From railway mergers to intelligent logistics networks
Integration Opportunities and Public Choices in the AI Era
Author: Zijun Fu | Produced by the Civilization Leap Research Group
CCM Real-World Application Case: Public Recommendations, Main Text V1.2.1 | September 16, 2026
Public research proposal | Docket FD 36873 | Fact-checking cutoff: September 16, 2026
View V1.3 · Recalculate in the browser · Submit a counterexample
A railway merger appears to connect two companies' lines. The deeper question is whether it can connect factories, warehouses, trains, trucks, and recipients that once waited for one another into a more effective cooperative network.
If so, the opportunity extends beyond railways. But if connection gradually strips every participant of choice, efficiency may become another form of dependency.
The proposed merger of Union Pacific (UP) and Norfolk Southern (NS) provides a real entry point into this question. This article neither recommends approval nor rejection of the transaction. It makes a constructive proposal: move merger discussion from asset scale to cooperative capability, compare forms of integration under the same technological conditions, and bring added value, distribution, and long-term consequences into the decision together.
Using this real case, the article studies how AI can expand cooperation and uses CCM to compare integration arrangements. Opportunities belong to arrangements that can create, share, and continually improve cooperative capability. Evidence must determine which arrangement is better. A transaction's long-term value also depends on whether more subjects can use the resulting capability and whether it becomes a reliable foundation for social production and circulation.
1. What Is the Merger Actually Meant to Solve?
As verified for this article on September 16, 2026, the STB had resumed review and set the procedural schedule on August 18; this did not establish transaction approval.[1] On April 30, the applicants projected $3.5 billion in annual shipper transport savings from shifting about 2.1 million truckloads to rail. Inventory and equipment savings from faster, more reliable service were additional.[2] These benefits must be accounted for separately. They are applicant projections, not realized gains, and have not been shown to require a merger in their entirety.
Operationally, continuous lines may reduce cross-company coordination, unify some investments and operations, and improve cross-regional service. This makes the merger worth serious study. Expanding market coverage, attracting freight, and improving capital returns are understandable business aims. But logically possible motives must not be presented as facts about undisclosed corporate intentions.
Industry competition ultimately concerns customers' choices: Can the shipment arrive as promised? Who handles exceptions? Does the total cost make sense? Even smooth rail-trunk operations may leave customers waiting if warehousing, feeder transport, or handoffs are mismatched. Integration value therefore cannot be assessed from a railway map alone.
Another fact matters: in their joint September 15, 2025 announcement, the companies proposed a bidirectional domestic intermodal product linking Louisville through Kansas City to western and southern US markets, expected to start in mid-October. They described it as another cross-company product development and listed existing Los Angeles–Charlotte and Los Angeles–Jacksonville services.[3] After announcing the merger, they continued cooperation between independent companies. This supports retaining a no-merger, continually improving intermodal comparison, rather than treating it as a stagnant baseline. The announcement alone does not prove timely launch or promised performance.
2. AI Changes What Cooperation Can Achieve
Traditional coordination problems often involve more than a slow vehicle. Parties see different information, accept different constraints, and bear different responsibilities. A factory changes production without the warehouse knowing in time; a train arrives early without feeder vehicles ready. Each party optimizes itself while the whole chain may pay the price.
AI can help address these problems: predict arrivals and demand, detect anomalies, compare dispatch plans, and coordinate appointments and operations with authorization. This article proposes testing such uses; it does not treat unmeasured corridor effects as established facts.
AI may make cooperation feasible where information and coordination costs previously prevented it. But it may also help powerful actors price more finely, allocate capacity, and exclude competitors. Technology strengthens the ability to act; the purposes that ability serves still depend on rules and choices.
A fair comparison is therefore essential: how does an independently operated network using open interfaces, joint scheduling, and AI cooperation compare with a merged network using similar technological capability? The merger scenario cannot receive future technology while the no-merger scenario remains in yesterday. Claims of additional merger–AI synergies should explain why independent cooperation cannot readily reproduce them.
3. From Container Trunks to End-to-End Service
The following is exploratory research beyond the implementation scope disclosed for this case. Rail handles suitable long-distance trunks; feeder vehicles connect industrial parks and warehouses; distribution systems consolidate and unpack shipments; last-mile networks deliver. Autonomous trucks and low-altitude aircraft may enter specific segments under appropriate operating conditions. Customers buy one transport service with accountable responsibility, rather than assembling unfamiliar services themselves.
This also opens research space for smaller shipments. Many small orders may be consolidated regionally, moved along a trunk, and distributed after arrival. But rail's ability to carry containers does not make every parcel suitable for rail. Freight density, load factors, sorting and return costs, and delivery windows determine feasibility. Low-altitude transport also faces payload, weather, noise, safety, and authorized-operation limits; it must not be presented as a general replacement for heavy trunk transport.
Extending toward production, logistics information can help factories plan materials and production, and farms align harvesting, cold chains, and markets. More automated factories still face logistics interruptions, raw-material fluctuations, and changing customers. What deserves development is mutually intelligible, coordinated interfaces, not the assumption that every factory and farm must join one ownership system.
Greater capability may enable broader autonomous coordination. We need not predict when AGI will mature before leaving today's interfaces open, nor wait until systems become highly autonomous to ask who can change objectives and stop mistaken instructions.
4. Intelligent Networks Can Advantage Further Integration
Once a network has reusable scheduling, settlement, tracking, and exception-handling capabilities, the coordination cost of adding a warehouse, carrier, or line may fall. This can strengthen its ability to expand through acquisitions, alliances, or service access. This is a mechanism-based inference, not a finding that this transaction has already created such capability.
A positive cycle is possible: access expands available capabilities, better matching reduces idle activity, more reliable service attracts customers, and further investment follows. A negative cycle is also possible: more customers make alternatives harder to sustain; deeper interfaces make exit costlier; the rule-setting party increasingly determines whether others receive orders.
To distinguish these cycles, continuously record processing time from complete third-party access applications to decisions, rejection rates and reasons among decided applications, and the separate share remaining unresolved for long periods. Compare changes in fees and terms for similar applicants. Record the time and cost from a customer's request to migrate data and switch services until usable data and the alternative service's first actual shipment. Group by corridor, application type, and period to distinguish capacity or safety limits from exclusionary barriers. These are proposed observations; this draft has no measured values for the case, and no single metric independently proves causation.
The same intelligent network may become shared industry infrastructure or a closed gateway charging for passage. Scale cannot make this judgment for us. Each subsequent integration should recheck which capabilities it adds and which choices it removes.
These observations should constrain long-term planning in return. Actual use must determine whether interoperable interfaces shorten entry and backup capabilities support switching. Whether through merger, alliance, or service access, CCM should reexamine incremental value, burden-bearers, and exit conditions at each expansion, allowing effective practices to transfer across networks. This is a proposed future testing path; subsequent merger cycles in this case have not been simulated.
5. CCM Brings the Parties into the Same Account
Computable Cooperation Mechanisms (CCM) explicitly records subjects, benefits, costs, constraints, and unknowns, then performs reproducible checks on the parts with definitions and data. It does not replace commercial judgment or public decisions, or compress everyone's rights and interests into an aggregate score.
For shareholders and operators, examine added revenue, capital investment, integration risk, and timing. For shippers, total price, reliability, and alternatives. For employees and drivers, job changes, skills transitions, work intensity, and who bears waiting. For connecting carriers, fair access and settlement. For communities, noise, emissions, traffic barriers, and accident risks. These are interests to examine, not predictions of actual gains and losses.
Transport-shift account
Dividing $3.5 billion by 2.1 million loads implies average transport savings of about $1,667 per load. This is aggregate division, not a corridor quote; first confirm that both totals share scope and period. Testing the applicants' projection requires corridor-level transferable volumes, end-to-end truck–rail price differences, first- and last-mile costs, and weighting methods.[2]
If both volume and the price gap weaken, aggregate savings decline accordingly. FreightWaves reported on April 30 that the shift projection rose from 2.0 million to 2.1 million loads because truck rates increased.[5] This is a discovered secondary attribution, not verified against application workpapers. Rate cycles, rail service quality, and feeder capability all affect outcomes. Test these changes instead of automatically attributing them to the merger.
Inventory-time account
Inventory and equipment savings are separate. Assuming $100,000 in goods per container and a 20% annual carrying rate, saving 25 hours reduces in-transit inventory costs by about $57. This shows one benefit channel under those assumptions; it neither validates nor refutes $3.5 billion. Avoided production stoppages, stockouts, reliability, and equipment turnover may add value, requiring separate evidence without double counting. Full calculations and scale comparisons follow below.
The public reference for 25 hours comes from the September 2025 joint announcement: UP said network operating improvements could save up to 25 hours between Southern California and its Kansas City intermodal terminal.[3] In April 2026, the applicants argued that eliminating cross-company interchange through merger could remove the 24–48 hours existing interchanges may impose.[2] Both are corporate statements; the latter is not a measured corridor average. The mechanisms can coexist. Whether they affect the same orders or can be added requires checking corridors, segments, timing, and overlap, not assuming addition or subtraction. Even if total improvements are additive, improvements already in the independent-operation baseline cannot be counted again as merger gains.
Comparison at equivalent technology
Under the same AI capability, market prices, and demand, how much freight could independent cooperation A1 already attract, and how much would merger B1 add? Compare transferable volume and end-to-end price differences, then account separately for inventory and reliability benefits. The existing corridor model has no network-wide freight-shift model: it cannot verify $3.5 billion or assume all 2.1 million loads are merger-exclusive gains. CCM turns a large benefit promise into individually answerable questions, identifying ways to improve cooperation.
More importantly, shippers' savings cannot automatically offset drivers' lost rest, and corporate profits cannot stand for community consent to new risks. Compensable losses require a payer, timing, and enforceable arrangements. Safety and rights boundaries must be set by authorized subjects under applicable rules, not silently drawn by a model.
CCM's most useful finding here is not a merger score, but three separate judgments: how much value technology creates, how much ownership integration adds, and who ultimately receives the value and bears the costs.
6. How We Use CCM to Analyze This Merger
The process is: clarify the problem, identify subjects, construct options, register evidence, calculate and check, and revise recommendations. The CCM reference program handles determinate calculations and checks of declared conditions. Researchers frame questions, construct options, and interpret results. The strategic judgments here are not automatically generated transaction conclusions.
Expand the question beyond whether to merge: As AI develops, which competitive and cooperative arrangements can improve end-to-end logistics while reasonably distributing benefits, costs, responsibility, and correctability? This admits mergers, open intermodal services, shared interfaces, joint scheduling, and phased integration into comparison, avoiding choosing an answer first and then seeking supporting data.
Next, register subjects and outcomes. Companies and investors correspond to revenue, capital, and integration risk; shippers to total costs, reliability, and alternatives; employees and drivers to jobs, income, and waiting costs; connecting carriers to access, settlement, and bargaining; factories and farms to production coordination, losses, and dependency; communities to safety, environment, and traffic effects. Registration is neither consultation nor completed quantification. The current cost program calculates only shipper costs, transport time, and in-transit inventory costs. Employees, communities, connecting carriers, and operators have unvalidated fields, not gain-and-loss models. Corporate capital returns and factory and farm outcomes are not numerically accounted for; factories and farms are not independent subjects in current checker inputs.
Construct four comparison scenarios: A0 is independent intermodal service with baseline technology; A1 is independent open cooperation with enhanced AI; B0 is a hypothetical merger with baseline technology; B1 is a hypothetical merger with comparable AI capability. The main comparison is B1 against A1; A1 versus A0 and B1 versus B0 investigate technology's effects. These four cells organize research questions. Four sets of real order data do not yet exist, so no empirical ranking of all four scenarios can be claimed.
Retain each input's type and source. Public announcements establish that an event or rule was disclosed. Corporate projections remain applicant claims. Cargo value, carrying rates, time savings, and other sensitivity parameters are research assumptions. Quotes, service times, and third-party effects without applicable evidence remain null. The model must not read null as zero or silently fill corridor performance with network averages.
The program has generated 36 hypothetical cost outcomes, while all empirical conditions remain unknown. It has made no satisfied or violated determination for any real option. It demonstrates a reproducible workflow, not discriminative validity for this case or justification of its condition selection. Operating details and old-code boundaries follow below. Real corridor pilots should next test and revise the method.
The condition hierarchy below is a revised design in this main text, not yet incorporated into the older model package. The older program's aggregate status does not establish implementation of the new design.
This revision divides conditions into two levels. Shipper cost and time increments become reports with direction, units, and uncertainty; requiring both not to increase is merely one optional comparison preference. When a shipper voluntarily pays for faster delivery, report the extra cost, service gain, and basis for consent, rather than rejecting the entire arrangement merely because cost rises. Labor transition, community safety, nondiscriminatory access, human takeover, and rollback drills are five proposed boundaries. Relevant requirements may become hard constraints only after applicable standards, authority, and verification methods are established.
The legitimacy of conditions must precede calculation. For each proposed boundary, register its proposer, affected subjects, source standard, authorized decision-maker, verifier, and validity period. Values 1, 0, and null mean evidence-supported pass, evidence-supported failure, and unknown under declared standards. A plan's existence does not establish adequacy; a completed drill does not guarantee safety. When standards are not established, mark them as awaiting establishment rather than hiding the missing normative basis behind an unknown value.
How did findings change the recommendations? Existing intermodal announcements preserved independent cooperation as an option. Separating technology from ownership prevented assigning all AI gains to merger. Cost accounting brought feeder transport, waiting, and surcharges back into analysis. Subject registration brought transition costs and community effects into recommendations. Questions about dependency and correction made open interfaces, human takeover, and alternative services part of planning. These findings combine documentary evidence, mechanism reasoning, and program checks; not all belong to the calculator.
Finally, retain feedback and revision paths. Applicable future orders can replace nulls, recalculate costs and service times, and support third-party checks under explicit standards. If shippers benefit while drivers wait longer unpaid, revise appointments, charging, or compensation before comparing again; do not hide the conflict in a higher total score. If parties agree to change a condition, retain the original, the reason, and recalculated results.
Unfinished work includes full simulations of long-term autonomous-transport expansion, subsequent merger cycles, shifts in market power, and society-wide net welfare. These remain mechanism analysis and long-term planning here. The current tools make computable parts reproducible, evidence gaps visible, and conditions open to questioning; they do not replace all judgments with a status.
Stakeholder analysis, cost–benefit analysis, and scenario planning already have established methods. CCM must demonstrate its value here by organizing these activities into a traceable, checkable, revisable process. A name alone does not establish uniqueness.
7. Public Recommendations for Companies and Other Participants
First, publish a testable end-to-end service proposal. Specify corridors, freight types, order definitions, and promised time windows, with baselines, charges, responsibility, and exception handling. Report modal-shift transport savings separately from inventory, equipment, and reliability gains, including volume and relative-rate sensitivity. Compare continually improving independent cooperation and explain which increments within the 2.1-million-load projection can be attributed to merger, giving supporters and critics equally clear objects to check.
Second, build capability through a reversible corridor pilot. Begin with stable freight flows, validating arrival prediction, appointment coordination, and cost warnings while retaining an independent-cooperation comparison. Add autonomous feeder transport and warehouse automation under appropriate conditions. Record efficiency and participants' benefits and costs at each stage, with responsible people, human takeover, and exit arrangements. Treat small-shipment consolidation and factory and farm connections as later extensions, proceeding only after freight density, service reliability, and costs support them.
Third, write open access into business arrangements. Establish explainable service terms, data-use boundaries, and dispute processes. Permit eligible third parties to join and support customers migrating their own data and changing services. Openness does not require exposing trade secrets; it avoids unnecessary information closure that manufactures dependency.
Fourth, assign responsibility for transition costs. Training, job changes, drivers' waiting and rescheduling, and community effects belong in plans and budgets. Automation cannot appear only in benefit forecasts while its adaptation costs are left to others.
Fifth, preserve alternatives for concentrated failures. Dispatch errors, network outages, and service disputes should permit local isolation, human takeover, and switching to other services. The larger the system, the less a single mistake should close everyone's way out.
These are research recommendations, not additional statutory approval conditions. As of September 16, 2026, this research had not obtained enough data for comparable order analysis and network-wide shift recalculation. The STB's August 18 decision required resubmission of unfiltered workpapers by August 28, with access for the Board and authorized persons under the protective order.[4] This draft has not verified compliance with resubmission; actual files, versions, and access conditions must be checked.
Subjects seeking continued participation in the docket should note the September 30, 2026 notice-of-intent deadline and November 18 deadline for comments, protests, and requests for conditions.[1][4] To recalculate the shift projection, first request public workpapers and an index from the applicants. For confidential material, check protective-order eligibility, undertakings, and use restrictions; qualified persons can then apply for access.[6] Participation notices and confidential-access authorization are separate. The former does not automatically grant access, and September 30 is not a universal closing date for all information requests or public comments.
This is an ongoing review process. Preliminary Justice and Transportation Department comments, if submitted, are due December 3, 2026; responses and rebuttals supporting the application are due February 16, 2027; final briefs are due May 28, with public hearing dates to follow.[4] Later submissions must comply with stage-specific rules, not be understood as unrestricted additions at any time. Formal filings should identify FD 36873 and check service and current requirements on the STB case page. Public research can keep posing testable questions and improving methods; formal docket participation is for willing, resourced subjects to decide.
8. From One Case to a Different Choice
The approach extends beyond railways. Ports, energy, healthcare, supply chains, and public services adopting AI may face similar questions: Does local efficiency benefit the whole? With more concentrated capability, do participants still have choices? When promises fail, who can correct them?
CCM does not require researchers to build a large platform first. It provides a small runnable open-source foundation, inviting users to bring real problems, add evidence, challenge assumptions, and revise methods. The real test is whether it helps people find previously unseen solutions whose burdens they can jointly bear.
AI brings opportunities to reorganize cooperation as well as new risks to guard against. Realizing those opportunities depends on our willingness to include others' situations in the design.
When such practices enter transport contracts, business interfaces, and public rules, more than a logistics route changes. More efficient cooperation can become more widely available opportunity, and stronger capabilities can remain correctable.
Civilizational change begins when, despite growing capability, we still choose to preserve options for one another.
Participation and onward reading
Bring a familiar coordination problem: Who benefits, who bears costs, and what remains unknown? Start by recalculating through the open-source CCM project or proposing a counterexample. Do not submit unauthorized customer information, contracts, or personal data. For the relationship between these choices and how civilization operates, continue with the five-volume Civilization Leap work and the Living Civilization trilogy. Using CCM requires no prior acceptance of the entire theory.
Before recalculating, save a separate working copy. Existing run.py regenerates default null inputs and overwrites derived files in outputs. Do not edit only outputs and rerun. Real orders require changing input-generation logic in a traceable copy and preserving original evidence, units, and versions; the current entry point does not support arbitrary order imports. This main text's condition hierarchy is also not yet in the old program.
CCM open-source project and usage entry point
Calculation supplement
The inventory calculation is 100,000 × 20% × 25 ÷ 8,760 = $57.0776. At 24–48 hours, the same assumptions yield $54.7945–$109.5890, approximately 3.29%–6.58% of the implied $1,666.67 average. This compares scales with different accounting scopes. It does not establish actual savings of only $57 per shipment or refute $3.5 billion.
For transport-benefit sensitivity illustration only: if both volume and the per-load price gap are 20% below the implied baseline, $3.5 billion × 0.8 × 0.8 = $2.24 billion. This is neither a forecast nor an empirical output from the current corridor program.
Recalculation entry points and file relationships
Model V1.2 calculates shipper and inventory costs in dollars per container and hours; cost items and time segments must be deduplicated. Three cargo values, three carrying rates, and four additional time savings produce 36 purely parametric outcomes without real orders. The aggregate transport sensitivity below is separate arithmetic, not an implemented freight-volume forecast.
Historical V1.2 treated all seven conditions as hard constraints: any evidence-supported violation yielded overall VIOLATED; no violation with remaining unknowns yielded UNKNOWN; only all satisfied yielded SATISFIED. All seven empirical values were unknown, producing no empirical satisfied or violated determination.
Version boundaries must also be explicit. The two reports plus five proposed boundaries are revised design in main text V1.2.1; they do not rewrite corridor cost model V1.2's historical code or outputs. The old package still reproduces the old seven-hard-condition run, whose aggregate status cannot establish implementation of the new design. Subsequent implementation should update condition configuration and input structures and add targeted validation, such as voluntary payment for faster delivery not triggering overall rejection. Existing synthetic tests check program branches but cannot establish the appropriateness of the conditions for this case. With every empirical value unknown, real-world applicability also remains untested.
The historical run is reproduced with the CCM REAL MA 01 Corridor Cost Model V1.2 package. Its case_id is REAL_MA_01_CORRIDOR_V1_2, using coop.case.v0.1 and the CCM v0.1.0 reference checker. Main text V1.2.1 and model V1.2 are versions of different files. Revising the article does not imply a synchronized model upgrade, new real data, or new test results.
model.py defines cost accounting and case conditions; run.py generates inputs and derived results; mechanism_ref/core.py checks declared conditions. outputs/empirical_inputs.json retains empirical nulls; outputs/conditional_break_even.csv stores 36 hypothetical examples; outputs/ccm_input.json and outputs/ccm_result.json contain declared inputs and checker results.
In the extracted directory, run python -m unittest -v test_model to check implementation, then python run.py to reproduce historical outputs. The existing twelve tests cover unknown propagation, rejection of non-finite or negative values, cost accounting, condition violations, and cancellation of equal AI time savings, among other behaviors. Passing establishes only tested implementation behavior, not transaction effects, empirical discrimination, or appropriate condition selection. Input fingerprints trace checked files; they do not certify truthful data.
Sources and research boundaries
Fact-checking cutoff: September 16, 2026. Sources [1]–[4] and [6] are original regulator or applicant materials. [5] is secondary reporting whose attribution has not been checked against workpapers. Autonomous logistics, production-side integration, and subsequent merger cycles are exploratory analysis, not announced corporate implementation commitments. This article is neither an approval or rejection recommendation nor certification of net effects, and has not been submitted to regulators or companies. AI assisted material checking, scenario reasoning, and text organization.
[1] STB — UP NS Merger Resources: procedural status and August 31 deadline extension
[2] UP — April 30, 2026 revised-application announcement: transport savings and additional benefits
[3] UP and NS — September 15, 2025 joint intermodal announcement: official NS release
[4] STB — August 18, 2026 decision: workpaper access on page 5 and schedule on pages 13–14
[5] FreightWaves — April 30, 2026 report: secondary attribution for the higher shift projection
Research support: CCM REAL MA 01 Corridor Cost Model V1.2; Public Evidence Verification and Application Recommendations V1.3. The model retains historical calculations and old condition configuration. This is a targeted final correction of main text V1.2.1, not a code upgrade or acquisition of new empirical data.