The Doctrine Factory

TRADOC Pam 71-20-3 opens by drawing the line that organizes the whole factory. Doctrine "provides fundamental principles by which the military forces or elements thereof guide actions in support of national objectives. It is authoritative, requiring judgment in application." It describes the current and near-term force, provides a common language, drives how the Army is organized and equipped, and is the basis for all training and education. Concepts describe how future forces may operate: joint concepts cover 8–20 years out, Army concepts 6–18 years. Concepts originate from exactly two sources: "the rise of a new military problem that currently available DOTLMPF capabilities cannot adequately answer," or "the recognition that existing military problems can be addressed more effectively through employment of new technologies, new methods, new organizational approaches, or other forms of change." A concept is warranted only when professional military judgment concludes no viable solution exists and the problem cannot be solved by incremental change to doctrine, organization, training, materiel, leadership, personnel, or facilities (DOTMLPF).

March (1991) argued that adaptive systems drift toward exploitation because its returns are nearer, surer, and more legible; organizations that want exploration must protect it structurally. The Army Concept Framework (ACF) is that protection made bureaucratic. It comprises a fixed hierarchy: the Army Capstone Concept (TRADOC Pam 525-3-0, approved by the Chief of Staff of the Army), the Army Operating Concept (525-3-1, approved by the Commanding General of TRADOC), six Army Functional Concepts — mission command, intelligence, movement and maneuver, fires, protection, sustainment, each owned by a Center of Excellence and approved by the Director of the Army Capabilities Integration Center (ARCIC) — plus leadership-directed concepts and legacy concept capability plans. Below the framework sit two advisory genres, CONOPS and white papers, which any TRADOC organization may initiate and a general officer may approve. Note the gradient: the more exploratory and informal the artifact, the lower the approval authority and the weaker its claim on the institution. Exploration is licensed, but its license is stamped by rank.

The seam between the two artifact classes is load-bearing because concepts are the intake of the requirements machine. Concepts yield required capabilities, written in a mandated five-part grammar — who, what, where/when, why, plus a back-reference the pamphlet calls "integrity of intent" — and those capability statements feed the capabilities-based assessment (CBA) of the Joint Capabilities Integration and Development System (JCIDS), which identifies gaps and proposes DOTMLPF solutions. The guide is explicit that this provides "traceability of all Army system and non-system solutions back to overarching national strategic guidance." In Walsh and Ungson's (1991) terms from Session 02: the retention bins are wired to the procurement system. What gets written into a concept determines, years later, what the institution buys, how it trains, and eventually what its doctrine can say. Ideas that never clear the concept pipeline never become capabilities; the factory's intake filter is a filter on the possible.

Development begins only on authority: the CG of TRADOC orders a concept, a new assessment identifies the need, or the ARCIC Campaign Plan directs a periodic review — Centers of Excellence "do not initiate concept development without approval from CG, TRADOC." A signed program directive (PD) names the concept, the general-officer guidance, the milestones, and — critically — the writing team's core and supporting members; a terms of reference extends the roster across organizational boundaries. The team runs a design session, often commissions a concept study ("usually longer than the concept" — the argument that lets the product be short), then tests the key ideas in a limited objective experiment or wargame against a red team — "a sanity check to ensure the study's key ideas are worthy of further development," explicitly not a substitute for post-approval experimentation.

1. 0.1 Writing team
   Initial draft. Produced by the chartered core team after the design session, literature review, study, and limited objective experiment. Consensus among core members is required before any version advances — the team lead's version number is a certification of that consensus.
2. 0.3 Action-officer review
   First staffing. An official TRADOC tasking sends the draft plus a blank comment resolution matrix (CRM) to the whole community of practice; at least 30 days for review. Each organization consolidates its comments into one CRM. Optional seminars convene subject-matter experts between versions.
3. 0.5 Colonel (O-6) review
   Second staffing. Same mechanics, but each organization's consolidated CRM now requires O-6 approval before submission; 30 days inside TRADOC, 45 for outside organizations. The adjudicated CRM from 0.3 travels with the draft — reviewers see how their last round of objections was disposed of.
4. 0.7 General-officer review
   Final staffing. CRMs now require general-officer approval. Unresolved critical comments go to an adjudication board whose members' rank matches the version — colonels for 0.5 disputes, general officers for 0.7 — drawn from the PD's core organizations, deciding by majority vote.
5. 0.9 Approval authority
   Approval package. Version 0.9, the adjudicated CRM, a position paper on any unresolved critical comments, and a TRADOC Form 5 staffing sheet go to the approval authority. After signature: formal edit, Freedom of Information Act check, authentication by the G-6 (up to 30 more days), and publication as a 525-series pamphlet. Only web posting makes it official.

The comment typology is a three-level severity scale with real teeth. An administrative comment fixes typography; a substantive comment flags text that is "unnecessary, incorrect, misleading, confusing, or inconsistent" and equates to concur with comment. A critical comment is different in kind:

Read that as organizational design: any single stakeholder organization can block the entire artifact by filing one comment — but the veto is procedurally channeled. The comment must include specific replacement language ("relook" and "section poorly written" are rejected without action); the writing team must accept, partially accept, or negotiate; failing that, the rank-matched board votes. Galbraith (1974) gives us the vocabulary for what this machine is. The version ladder is a designed escalation path: cheap, low-rank information processing first, expensive senior attention reserved for residual conflict — exactly his rule that exceptions rise until they meet someone with authority to resolve them. The CRM is a formalized lateral relation: it forces horizontal contact between the writing team and every stakeholder, in a lean written medium, before vertical escalation. And the 25-page cap on a concept's main body, the acronym discipline, the mandated five-chapter format — these are strategies for reducing the need for downstream processing, rationing the scarce attention of hundreds of reviewers across the community of practice. The price of all this reliability is stated plainly in the document itself:

Notice what this process is optimized for: not speed, and not originality — consensus and traceability. Every objection is logged, adjudicated in writing, and archived with the draft; every capability statement carries its provenance reference. The factory produces memory that the whole institution has signed, which is precisely what makes the memory authoritative — and precisely what makes it slow. Hold that tradeoff; Brown is about to show us its cost in the field.

Brown's framing claim is historical: joint doctrine was never manufactured as joint knowledge. The services wrote their doctrine separately, and joint publications were produced by "melding" the service texts — a merge, not a synthesis. His foreword (by Michael Fischerkeller of the Institute for Defense Analyses) states the consequence without euphemism:

The paper's method is disciplined: he anchors every proposal in the nine command-and-control tenets of JP 0-2 (clearly defined authorities, information management, implicit communication, timely decision making, robust integration mechanisms, battle-rhythm discipline, responsive interoperable support systems, situational awareness, mutual trust) and the Joint Operations Concept's seven attributes for the 15–20-year force (fully integrated, expeditionary, networked, decentralized, adaptable, decision superior, lethal), then tests current doctrine against recent operations in five focus areas: supported/supporting relationships, establishing directives, synchronization of interdiction and maneuver, joint fires, and fire support coordination measures. The evidence table below compresses his case studies; each row is a place where the codified memory and the operating force diverged.

Now put the two learning loops side by side. The fast loop is field adaptation: before OIF, joint elements rehearsed time-sensitive targeting on the Nellis ranges for weeks and wrote their own tactics, techniques, and procedures; the air component then relied on those TTPs and "never produced an establishing directive" that doctrine expected. The 484th Air Expeditionary Wing ignored standard formulas and placed tactical air control parties by mission need. Marines put a liaison officer aboard AWACS to manage joint fires in real time — "it took several weeks before this ad hoc arrangement realized its full utility." In Argyris and Schön's (1978) terms this is single-loop learning: correcting action within the existing frame, and storing the correction locally — in TTPs, workarounds, and individual experience, the most volatile of Walsh and Ungson's retention bins. The slow loop is the factory: the ACCE liaison concept improvised in OIF was captured in JP 3-09.3 in September 2003 — but only for the air-to-land case; JP 3-31 gave land operations their first approved joint C2 doctrine in March 2004; the joint "battle handover" Brown wants did not exist in doctrine at all. Double-loop learning — revising the governing frame itself — happens only through the consensus machinery of Section 03, and it runs years behind the fast loop. Brown's foreword phrase for the residue is exact: recent successes "have been more reliant on the innovation and tenacity of our warriors than forward-thinking joint doctrine."

Cohen, March, and Olsen (1972) modeled organizations as "organized anarchies" in which problems, solutions, and participants flow semi-independently and attach to whatever choice opportunities happen to be open. Doctrine development is a nearly clinical instance. The choice opportunities are scheduled: publication revisions, rewrite conferences (Brown's field data arrives via briefings to an AFDD 2-1.3 Counterland rewrite conference), joint working groups. The solutions often pre-exist the problems: effects-based operations, prioritized effects lists, and effects tasking orders came out of the Millennium Challenge 2002 experiment and circulated for years looking for a doctrinal home. And the participants are the services, each carrying institutional interests into every vote. Consider the episode Brown recounts from inside the drafting of JP 3-30, Command and Control for Joint Air Operations:

Brown understands the game he is playing, and says so — his preface defines success not as adoption but as agenda-setting:

He is manufacturing a solution and dropping it into the can, hoping a choice opportunity drifts by. (It is worth pausing on the fact that the colonel who wrote this critique of the doctrine factory later ran the institutions that own it — as Air Force Chief of Staff and then Chairman of the Joint Chiefs. The factory's critics are its future managers, which is itself a retention mechanism.) Now reread TRADOC Pam 71-20-3 with garbage-can eyes and its strangest features become legible as anarchy dampers. Fluid participation is the model's chaos engine — so the program directive fixes the participant roster in writing before drafting begins, and CoE representatives "must have the authority to speak on behalf of their organizations." Unclear preferences flood choices with untraceable objections — so the CRM forces every objection into a typed severity class with mandatory replacement language, and vague comments are "rejected without action." Problems latch onto unrelated choices — so the two-source rule (new problem, or new means) and the "is a concept even required?" gate at Step 2 screen the intake. The staffing process is best read not as naive rational planning but as a deliberately engineered constraint on an organized anarchy the institution knows itself to be. The garbage can still operates — the JP 3-30 vote proves it — but it operates on a schedule, with a roster, under a comment typology, leaving an archived adjudication record. Consensus is the price of authority: memory that binds five armed services can only be written by a process every one of them can block, which is why every clause survives only in the form all parties could live with.

The joint doctrine factory is chartered by a single Chairman's instruction, CJCSI 5120.02E, whose Enclosure A is the joint counterpart to TRADOC Pam 71-20-3's opening definitions — but where the Army guide separates doctrine from concepts, the joint charter first fixes what joint doctrine is. It is "official advice and should be followed, unless the commander determines exceptional circumstances dictate otherwise"; it "focuses on how to think, not what to think, about operations"; and it takes precedence over Service and multi-Service doctrine. The charter names the machine that produces this advice — the Joint Doctrine Development Community (JDDC) — and enumerates its membership with the precision of a voting register: the combatant commands, the Services, the National Guard Bureau, and the Joint Staff J-7 are voting members; other Joint Staff directorates, combat support agencies, and National Defense University are non-voting. The definition that matters most for our lens is the charter's own account of what it is establishing:

Read that definition as an organizational chart written in the grammar of role-titles: the factory is its assignment of authorities — sponsor, agent, reviewer — and its "hierarchical framework." This is Weber's bureaucracy naming itself, and it makes the same design commitment the Army guide did in Section 03: memory is authoritative because a named ladder of offices signs it. The charter is deliberately thin, though — it establishes roles and delegates the actual process to a companion manual, exactly as the Army guide points past its own text to staffing regulations. That manual is the second document, and it is where Walker's 2003 seams meet the current process.

The current CJCS Manual 5120.01D (2025) specifies the joint process station by station, and its headline parameter is a cycle time — the joint analogue of TRADOC's 8–10 months:

Now run the comparison Walker could not. He wrote in 2003 against the process then published as JP 1-01 (2000) — the factory's self-description back when its own operating manual was itself a joint publication, before that self-description migrated into a Chairman's directive. His central charge was structural: joint doctrine was "melded" from separately written Service texts, and the resulting seams surfaced "on the battlefield," years after the field had already improvised around them. The 2025 manual answers some of that charge directly and leaves the rest untouched.

There is a genuine discrepancy between what Session 03 taught in Section 03 and what the joint manual specifies, and it should be surfaced rather than smoothed. The Army guide's comment typology is a three-level scale — administrative, substantive, critical. The joint manual runs four types: it inserts a "major" comment between substantive and critical, and it calibrates the veto differently. Where the Army's critical comment simply "indicates non-concurrence of the entire document," the joint critical comment is conditional on rank:

The joint manual then adds a rule the Army guide has no equivalent of: a critical comment submitted without a general-officer memo is automatically downgraded to a major comment, and even a properly memo-backed critical comment yields to the working group, because "the final text is determined by consensus of the JWG, as is the case for all comments." Read against Section 03's one-comment veto, this is a subtler valve. The Army guarantees any single organization a blocking objection; the joint system makes the block cost a flag officer's signature and then subordinates even that to a consensus vote. Neither is simply stronger — they are different settlements of the same tradeoff between letting any stakeholder stop the line and preventing lowest-common-denominator memory. Two factories in the same institution, having faced the identical design problem, wrote two different answers into their manuals. The seminar's job is not to reconcile them but to notice that the choice was available and made twice.

Finally, the factory's parts catalogue is itself primary evidence for this seminar's thesis. The official Joint Doctrine Hierarchy chart (status of 12 December 2024) lays out the whole library — capstone JP 1 Volumes 1–2, six keystone publications (1-0 through 6-0), the core publications hung beneath each keystone, the Joint Doctrine Notes, and the joint guides. Where TRADOC Reg 25-36 (2023) governs the Army's parallel publishing program, this chart shows the joint output in one frame. But read its legend closely.

The JDN specimen closes the loop with Section 02's exploration–exploitation seam. JDN 1-19 describes itself as "a publication intended to facilitate information sharing on problems and potential solutions as a supporting effort of formal joint doctrine development and revision," and states flatly that "The guidance in this JDN is not authoritative." It was the joint force's licensed speculation about competition below armed conflict — and by the 2024 hierarchy chart its number is gone, its content matured forward into a successor note (JDN 1-22, Joint Force in Strategic Competition) and into concepts and doctrine. That is the whole factory in one artifact's life: an idea admitted through the quarantined intake, circulated as explicitly non-binding, and either promoted into authoritative memory or rescinded. March's protected exploration, the garbage can's choice opportunity, and the slow double loop are all visible in a single publication number that appeared and then vanished from the map.

JP 5-0 defines joint planning as "the deliberate process of determining how to implement strategic guidance: how (the ways) to use military capabilities (the means) in time and space to achieve objectives (the ends) within an acceptable level of risk." Everything after that definition is organizational process, codified. Planning consists of four functions — strategic guidance, concept development, plan development, plan assessment — "generally sequential, although often run simultaneously to deepen the dialogue between civilian and military leaders." Inside them runs the seven-step joint planning process (planning initiation; mission analysis; course-of-action development; COA analysis and wargaming; COA comparison; COA approval; plan or order development), explicitly "recursive, assessment-informed … and not linear." The document even specifies the sociology of the work: planning teams are cross-functional groups led by the J-5 or J-3, with named seats for intelligence, logistics, communications, medical, legal, components, and — where authorized — allies. The publication is a communication-design artifact about decision rights and information flow, produced by the very machinery of Sections 03 and 05.

Two features matter most for our lens. First, the four levels of planning detail are a graduated codification scale — an explicit institutional answer to the question "how much shall we write down in advance?" Level 1, the commander's estimate, is little more than ranked courses of action; Level 2, the base plan (BPLAN), adds a concept of operations, major forces, and timelines but "normally does not include annexes"; Level 3, the concept plan (CONPLAN), is "an OPLAN in an abbreviated format" with named annexes, upgraded to 3T if time-phased force and deployment data is directed; Level 4, the operation plan (OPLAN), is "a complete and detailed plan" down to a transportation-feasible deployment database. In Galbraith's terms, pre-planning is pre-computation: each level buys shorter crisis-response latency at the price of authoring and — crucially — maintaining more codified material as the world drifts away from it. Hansen, Nohria, and Tierney's (1999) codification-versus-personalization economics apply line for line: codify what will be reused under time pressure; leave the rest in planners' heads and in the process.

Second, plans are governed as living memory with a standing revision loop. The plan-assessment function requires commanders to keep every plan under continuous evaluation against a changing environment:

Refinement flows continuously through fragmentary orders; adaptation reopens the operational approach when the problem itself has changed; termination requires permission to archive, staffed through the Joint Staff to SecDef or the CJCS depending on which tasking document created the plan. Even forgetting is a governed, multi-echelon process. This is the doctrine factory's answer to its own Session-04 problem from Brown: having been burned by shelfware plans and lagging publications, the institution wrote a maintenance schedule for its memory into the memory itself. And it flags the deepest caveat in its own epigraph — Eisenhower's "Plans are worthless, but planning is everything." The JPEC's civil-military dialogue, the shared staff estimates, the wargamed assumptions — much of the factory's real product is the shared understanding built in the participants, not the document. Levitt and March (1988) would recognize the design: routines encode lessons, but the encoding process is itself where much of the learning lives. The Spanish-language edition of JP 5-0 on our supplementary list completes the picture: the artifact is also an export good, a coordination standard shipped to allies so that two nations' staffs can mesh their planning processes — memory as interoperability infrastructure.

1. Is the latency load-bearing? TRADOC's cycle is 8–10 months plus editing, with a quarter of it spent waiting out 30–45-day review windows. Argue both sides: the delay is waste the institution should engineer away; the delay is the validation — consensus time is what converts an opinion into memory the whole institution will obey. Which properties of the military's environment (Session 01) decide the question, and would your answer survive transfer to a firm?
2. The one-comment veto. A single critical comment non-concurs an entire concept, but must carry replacement language and can be overridden by a rank-matched majority vote. Is this closer to a liberum veto that guarantees lowest-common-denominator memory, or to a well-designed andon cord that lets any stakeholder stop the line? What would Cohen, March, and Olsen predict happens to the content of documents written under such a rule?
3. Serendipity as evidence. Brown's operations succeeded despite the doctrine gaps — "innovation and tenacity" carried the load, and TTPs written at Nellis substituted for the establishing directive doctrine required. If the fast loop reliably compensates, what exactly is the argument for spending the slow loop's cost at all? Construct the strongest case that an organization should deliberately let TTPs lead and doctrine trail — then identify the failure mode Session 02's memory argument predicts for it (personnel rotation, cross-unit transfer, allied interoperability).
4. Quarantined exploration. The ACF gives exploration its own artifact class, pipeline, and "DRAFT — NOT FOR IMPLEMENTATION" stamp, while doctrine "warrants" no re-proof of its own validity. Does this structure protect exploration, as March requires — or contain it, ensuring the exploitative memory is never contaminated by an unapproved idea? Note the asymmetry of burden: concepts must "accept the burden of proof"; doctrine never re-argues its case. Who, in this system, can falsify standing doctrine, and with what evidence?
5. Galbraith audit. Map the staffing chain onto Galbraith's four strategies: where has the Army bought information-processing capacity (CRMs, adjudication boards, seminars), and where has it reduced the need for processing (page limits, format templates, capability-statement grammar)? Brown proposes digitizing the establishing directive and coordination measures. Does digitization change the design, or only the medium — and what does Daft and Lengel's richness hierarchy predict about resolving doctrinal disputes through a comment matrix?
6. Why keep Level 4 plans? Eisenhower's epigraph says plans are worthless and planning is everything; JP 5-0 nonetheless mandates OPLANs with transportation-feasible deployment databases that are costly to build and age from the moment of approval. Using Hansen's codification economics and the RATE loop, defend the Level 4 investment for some contingencies and Level 1 for others. What is the organizational equivalent in your world — and do you actually maintain your "OPLANs"?
7. Reading negotiated text. The JP 3-30 supported-commander language was settled by presenting two options to the service operations deputies; theater kill-box conventions were never settled at all. Given only a finished doctrinal text, how could an analyst detect which passages are validated learning and which are negotiated residue? Propose concrete textual markers (hedged verbs, theater exceptions, dual definitions like interdiction vs. air interdiction) — then ask what this implies about reading any organization's official documents.
8. Did the reform touch the seam? The 2025 joint manual answers Walker's 2003 critique by adding a fast lane (100-day, administrative, urgent changes; 7-day qualification) and a formal Joint Doctrine Note to capture emerging practice — yet keeps the nominal 300-day base revision and the one-vote-per-service consensus that produces the "melded" doctrine he attacked. Did the institution fix the seam or only the latency around it? Argue whether an expedited change track and a non-authoritative bridging note actually address "melding, not synthesis," or whether they leave the structural cause untouched while relieving its most visible symptom. What would a reform that touched the seam itself have to change?
9. Two factories, two vetoes. The Army guide (Section 03) gives any organization a one-comment veto that blocks the whole document; the joint manual (Section 06) runs a four-type typology in which a critical comment is downgraded to "major" without a general-officer memo and, even memo-backed, yields to joint-working-group consensus. Same institution, same design problem, two settlements. Which better balances letting any stakeholder stop the line against lowest-common-denominator memory — and what does the difference between an Army audience and a five-service joint audience predict about why each chose as it did? Would you expect the content of joint doctrine to be blander or sharper than Service doctrine as a result?
10. Transfer. Your organization's doctrine factory is its RFC process, architecture decision records, postmortem repository, style guides. Choose three TRADOC mechanisms — the version-rank ladder, the comment severity typology with mandatory replacement language, the rank-matched adjudication board, the 25-page cap, the PD's fixed participant roster — and argue which would improve your process and which depend on properties (severe environment, obedience norms, mass rotation) your organization lacks.