A Legal Transaction Has a Bill of Materials. Here Is One.

Two rows, condensed from a published table:

Node 32 · Draft proyecto de estatutos: objeto social, capital, partes sociales, admission rules, foreigner-admission clause, governance · Owner: MX counsel · Depends on: 30, 31 · Call: AUTOMATE (lawyer review retained) · Unit cost: $1,000 (modeled). Node 45 · Execute escritura constitutiva before notario público (LGSM art. 5) · Owner: notario · Depends on: 44 · Call: HUMAN-ONLY · Unit cost: $1,750 (modeled honorarios) · Resistance: INSTITUTIONAL.

Node 45 does not move with model capability. Only a treaty or a statute can move it · and it sits on the critical path. Every model release that improves the draft in Node 32 changes nothing about the appointment in Node 45, because the act that makes a Mexican company exist is not a document. It is a fedatario giving a document public faith, and that is a fact about the Ley General de Sociedades Mercantiles, not about software.

The table these rows come from is the legal BOM: a complete bill of materials for one legal transaction, published in full alongside this memo. This piece is not an essay about it. The artifact is the argument. What follows is the instruction manual.

Eighty-four numbered tasks

The matter is Project Saltillo, a representative cross-border matter modeled for publication · a hypothetical built to be typical, not drawn from any client file: a US parent entity (a Delaware LLC) plus a Mexican operating subsidiary (an S. de R.L. de C.V.), taken from engagement through operational closing, both entities formed, tax-registered, banked, and intercompany agreements in place.

That matter decomposes · completely, not illustratively · into 84 numbered task nodes across five phases. Every node carries an owner, its dependencies, modeled hours, a modeled fully-loaded unit cost, an automation call (AUTOMATE, ASSIST, or HUMAN-ONLY) with a one-line rationale, and, where the call is HUMAN-ONLY, a resistance class that says why. Here is a twelve-row excerpt chosen to straddle the seam the whole piece turns on:

The Legal BOM · Project Saltillo · 84 tasks · $23,786 all-in (modeled) · full table published with this memo

Two prior efforts deserve to be named, because each got halfway here. LawFlow, an academic project published in 2025, mapped entity formation as task workflows · tasks, subtasks, dependencies · and published no dollar column (arxiv.org/abs/2504.18942). The California Management Review ran a careful piece in April 2026 on AI in M&A that decomposes a deal into six phases · and stops at phases, never reaching tasks (cmr.berkeley.edu). Workflows without dollars; phases without nodes. The missing column was never hard to build. It is assembled from public statutes, published government fee schedules, and a stated rate card, and the full methodology is printed so that any operator can recompute or refute any row. The interesting question is why no one had published one.

The silence is the evidence

The constraint was never measurability. It was publishability.

Task-level economics are the most competitively sensitive data a law firm holds. The billable hour is priced precisely so that the client never sees the parts list: the invoice says corporate formation, 42 hours, not Node 10, one-page statutory form, fully templated, $48. No incumbent will publish a BOM, and not because they cannot · the incentive runs the other way. Anyone whose pricing power rests on the bundle being opaque is structurally committed to the bundle staying opaque.

You can see the same silence from the buy side. The diligence literature on legal MSOs is genuinely good on contract design and fee mechanics · L.E.K., advising investors on how to evaluate these deals, gets to the heart of the structure in one sentence: "Since the MSO fee is paid before partner distributions, successful deals build alignment through fair market value fees, scope adjustment terms and rollover equity." But nowhere in that literature, including the most-circulated investor whitepaper in the category, does a single task-level number appear. Even the AI-native entrants who publish anything publish prices · what they charge per step · never costs, which is what it takes. Prices are a marketing decision. A BOM is an operating confession.

So the empty record is not evidence that the artifact is impossible or worthless. It is evidence of what it costs the publisher. Which is exactly why the natural publisher is an MSO: a management services organization sells operations, not opacity. Our pricing interest runs toward the parts list being public. That disclosure is doing credibility work here, and we intend it to.

An exposure percentage is wrong even if it is true

Into the vacuum flowed the aggregate exposure percentage. Clio's 74 percent and Goldman's 44 percent are the foils here, named once and used for nothing: marketing-grade aggregates that this memo builds no arithmetic on. But the deeper problem is not that any given percentage is inflated. It is that an exposure percentage is the wrong kind of number, and it stays wrong even when it is accurate.

Automatability is a property of a node's position in the graph, not of the node alone. An AUTOMATE node that sits downstream of a HUMAN-ONLY gate contributes zero cycle-time savings and only partial cost savings: Node 32's automated draft still waits for the name authorization in Node 30, and everything after it still waits for the notario in Node 45 and the in-person SAT inscription in Node 51. A percentage cannot see dependencies. It treats the matter as a bag of tasks when the matter is a graph, and savings live on the graph. Scholars saw the task-level structure a decade ago · Remus and Levy famously scored lawyering by task category · but scoring categories is not assembling a matter, and the unit of analysis that survives contact with a real matter is the positioned node: this task, with these dependencies, owned by this role, at this cost.

Project Saltillo makes the point with its own headline numbers. 54 of its 84 nodes · 64.3 percent of tasks by count · are automatable. Hold that number next to what the curve below says it is worth.

Unit cost includes the checking

Position is half the failure of the percentage. The other half hides inside every "automated" node, and it is the review.

In an assisted workflow, the dominant cost term is not generation. It is the lawyer reading what was generated, and the escalation path for the cases the system gets wrong. So every honest unit cost in the BOM includes the checking, and the model is published rather than assumed:

Methodology, in the open (everything here is modeled, and says so). Labor is priced from a stated rate card for a US-Mexico cross-border boutique: US attorney $450/hr, Mexican attorney $250/hr, paralegal-ops $95/hr · modeled market assumptions, not billing data. Government fees come from published official schedules; market-priced items (notario honorarios, courier, certified translation) are modeled and labeled in their rows. The automation model: an adopted AUTOMATE node eliminates 80 percent of its hours, an adopted ASSIST node 50 percent · and then 30 percent of every saved hour is added back as human review and escalation, priced at the node's own rate. Net of the review tax, an automate node yields 56 percent of its labor cost as savings; an assist node, 35 percent. Fixed fees and HUMAN-ONLY nodes never move. Adoption at X percent means the X percent of automatable nodes with the largest net savings are adopted first. Project Saltillo is a reference matter: no client, no engagement, no invoice behind it. Every assumption is printed so the table is falsifiable.

This is the arithmetic that quietly collapses every savings model built on exposure percentages: a task can be fully automatable in the generation sense and still return less than half its cost, because the checking was always part of the cost and the percentage never counted it.

What 0 / 25 / 50 / 75 looks like

Sum the honest node costs along the graph at four adoption levels and you get a curve nobody had plotted. This is original output of the published model; no external citation exists for it, which is the point.

64 percent of this matter's tasks are automatable by count. Adopting automation across three quarters of them saves 22.0 percent of modeled cost. Full adoption saves 23.5.

The curve is concave, and for two stacked reasons. By construction, rational adoption takes the cheap wins first: the 0-to-25 step is worth $3,780, the 50-to-75 step is worth $518. And by topology, the floor never moves: $9,506 · 40.0 percent of the matter · sits in human-only judgment nodes, pass-through fixed fees, and institutional gates that automation cannot touch under this model.

Cycle time behaves the same way without needing a chart: the clock on this matter is set by the Secretaría de Economía's response window, the IRS's nonresident EIN queue, the SAT appointment supply, and two banks' KYC processes, all of which sit on the critical path and none of which run faster because a model improved. The savings are a topology property of one matter graph, not an arithmetic property of a cost base. That single sentence is the whole correction to the exposure-percentage genre.

One pointer, and only a pointer: this memo prices what automation saves on a matter. Who captures those savings · firm, MSO, or client · is a fee-clause question, and it is the entire subject of "The AI Dividend Has an Address," an earlier memo in this series.

The notario line

Now the seam the twelve-row excerpt was chosen to straddle. The BOM's 30 human-only nodes are not one population. They split into two resistance classes, and the split is the most decision-relevant distinction in the artifact.

Capability nodes resist automation because of what models cannot yet be trusted to do: the structuring memo with long-tail tax consequences (Node 5), the check-the-box election that binds the client for sixty months (Node 62), the thin-cap judgment on the capitalization mix (Node 68), the final attorney certification (Node 81). These nodes may move. Each one carries, implicitly, a forecast tied to model generations, and an honest analyst should expect some of them to migrate toward ASSIST over time.

Institutional nodes resist automation because a statute, a treaty, or a government interface says so, and they carry no model-generation forecast because none applies:

• Notario execution. A Mexican company is constituted before a fedatario público; LGSM article 5 puts the notarial act, not the document, at the center. Drafting the estatutos is automatable. Giving them public faith is not a software function.
• Apostille. Cross-border recognition of the US corporate documents runs through the 1961 Hague Apostille Convention; only the competent authority · here the Delaware Secretary of State, at $30 per document · can issue one.
• SAT registration. The subsidiary's RFC inscription starts and concludes in person at a SAT office, by rule (ficha 43/CFF), and the entity's e.firma requires biometric capture. There is no API to a trámite.
• The IRS nonresident route. An EIN for a foreign-owned entity without an SSN/ITIN responsible party travels by phone or fax, per the SS-4 instructions. The form fill is deterministic; the channel is institutional.
• Bank KYC. Two accounts, two AML regimes (31 CFR 1010.230 on the US side, CNBV rules on the Mexican side), and in both cases the clock belongs to the bank.

We call the boundary between these two classes the notario line, and we have not found a prior piece, in English or Spanish, that draws it. It matters because the two classes answer to different forces. Capability nodes answer to model releases, which nobody can schedule. Institutional nodes answer to treaties, statutes, and agency procedure, which move slowly, visibly, and in public. Of the 30 human-only nodes in this matter, 21 are institutional. Which means the residual · the 40 percent floor under the curve · is mostly the kind of thing you can actually rely on, because the only events that move it are published in the Diario Oficial or the Federal Register. To be precise about scope: every institutional claim here is procedural. Who may own what, on either side of the border, is a different subject, and it is not this memo's.

What survives every model release

Read the BOM as an investor reads anything: where is the durable position?

It is not in the AUTOMATE column. Drafting nodes are where competition will be fastest and margin thinnest, precisely because they move with every model release and every entrant gets the same models. The durable position is the residue: the statute-backed, treaty-backed nodes on the critical path. They are the chokepoints that survive every model release, and operating them · the notario relationships, the apostille logistics, the SAT appointment pipeline, the bank KYC choreography, the custody of single-instance original instruments (Node 80) · is a different business from generating drafts. That is an observation about where defensibility lives in a matter graph, not guidance to anyone about what to buy.

And it explains a gap in the diligence record that should now look stranger than it did. The investor literature on legal MSOs evaluates fee design, contract durability, and alignment, and it does this well. It contains no task economics at all: no node costs, no dependency structure, no cost-to-serve curve, nothing that would let a reader distinguish a deck that says "AI-powered legal services" from an operation that is one. Those claims cannot be evaluated from a deck. They can be evaluated from a parts list in an afternoon: which nodes, what calls, what review overhead, where the floor sits, which side of the notario line the moat is on. A target that operates has one, whether or not it has ever written it down; this memo is an existence proof that publishing one is survivable. The presence of a BOM, or the silence where it should be, is the tell.

So the close of this memo, and of this series, is one sentence of diligence hygiene · the request that belongs at the data-room floor, right beside the deck: ask for the BOM.

The objection, at full strength

Here is the strongest version of the case against this artifact, stated the way a skeptic would state it. This is one matter, modeled by the party publishing it, with no auditor. The rate card is assumed, the 80/50 elimination rates are assumed, the 30 percent review tax is assumed. Pick a friendlier matter, tune the overhead, and the curve says whatever you want. And the publisher is an MSO whose business improves if readers believe operations are the moat.

Four answers, none of which is a dodge.

First, one fully specified, falsifiable matter beats any unfalsifiable aggregate. Every row prints its hours, rate, call, and rationale; the automation model is stated to the percentage point; the procedural facts cite official sources. A reader who thinks the review tax is 15 percent, or 60, can recompute the curve tonight. No exposure percentage in circulation extends its critics that courtesy.

Second, the artifact is an open spec, and the invitation is genuine: publish a competing BOM. Cost a different matter, a different corridor, a different rate card. This one is a reference spec, modeled in the open; the confession it solicits is every operator's, and a published spec is what makes measured ones comparable when they come. The category wins the moment a second one exists, and so, frankly, do we · a market that prices legal work from parts lists instead of percentages is the market an operations business wants to live in.

Third, the interest is disclosed, not hidden. Jopese is an MSO; we sell operations, not opacity, and that is exactly why we can publish what incumbents will not. Discount for it as you see fit. The discount rate on an undisclosed interest is higher.

Fourth, and decisively: the part of the artifact that carries the thesis is the part no modeling choice can fake. The notario requirement is LGSM article 5. The apostille is a 1961 treaty. The SAT trámite is ficha 43/CFF. The 40 percent floor is built mostly out of public law, and you do not have to trust our rate card to check a statute.

The exposure percentage told a clean story: count the automatable share, multiply, wait for better models. The BOM tells the true one: the savings live on a graph, the checking is part of the cost, and the floor is made of law. One of these can be forwarded to an investment committee with a straight face. It is the one with eighty-four rows.

---

This memo is published by Jopese, a legal management services organization operated by HIRO PARTNERS LLC, a Texas limited liability company. It is offered for educational and analytical purposes only. It is not legal, tax, or investment advice, and it is not an offer to sell or a solicitation of an offer to buy any security or service. Jopese is not a law firm and does not provide legal advice or legal services; legal services are delivered by an independent law firm under a separate engagement in which Jopese does not participate. Project Saltillo is a modeled, representative matter, not a client engagement; no client information appears in this memo or in the accompanying table. References to specific funds, firms, publications, transactions, and regulatory developments are drawn from public sources and are provided as market commentary, not as an endorsement, a recommendation, or a representation of any relationship. Questions and competing BOMs: rene@jopese.com.

---

Appendix · The complete bill of materials (84 tasks · modeled)

The full reference BOM behind this memo. Representative matter · modeled estimates from the stated rate card · not measured client data.

Reference BOM · representative matter · modeled estimates from the stated rate card below · NOT measured client data.

Published by Jopese, a legal MSO operated by HIRO PARTNERS LLC. Analysis only · not legal, tax, or investment advice. No client information appears in this document. Every dollar figure and hour figure is modeled from the assumptions stated in section 2; government fees and statutory procedures are grounded in the public official sources listed in section 6. Contact: rene@jopese.com.

---

1. The matter

One representative cross-border matter, taken end to end:

US parent entity (Delaware LLC) + Mexican operating subsidiary (S. de R.L. de C.V.), from engagement through operational closing: both entities formed, tax-registered, banked, and intercompany agreements in place.

The matter decomposes into 84 numbered task nodes across five phases: engagement and structuring (Phase A), the US workstream (Phase B), the Mexico workstream (Phase C), the cross-border and intercompany workstream (Phase D), and closing (Phase E). Each node carries an owner role, dependencies, modeled hours, a modeled fully-loaded unit cost, an automation call with a one-line rationale, and · for non-automatable nodes · a resistance class.

Resistance classes. INSTITUTIONAL: the node resists automation because of a statute, treaty, or government/institutional interface (notario execution under the LGSM, apostille under the 1961 Hague Convention, SAT/RFC registration, the IRS SS-4 nonresident route, bank KYC). It does not move with model capability; only a treaty, statute, or institutional change can move it. CAPABILITY: the node resists automation because of current model limits (judgment, accountability, risk allocation); it may move with model capability.

2. Rate card and modeling assumptions (all modeled)

Labor rates (modeled from stated market assumptions for a US-Mexico cross-border boutique; not billing data):

Fixed fees. Official fees are taken from published schedules (Delaware filing $110, 24-hour expedite $50, certified copy $50, good standing $50, apostille $30/document; SE name authorization free; RFC, e.firma, and RNIE inscription free; EIN free). Market-priced items are modeled and labeled as such in the row: registered agent $125/yr, US mail/office services $300/yr, courier $120, perito traductor $400, notario honorarios MXN 32,000 converted at a modeled 18.3 MXN/USD (~$1,750), Registro Público de Comercio derechos $350, testimonio copies $150. Total fixed fees: $3,595 (modeled).

Automation model (stated, recomputable):

• An adopted AUTOMATE node eliminates 80% of its modeled hours; an adopted ASSIST node eliminates 50%.
• Review overhead: 30% of every saved hour is added back as human review and escalation, priced at the node's own fully-loaded rate. Net effect: an adopted automate node retains 56% of its labor cost as savings; an adopted assist node retains 35%.
• Fixed fees and HUMAN-ONLY nodes never move. Client, government, notario, and bank time is not billed, so automation cannot save it (it can only be waited on).
• Adoption at X% means the X% of the 54 automatable nodes with the largest net savings are adopted first (rational greedy adoption, rounded to the nearest node). This is why the curve is concave by construction as well as by topology: the cheap wins go first, and the floor never moves.

3. The bill of materials · 84 nodes

Column legend: Hrs = modeled billed hours (a "·" means the actor's time is real but not billed); Unit cost = modeled hours x rate + fixed fees; Call = AUTOMATE / ASSIST / HUMAN-ONLY; Resistance = INSTITUTIONAL / CAPABILITY for human-only nodes, n/a otherwise.

Phase A · Engagement and structuring

Phase B · US workstream (Delaware LLC parent)

Phase C · Mexico workstream (S. de R.L. de C.V.)

Phase D · Cross-border and intercompany

Phase E · Closing and handoff

---

4. Roll-up (modeled)

Node counts by call: 25 AUTOMATE · 29 ASSIST · 30 HUMAN-ONLY (21 INSTITUTIONAL, 9 CAPABILITY). 54 of 84 nodes · 64.3% of tasks by count · are automatable.

Modeled matter cost at 0% adoption: $23,786 = $20,191 labor (US attorney 30.5 hrs / $13,725 · MX counsel 11.5 hrs / $2,875 · paralegal-ops 37.8 hrs / $3,591) + $3,595 fixed fees.

The non-movable floor: $5,911 of human-only labor + $3,595 of fixed fees = $9,506 · 40.0% of the matter · sits in nodes that automation cannot touch under this model: judgment nodes (CAPABILITY) and statute/treaty/government-interface nodes (INSTITUTIONAL), plus pass-through fees.

5. Cost-to-serve at 0 / 25 / 50 / 75% automation adoption (modeled)

The curve is concave: the 0-to-25% step is worth $3,780; the 50-to-75% step is worth $518. Savings are a topology property of the matter graph, not an arithmetic property of the cost base.

Headline finding (one sentence): 64% of this matter's 84 tasks are automatable by count, yet 75% adoption saves only 22.0% of modeled cost (full adoption, 23.5%) · because 30% of every saved hour returns as review overhead and a $9,506 floor (40% of the matter) sits in human-only judgment nodes, fixed fees, and institutional gates · notario execution, Hague apostille, SAT/RFC registration, the IRS nonresident EIN route, bank KYC · and that institutional residue is fixed by statute and treaty, so it moves only with treaty, statute, or institutional change, never with a model release.

6. Sources (procedural facts · public official sources, fetched 2026-06-10)

US workstream:

1. IRS, Instructions for Form SS-4 (Rev. December 2025): international applicants without SSN/ITIN apply by phone +1-267-941-1099, by fax (Fax-TIN, generally ~4 business days), or by mail (EIN International Operation, Cincinnati, OH 45999); no online EIN without a valid SSN/ITIN · https://www.irs.gov/instructions/iss4
2. IRS, About Form SS-4 · https://www.irs.gov/forms-pubs/about-form-ss-4
3. Delaware Division of Corporations, Corporate Fee Schedule: LLC Certificate of Formation $110; expedited tiers from 24-hour service through 1-hour ($1,000) · https://corp.delaware.gov/fee/
4. Delaware Division of Corporations, LLC/LP/GP (Alternative Entity) Franchise Tax: flat $300/yr, due June 1, no annual report · https://corp.delaware.gov/alt-entitytaxinstructions/
5. Delaware Division of Corporations, Certifications, Apostilles and Authentication: $30 per document; apostilles issued under the 1961 Hague Convention, to which Mexico and the United States are parties · https://corp.delaware.gov/apost_info/

Mexico workstream:

1. Secretaría de Economía, Autorización de uso de denominación o razón social: free, response within a maximum of 2 business days, requires e.firma, authorization valid 180 calendar days; filed via the MUA portal (mua.economia.gob.mx) · https://www.gob.mx/public/tramites/detalleTramite.xhtml?homoclave=SE-09-041 · https://e.economia.gob.mx/tramites/autorizacion-de-uso-de-denominacion-o-razon-social/
2. Ley General de Sociedades Mercantiles: art. 5 (constitution and amendments before fedatario público), arts. 58-61 (S. de R.L.: partners liable to the amount of their contributions; minimum 2, maximum 50 socios), art. 64 (at least 50% of each parte social paid at constitution), art. 73 (libro especial de socios) · https://mexico.justia.com/federales/leyes/ley-general-de-sociedades-mercantiles/ · official text: https://www.diputados.gob.mx/LeyesBiblio/doc/LGSM.doc
3. SAT, ficha 43/CFF, Solicitud de inscripción en el RFC de personas morales en la ADSC: started and concluded in person at a SAT office by appointment; requires protocolized constitutive document, proof of tax domicile, notarial power and official ID of the legal representative; partners' RFCs (or written manifestation) per CFF art. 27 · https://www.sat.gob.mx/gobmx/Paginas/Ficha_43_cff.html
4. SAT, ficha 105/CFF, e.firma: in-person issuance with biometric capture · https://www.sat.gob.mx/gobmx/Paginas/ficha_105_cff.html
5. Secretaría de Economía, RNIE preguntas frecuentes: Mexican companies with foreign investment must register within 40 business days of constitution / foreign-capital entry (Ley de Inversión Extranjera art. 32); late filing sanctioned at 30-100 UMA per day · https://rnie.economia.gob.mx/RNIE/faces/preguntas.xhtml
6. Reglamento de la Ley de Inversión Extranjera y del RNIE · https://www.diputados.gob.mx/LeyesBiblio/regley/Reg_LIERNIE_170816.pdf

Items NOT grounded in official sources and therefore modeled and labeled as such: all labor hours and rates; registered agent, office services, courier, translation, notario honorarios, RPC derechos, and testimonio fees; the FX rate; the automation elimination rates and the 30% review-overhead assumption.

7. Honesty block · what this artifact is and is not

• This is a reference BOM for a representative matter, built entirely from the modeled rate card and assumptions in section 2. It is not measured client data, not an invoice, and not a price quote. No client, no affiliated firm, and no attorney is identified, and none of the figures derive from any client engagement.
• Procedural steps and official fees are grounded in the public sources of section 6; everything else is modeled and says so. Any operator can refute any row: the methodology is published precisely so the table is falsifiable.
• The adoption curve is a cost-to-serve property of this one matter graph under the stated assumptions. It is not a forecast of any firm's profits and not an extrapolation to any share of all legal work.
• Resistance-class labels state where a node's resistance comes from today. Institutional nodes do not move with model capability; only a treaty, a statute, or institutional change can move them. Capability nodes may move.