incentives "selects apparently-safe projects in order to capture short-term rewards at the cost of placing the organization at risk of catastrophic failure," with AIG and BP Texas City as cases and the Baker Report quoted on BP's short-term performance contracts. This is Paper 3's career-calculus mechanism with a formal result and two named disasters attached, acquired at no cost. The route is not closed — firm-level panel evidence on CEO horizon and capability outcomes is still ungated-unreachable — but it now has a theory anchor the base lacked, and the gap is narrower and better specified: what is missing is the empirical test of a mechanism that is already formalized, not the mechanism.

strategy — add a source, and it is a journal rather than an outlet. PROPOSED-UPDATE to research-evidence/knowledge/sources.md, with the probation exception stated.

Proposed 2026-11-02 (Stafford): add the Journal of Economic Literature and, through it, the organizational-economics survey literature, cadence irregular. This breaks the probation rule (never add a source to sources.md on first contact) and the breach should be explicit so Brandon can refuse it. The argument for the exception: probation exists to keep unproven outlets out of a standing list, and the thing being added is the review journal of the field that just supplied the strongest antecedent to the framework and the formal result behind two of Brandon's own instruments. One contact produced a KB file, a staged entry, three challenge updates and an asset amendment. If that is not signal-rich, nothing is. Alternative if the rule should hold: log it as a candidate with the signal count already at one and a note that a second find promotes it. I recommend the exception and I am flagging it rather than taking it.

governance — change one line in §Source Discovery, because the most productive acquisition channel in this system is not being run. PROPOSED-UPDATE to research-evidence/CLAUDE.md §Source Discovery, step 1.

Citation chasing is specified as an outlet-discovery step and should be a work-acquisition step. The rule currently reads: "if it cites a study or report, follow to the origin. Is the originating outlet worth monitoring?" — so the output is a candidate outlet. Today the reference list of the day's own ingest contained the single most consequential document this base has acquired in nine weeks, and the same morning's exploratory slot was spent web-searching for a literature that was cited in it. Proposed amendment, one sentence: "Read the reference list of every ingested source for works, not only outlets. A cited work that bears on an open challenge, a case-file gap or a blocked asset is an acquisition item, logged by name; the outlet question comes second." Cost: minutes per ingest. Demonstrated value today: one gate asset unblocked and one acquisition route discharged that the same day's exploratory query had declared gated.

governance — yesterday's proposed rule needs one more line, and today supplied it. Yesterday I proposed: state the population the number describes, and state what was held constant; if nothing was held constant, the number describes the groups and not the cause. That would not have caught today's failure, because today's failure is a number that does not exist. The rule as it should now read: state the population, state what was held constant — and if the claim is that something had no effect, state the interval. A null without an interval is a failure to detect, not a finding. Five dated instances in six days now attach to it (10-28, 10-30, 10-31, 11-01, 11-02), four of them misread numbers and one of them a missing one. Still the best material asset 12's checking section will ever have, and still unanswered.

personal — unchanged, now twenty-three days old, and today makes one of them cheaper and one of them harder. master-plan.md Tier 0.1 (one paragraph, fifteen minutes, forty-six days old) and Tier 0.2 (the post-engagement measurement, the only time-gated item on the plan, no decision recorded). Tier 0.1 is now easier to write than it has ever been — the hostile reader who knows Kotter cold has been joined by a hostile reader who knows Garicano, and the paragraph that answers both is stronger than the paragraph that answers one. Tier 2.1 is harder and more valuable than it was this morning. And the new ask is small: get the published JEL article, or tell me to draft from the 2015 version with every quotation flagged. Also still open: the OPENCLAW_AGENT_ID provenance defect, now seven days old — one line in the script or the environment, and the two-writer split depends on being able to tell the writers apart.

MERGED 91a3112 main. Research-evidence addendum merged to its main as 04d5016. Brief emailed to brandon.freitag@gmail.com.