CCM · REAL MA 01
Rail Integration in the AI Era
Public recommendations for comparing merger and cooperation in the UP–NS case
Compare what independent cooperation can deliver with comparable AI capabilities before attributing additional benefits to common ownership. Then examine who receives the gains, who bears the costs, and whether participants retain practical choices.
A faster train does not necessarily produce a better delivery. If the warehouse cannot receive it or the connecting truck is unavailable, an improvement in one part of a supply chain becomes waiting in another. The opportunity for AI is to help participants coordinate those dependencies. The institutional question is which arrangements make that coordination effective and accountable.
The proposed Union Pacific–Norfolk Southern combination offers a concrete setting for examining that question. This paper uses Computable Competition–Cooperation Mechanisms (CCM) to structure a public comparison. It makes no recommendation to approve or reject the transaction. The aim is to identify improvements worth testing, evidence needed to distinguish alternatives, and conditions under which a more capable logistics network can remain accessible and correctable.
Start with a moving counterfactual
As checked on 16 September 2026, STB materials describe an ongoing review, not an approval. The applicants estimate $3.5 billion in annual shipper transportation savings associated with shifting about 2.1 million truckloads to rail; inventory and equipment savings are additional. These are applicant forecasts, not realized outcomes. [1][2]
A joint September 2025 announcement described another UP–NS interline product and cited earlier services. It supports including continuing cooperation in the no-merger scenario; it does not establish the new service’s actual performance. [3]
The relevant comparison therefore cannot assign advanced AI to a merged network while leaving independent operators with yesterday’s technology. A merger may create capabilities that contracts and interfaces cannot readily reproduce. If so, identify those capabilities and test the claimed difference against an improving alternative.
Separate three questions
First, what can technology improve? Arrival prediction, appointment coordination, exception detection and authorized scheduling are candidates for corridor trials. Their value depends on data quality, operating constraints, adoption and the allocation of responsibility. An attractive use case is not a measured result.
Second, what does ownership add? Compare independent interline cooperation at a baseline technical level (A0), stronger AI-assisted independent cooperation (A1), hypothetical common ownership at the baseline level (B0), and hypothetical common ownership with comparable AI capabilities (B1). B1 versus A1 is the main comparison for incremental integration value. The four cells organize inquiry; this study has no four matched sets of actual orders and does not rank the scenarios empirically.
Third, who experiences the result? A shipper’s lower bill does not itself compensate a driver for unpaid waiting. Higher operating profit does not establish community acceptance. Shippers, workers, connecting carriers, operators, production businesses and communities need distinct outcomes and evidence. Costs that can be compensated require an accountable payer; non-negotiable boundaries require legitimate standards and authority.
Keep the economic channels distinct
Transportation savings from modal shift depend on the volume actually diverted and the corresponding door-to-door cost difference. They should be examined by corridor and commodity, including terminal and connecting services. Rising or falling truck rates can change both diversion and the price advantage. Inventory, equipment and reliability benefits require separate estimates with checks against double counting.
A carrying-cost calculation cannot validate or refute the entire modal-shift forecast. Nor does dividing an aggregate savings estimate by an aggregate load estimate establish a lane-specific quote. The practical question is how much traffic the improving A1 network could already attract, and how much additional value B1 would produce under comparable demand, price and technology assumptions.
What CCM contributes and what has been tested
CCM organizes the work as a revisable record: define the decision, register affected actors, specify alternatives, distinguish observations from forecasts and assumptions, calculate well-defined quantities, check declared conditions, and revise the arrangement when evidence exposes a conflict. AI can assist research and option generation; people remain responsible for selecting and justifying the conditions.
Model V1.3 now implements the revised split: shipper cost and transit time remain directional reports, while five proposed boundaries—labor transition, community safety, non-discriminatory access, human override and rollback drill—alone determine the declared-boundary status. A voluntary choice to pay more for faster service therefore does not automatically veto a proposal.
The public default still contains no matched A1/B1 order evidence and all five boundary fields remain UNKNOWN. Each proposed boundary still needs an applicable standard, legitimate authority and verification method. The implementation demonstrates reproducible logic, not transaction benefits, empirical discriminatory power, criterion legitimacy or real-world authorization.
The historical V1.2 package is retained for reproducing its original seven-hard-condition behavior. It has not been rewritten or presented as if it had already implemented the revised design. The current V1.3 source, inputs and tests are separate and traceable.
Five recommendations for a corridor trial
1. Publish a comparable service proposition. Specify the corridor, freight type, service window, full charges, responsibility for exceptions and evidence sources. Include an improving independent-cooperation alternative. Publish enough to test the claims without requiring unrestricted disclosure of commercial secrets.
2. Begin with improvements that can be reversed. Test prediction, appointments and exception handling first, then consider bounded autonomous drayage and warehouse automation where operating conditions permit. Assign responsibility and preserve human intervention. Expansion should follow demonstrated service and cost performance.
3. Measure access and exit. Record time from a complete third-party access application to a decision, refusal rates and reasons, and unresolved applications. Track changes in comparable access terms. Measure the time and cost of moving customer data and completing the first shipment with an alternative service. Group observations by corridor, application type and period; these indicators require context and do not alone establish exclusion.
4. Fund transition costs explicitly. Include training, job changes, driver waiting and affected communities in the plan and budget. A projected automation benefit should not depend on silently moving adjustment costs outside the analysis.
5. Retain alternatives when coordination fails. Test isolation of an affected segment, human override and fallback service. An interface or backup that exists on paper is insufficient if no authorized participant can use it when needed.
A longer horizon for logistics infrastructure
Beyond this transaction’s disclosed implementation scope, coordinated rail, road and warehouse networks could make consolidation and distribution of smaller consignments more practical. Autonomous vehicles and low-altitude aircraft may serve suitable segments, subject to operating limits. Connections to factories, highly automated plants and agricultural cold chains are further research directions—not announced commitments of the merger parties.
Reusable coordination may lower the cost of later acquisitions, alliances and service access. It can also deepen dependence on a dominant operator. The same access and switching measures should therefore follow each expansion. This study has not simulated that long-term cycle. Its recommendation is to build interoperability, responsibility and correction into development while testing whether they work in practice.
More capable logistics could become part of the infrastructure through which society produces and exchanges essential goods. The desirable outcome is a growing capacity to cooperate that more participants can use, challenge and improve. The route to that outcome need not require common ownership of every connected business.
Participation and evidence
Bring a coordination problem, a counterexample or an authorized observation. Identify who benefits, who bears an unrecorded cost, and what remains unknown. CCM is a small open-source foundation intended for reuse and extension; participation does not require accepting a broader theory of civilization.
For current reproduction, use V1.3: its runner reads the source input and writes derived files to an output directory without overwriting the input. The historical V1.2 package retains its original overwrite behavior. Neither route is an arbitrary order-import workflow. Preserve original evidence, units, source scope and authority.
Readers considering formal participation should consult the current STB docket and its filing, service and data-access requirements. Publishing a paper or opening a GitHub issue does not enter a submission into that docket or authorize access to protected material. This paper has not been filed with STB or sent to the companies.
Author: 子君赋 (Zijun Fu). Produced by 文明跃迁研究组 (Civilization Leap Research Group). English public recommendations V1.0, based on Chinese main manuscript V1.2.1. Evidence cut-off: 16 September 2026. AI assisted source checking, analysis and drafting. No endorsement by the named companies or regulator is implied.