XSci

A Cross-Domain TARA-to-NAC Rule Mapping Methodology: From Automotive OEM Networks to Energy Storage Systems

Krishna Teja Medam, Thiago Quinelato

Submitted by Krishna Teja Medam

Published November 13, 2026 · Version v1, November 13, 2026 · DOI 10.66977/xsci.2611.0022

Cryptography, Cybersecurity, Manufacturing

Abstract

Threat analysis and risk assessment under ISO/SAE 21434 ends with cybersecurity goals, but existing methods give little structured guidance for turning those goals into enforceable network access control rules. Compliance standards such as IEC 62443, ISO/SAE 21434, and ISO/IEC 27001 are usually applied as parallel efforts rather than as one enforcement architecture. This paper presents a six-step methodology that maps cybersecurity goals from a threat analysis and risk assessment to network access control rules, each placed at a named enforcement point and traceable to the threat records behind it. It also presents a unified compliance framework that treats IEC 62443 and ISO/IEC 27001 as a fixed substrate and domain-specific regulation as an interchangeable overlay. The methodology is applied to an automotive original equipment manufacturer plant network, addressing multi-tier supplier authorization and legacy operational technology devices, and then to utility-scale battery energy storage, where North American Electric Reliability Corporation Critical Infrastructure Protection standards and ISO/IEC 27019 replace the automotive-specific standard while the substrate stays constant. In the automotive example, 37 threat entries and 24 cybersecurity goals yield 23 rules covering 11 of the 13 in-scope goals; the energy example yields six rules. Three derived rules have no equivalent in the cited standards. Both examples are desk-based: the methodology is applied to them, not tested on them. The result is a reusable pattern for translating risk assessment into enforcement across critical infrastructure sectors, with empirical validation left as future work.

View PDF

Review reports

xPeerd (AI review)

November 13, 2026

AI Review (xPeerd — DAReview)

Summary

This manuscript presents a conceptually clear and potentially useful methodology for translating TARA-derived cybersecurity goals into concrete NAC rules across two domains: automotive OEM networks and utility-scale battery energy storage systems. Its central claim is that existing TARA and standards-mapping work often stops at goals, requirements, or control catalogs, whereas this paper derives device-relevant access-control rules placed at named enforcement points and tied back to threat records. The paper’s strongest features are its explicit problem framing, traceability orientation, and structured cross-domain worked examples. The authors also state important boundaries candidly, including that “Both worked examples are desk-based” and that “The method is applied to them rather than tested on them” on page 1, Introduction .

The main weaknesses concern validation, generalizability, and evidential sufficiency. The manuscript repeatedly acknowledges that derivation is manual, examples are not experimentally validated in this paper, attack feasibility is not rated, and many goals remain deferred or only partially discharged. On page 9, Applied TARA Outputs, the authors state that “Attack feasibility is not rated” and that every threat is treated as requiring treatment on impact alone . On page 10, they also acknowledge that CG-17 is only partly discharged because internal OT sessions remain out of scope and are deferred to future work under CG-22 . Overall, the manuscript has a coherent contribution, but it currently reads more as a structured methods paper with worked illustrations than as a fully validated research study.

Potential Major Revisions

  1. Reproducibility and empirical support remain insufficient for the strength of the claims

The manuscript’s principal weakness is that its contribution is demonstrated only through desk-based examples, not through systematic empirical validation. In the Introduction on page 1, the authors explicitly write: “Both worked examples are desk-based. The method is applied to them rather than tested on them” . Later, on page 15, the Discussion/Future Work material reiterates that “the companion paper tests the automotive enforcement points, and the energy rules still need a site or testbed of their own” and that “the derivation here is done by hand” . This sharply limits the manuscript’s evidential basis.

If the manuscript claims a “systematic methodology” that is reusable across sectors, then the current evidence is too narrow to establish robustness, transferability, or operational reliability. A stronger version should include at least one of the following within the present paper: a reproducible validation protocol, rule-generation inter-rater consistency, formal correctness criteria, deployment testing, or quantitative evaluation against baseline approaches.

  1. The treatment of attack feasibility introduces a substantial methodological distortion

On page 9, Applied TARA Outputs, the manuscript states: “Attack feasibility is not rated, because it needs site-specific operational data a general method cannot assume; every threat is treated as requiring treatment on impact alone” . This is methodologically consequential. TARA frameworks ordinarily combine impact and feasibility to prioritize controls. Omitting feasibility may be acceptable for illustration, but the manuscript does not sufficiently analyze how this simplification may bias rule selection, overproduce rules, or mis-rank mitigation priorities.

This choice risks weakening the paper’s practical credibility. If all threats are treated as requiring treatment, the derivation pipeline may become insensitive to realistic operational risk thresholds. The paper should explicitly evaluate how rule outputs would differ under at least one alternative feasibility-sensitive scenario.

  1. The manuscript overstates closure of the traceability gap while still leaving core security goals unresolved

The manuscript’s framing is persuasive, but some key goals remain unmapped or only partially addressed. Table A4 states that “Rules derived: 11 goals. In scope, unmapped: CG-14 and CG-24... Deferred to future work: 11 goals, including CG-22” on page 23 appendix material . This is important because some of the deferred items are not peripheral. For example, OT lateral movement restrictions, flooding protection, integrity protection for HMI communications, and tamper-evident logging are all materially important security controls.

The most significant limitation appears in the worked derivation for CG-17/CG-18. On page 10, the manuscript states: “Sessions inside the OT zone are out of scope here... The proxy does not see these sessions, so CG-17 is only partly discharged in this example. Restricting them falls under CG-22” . This means an important command-authentication goal is not fully handled in the architecture presented. The paper should moderate claims about end-to-end traceability and more explicitly distinguish solved cases from partial cases.

  1. Cross-domain transfer is promising but not yet strongly substantiated

The paper’s cross-domain ambition is interesting, but the evidence for transferability is still limited. On page 1, the authors claim that the methodology demonstrates “cross-domain applicability” through a BESS example . However, page 15 clarifies that the BESS case produced only six rules and remains desk-based , while page 23 notes that this paper’s cross-domain transfer still has the limits “Derivation is manual; examples are desk-based; deployment is tested in the companion paper” .

The transfer may be real, but the manuscript currently shows compatibility more than validation. The paper should either narrow its claim to “illustrative cross-domain transfer” or provide stronger evidence that the same procedure remains stable under different role vocabularies, protocols, regulatory overlays, and asset classes.

  1. Rules without standard equivalents require more rigorous justification

The manuscript highlights flagged rules with “no direct equivalent in the cited standards,” including policy-version binding and quarantine of undefined roles. On page 9, the authors carefully note that these flags “are not a claim of novelty against practice” and only indicate that the rule is not directly recoverable from the standards text . That is a fair distinction, but the paper still needs a more rigorous normative defense for why these flagged rules are necessary, sufficient, and non-arbitrary.

Without that additional argument, readers may reasonably wonder whether these are principled derivations or design preferences chosen by the authors. The paper would benefit from a formal rule-justification schema showing why the trigger/action pair is required by the threat-goal chain even when standards do not state it explicitly.

  1. Partial dependence on a companion paper weakens the manuscript’s self-sufficiency

The Introduction says that “A companion paper evaluates an implementation of the automotive enforcement points, and none of its results are used here” on page 1 . While this is transparent, it also leaves the present submission in a somewhat incomplete state. The paper asks the reader to accept the practical plausibility of the architecture without seeing implementation outcomes. If the authors choose not to include those results, they should compensate with stronger internal validation: formal examples, consistency checks, sensitivity analysis, or a procedural audit trail showing that independent analysts can reproduce the same rules from the same TARA inputs.

  1. Anti-fraud / coherence assessment

I do not see strong paper-mill or fabrication signals. The manuscript is coherent, domain-specific, and internally structured. It acknowledges limitations instead of masking them, which argues against obvious fraudulent construction. The references, standards framing, and derivation examples appear thematically aligned. However, one caution remains: because the core examples are desk-based and manually derived, the paper’s claims should not be interpreted as experimentally established. The main risk here is not apparent fabrication, but over-interpretation of illustrative results as validated ones.

Potential Minor Revisions

  1. Claims wording should be tightened to avoid overstating demonstrated performance

In the abstract on page 1, the manuscript states that the method “is applied to a primary worked example” and “demonstrates its cross-domain applicability through a secondary worked example” . Given the acknowledged lack of experimental validation, “demonstrates” is slightly too strong. “Illustrates” or “shows preliminary cross-domain transferability” would better fit the evidence.

  1. Some contribution language should be better aligned with stated limitations

On page 2, the manuscript says: “The method in this paper supplies that missing traceable path” . Since multiple goals remain unmapped or only partially discharged, this should be softened to something like “supplies a traceable path for the in-scope worked cases.”

  1. The manuscript should more explicitly define evaluation criteria for a “good” derived rule

The text is clear about traceability, but less clear on quality criteria beyond traceability. For example, on page 3 the paper distinguishes the identity layer from the protocol layer and explains that neither cited NIST publication “derives the rule from a TARA” . However, the paper should define how derived rules are judged for minimality, necessity, conflict avoidance, maintainability, and false-positive operational burden.

  1. Internal consistency of scope handling should be made more explicit

The paper argues for mapping goals to enforceable rules, yet it also leaves several high-impact goals deferred. Table A4 on page 23 clearly records deferred goals including CG-19, CG-21, CG-22, and CG-23 . It would help readers if the manuscript added a brief rationale for the exclusion criteria that determined which goals were treated now and which were deferred.

  1. Typographic and stylistic matters

I did not detect many obvious typographical or grammatical errors in the retrieved text. The prose is generally polished and consistent. Minor style issues remain:

  • Repeatedly using declarative contrastive formulas such as “What is protected,” “Who is the subject,” and “How” is effective, but occasionally becomes mechanical, especially in the derivation sections on pages 9–10 .
  • Some sentences are long and densely packed with standards references, especially in Related Work and Background, which may reduce readability for non-specialist reviewers .
  • The phrase “The contribution is real, but it is bounded” in the Related Work discussion on page 1 has a slightly rhetorical tone for a neutral research article . A more formal comparative wording would be preferable.
  1. Formatting and presentation

The appendix tables appear central to evaluating the method. Because the paper’s contribution depends heavily on traceability, the main text may be too dependent on appendix material. For example, page 9 states that the full TARA is in Tables A1 to A4 . Consider moving a compact but representative threat-to-goal-to-rule trace matrix into the main paper.

  1. AI-assisted content analysis

The manuscript includes a disclosure: “Claude (Anthropic) was used for reviewing manuscript text and figures, for checking standards identifiers and citations against the published sources, and for identifying inconsistencies in drafts” on page 18, Declaration of AI-Assisted Technologies . This is appropriate and transparent.

Estimated AI-generated content: low to moderate, approximately 10–20%.

Basis for estimate:

  • The manuscript has a consistently polished style, regular structural patterns, and highly standardized expository phrasing.
  • However, it also contains domain-specific derivation logic, specific standards anchoring, and localized methodological limitations that suggest substantial human authorship and control.
  • The disclosure itself limits the need for speculative concern, because AI use is openly acknowledged for editorial and consistency-checking tasks rather than concealed.

Highlighted AI-susceptible sections:

  • The highly regularized comparative prose in Related Work on pages 1–3 may have been AI-assisted at the editorial level, especially recurring formulations such as what prior work “does not produce” or where a process “stops” .
  • The structured rhetorical cadence of the derivation subsections on pages 9–10 may also reflect AI-supported stylistic harmonization rather than substantive generation .

Assessed epistemic impact:

  • Low, based on the disclosed use and the manuscript’s internal coherence.
  • The AI use, as disclosed, does not appear to undermine the stated contribution, but it reinforces the need for careful verification of all standards mappings, especially where the paper claims no direct standard equivalent.

Recommendations

  1. Reframe the paper explicitly as a validated methodology proposal with illustrative worked examples, unless new empirical evidence is added.
  2. Add a reproducibility section specifying the exact inputs, transformation steps, and decision criteria required for an independent analyst to derive the same rule set.
  3. Include a sensitivity analysis showing how outputs change when attack feasibility is incorporated, even hypothetically.
  4. Distinguish more sharply between:
    • fully discharged goals,
    • partially discharged goals,
    • in-scope but unmapped goals,
    • goals deferred to future work.
  5. Justify flagged non-standard-equivalent rules with a formal derivation logic or necessity argument.
  6. Bring one compact threat-to-goal-to-rule trace table from the appendix into the main text.
  7. Temper claims of cross-domain generalizability unless additional domains, testbeds, or evaluative criteria are added.
  8. If possible, integrate at least a minimal subset of the companion implementation results so the manuscript stands more convincingly on its own.

Overall recommendation

Promising manuscript with a clear methodological contribution, but it requires major revision before it can support its broader claims with sufficient rigor.

Reviewer3 (AI review)

November 13, 2026

AI Review (reviewer3.com)

Overview

TARA-to-NAC Rule Mapping Methodology

Contribution

The authors present a systematic six-step methodology that maps Threat Analysis and Risk Assessment cybersecurity goals directly to concrete, traceable Network Access Control rules at distinct identity and protocol enforcement layers.

Unified Substrate-Overlay Compliance Framework

Contribution

The study introduces a compliance framework that treats IEC 62443 and ISO/IEC 27001 as a fixed management and system substrate while applying domain-specific regulations as modular, interchangeable overlays.

Cross-Domain Critical Infrastructure Application

Contribution

The authors demonstrate the methodology's sectoral transferability through desk-based worked examples in an automotive OEM manufacturing plant network and a utility-scale battery energy storage system.

Comments (11)

1. Inconsistent ISO/SAE 21434 overlay entry for rule NAC-R-02b in Table 2 — Editorial

In Table 2 ('Automotive overlay', page 8), rule 02b is listed under the ISO/SAE 21434 column as having No 21434 NAC duty. However, in 'Derivation 3' and 'Remaining In-Scope Goals' (page 11) as well as Table 4 (page 12), NAC-R-02b traces to cybersecurity goals CG-02, CG-07, and CG-11. In particular, Table A4 (page 22) defines CG-11 as: 'IT/OT segmentation rules SHALL be version-controlled and mapped to IEC 62443 and ISO/SAE 21434 clause requirements.' Moreover, the other audit/supporting rule 08e (which traces to CG-07 and CG-08) is mapped to Cl. 9 goals [1] in Table 2, whereas 02a–02d (which all trace to CG-02, and 02b to CG-02, 07, 11) are listed as having No 21434 NAC duty.

2. Rule taxonomy count: two categories in prose versus three in Table 1 — Editorial

The rules fall into two categories, each defined by the decision it makes rather than by the product that enforces it (Table 1).

In the 'Rule Taxonomy' subsection (page 6), the opening sentence states: 'The rules fall into two categories, each defined by the decision it makes rather than by the product that enforces it (Table 1).' Similarly, the 'Applied TARA Outputs' subsection (page 10) states: 'Three derivations are walked through below, chosen to cover both rule categories and both flagged rules.' However, the fourth paragraph of 'Rule Taxonomy' refers to 'the other two categories' when introducing the supporting rules, Table 1 lists three rows under the 'Category' column ('Identity and segmentation', 'Protocol and command', and 'Supporting'), and the third walkthrough ('Derivation 3: Governing the Rule Set (CG-02)') specifically derives the 'Supporting' category (which contains the second flagged rule, NAC-R-02e). The prose should distinguish between the two session-decision layers and the three rule categories shown in Table 1.

3. Enforcement point mismatch for assignment-logging rule NAC-R-08e — Editorial

Policy integrity, change control, and policy - version binding (02a – 02e) and assignment logging (08e) apply at the site’s policy host as written, so they are not renumbered.

In Figure 2 (page 11) and Table 4 (page 12), the assignment-logging rule NAC-R-08e is placed at the RADIUS server and boundary (RADIUS; boundary in Table 4), whereas the supporting rules NAC-R-02a, 02b, 02c, and 02e are placed at the Policy host. However, in 'Worked Example 2' under 'Placement' (page 14), the text states that 08e applies at the policy host: 'Policy integrity, change control, and policy-version binding (02a–02e) and assignment logging (08e) apply at the site’s policy host as written, so they are not renumbered.'

4. Asymmetric application of the strict trigger-and-action flag test — Major

A substrate binding names the requirement a rule discharges. An overlay entry names a duty the rule also serves. Neither has to state the rule’s trigger and action. The flag test is stricter: a cited requirement must cover the trigger and the action together, and support for one half is noted without removing the flag. A rule with no such requirement keeps the

The flag test states that a rule is flagged when no requirement in the cited standards mandates its trigger and action together, yet the preceding paragraph states that a substrate binding does not have to state the rule's trigger and action. Because rules like NAC-R-FC-03/04 (permitting Modbus FCs 3 and 4) or NAC-R-08a (assigning Tier-1 to VLAN 10) also have triggers and actions that are much more specific than IEC 62443-3-3 SR 2.1 or RFC 3580, the abstraction level at which a standard is judged to cover 'both trigger and action together' is not defined precisely enough to distinguish generalizable standard requirements from the flagged rules without analyst subjectivity.

5. Algorithm 1 generates redundant audit rules per trigger class — Major

Step 4, write the rules. Each distinct trigger at an enforcement point gets one rule. Every trigger class gets a default branch for the no - match case, so an input the policy does not recognize is denied or quarantined rather than passed. Every decision rule gets an audit rule. Each rule is bound to the

In Algorithm 1, line 11 ('add audit rule at p') is nested inside both the goal loop (line 2) and the trigger-class loop (line 8), which would generate an audit rule for every trigger class of every goal rather than one per decision rule or enforcement point. In the worked examples, however, only a single assignment-logging rule (NAC-R-08e) is created for the identity layer and no separate audit rules are created for 01a, 05a, 05b, or 12a, indicating a mismatch between the pseudocode loop structure and how Step 4 is actually executed.

6. Missing explicit rule count breakdown for grouped rows — Minor

Table 4 lists the result . The sixteen rules derived for CG - 02, CG - 06, CG - 07, and CG - 08 consolidate to twelve. The protocol proxy adds seven: one per listed function code and a default deny for any other. The remaining goals add four more (01a, 05a, 05b, 12a). That gives 23 distinct rules covering eleven goals, with two in - scope goals unmapped for stated

The text states that Derivation 1 and Derivation 3 consolidate to 12 rules, Derivation 2 adds 7 protocol proxy rules, and the remaining goals add 4 rules for a total of 23 distinct rules in Table 4. Because Table 4 contains only 19 rows by combining two read function codes into row NAC-R-FC-03/04 and four write function codes into row NAC-R-FC-05/06/15/16, explicitly noting this row grouping in the table footnote or text would prevent confusion when readers count the rows.

7. Compound BESS rules bypass Step 4 atomic trigger-per-rule structure — Major

NAC-R-E-02 admits vendor sessions only through the intermediate system with multi-factor authentication, then places each by role: the PCS vendor in the PCS zone and the BMS vendor in the BMS zone, with no route between them or to the EMS control functions.

Step 4 of the methodology requires writing one rule per distinct trigger at an enforcement point, plus a separate default branch and audit rule, which in the automotive example produced distinct rules for Tier-1, Tier-2, Tier-3, cross-tier blocking, and per-function-code filtering. In the BESS example, however, NAC-R-E-02 bundles MFA admission, PCS-vendor zone assignment, BMS-vendor zone assignment, and cross-zone blocking across two enforcement points (RADIUS and zone boundary) into a single rule record, and NAC-R-E-04 similarly combines DNP3 control, read, and default-deny branches into one rule, departing from the atomic coding granularity used in Worked Example 1.

8. Inconsistent classification of ISO/IEC 27001 as substrate versus overlay — Major

threat entry, and every trigger class has a default branch. The bindings shown are the substrate bindings to IEC 62443 and NIST guidance; the Automotive Overlay subsection applies the ISO/SAE 21434 and ISO/IEC 27001 overlay to the same rules.

Throughout the paper, ISO/IEC 27001 is defined as half of the fixed compliance substrate alongside IEC 62443, yet the Automotive Overlay text and Table 2 present ISO/IEC 27001 as part of the overlay applied to the automotive rules, and Table 3 omits the ISO/IEC 27001 column entirely in favor of ISO/IEC 27019 scope elements. Presenting the ISO/IEC 27001 bindings consistently as part of the invariant substrate across both worked examples would better support the claim that the substrate remains fixed while only the domain overlay changes.

9. Algorithm 1 omits rule-layer assignment and conflict logging — Major

from several threat records. A rule record is r = ( id, G, c, act, p, l, b ): an identifier, the set of goals the rule serves, the trigger condition, the action, the enforcement point, the rule layer (see Rule Taxonomy), and the binding to a standard requirement, or a flag where none exists (see Rules Without a Standard Equivalent). Triggers are things an enforcement point can

While a rule record is defined as r = (id, G, c, act, p, l, b), Algorithm 1 never assigns the rule layer l during rule construction, and line 14 ('on conflict keep stricter action') omits the Step 5 requirement that the conflict be recorded against both goals and that the goal sets G be unioned during consolidation. Updating the pseudocode to explicitly populate all fields of r and record conflicts will align the formal algorithm with the prose procedure.

10. Unexplained goal-trace merges during consolidation in Table 4 — Minor

Step 5, consolidate.

Step 5 specifies that rules from different goals are merged only when they share the exact same trigger, action, and enforcement point, and the text states that every merge was an exact duplicate. However, Table 4 lists CG-12 in the trace of NAC-R-08d (a RADIUS unknown-role quarantine rule) and CG-20 in the trace of NAC-R-08f and NAC-R-08b (Tier-2 and Tier-3 VLAN assignment rules), even though CG-12 and CG-20 concern OT boundary and PLC register read restrictions rather than generating identical RADIUS role-assignment triggers.

11. Inconsistency between two-category prose and three-category Table 1 — Major

The rules fall into two categories, each defined by the decision it makes rather than by the product that enforces it (Table 1).

In the Rule Taxonomy subsection, the text states that rules fall into two categories and later describes a supporting class, whereas Table 1 lists three distinct categories under the Category column. Clarifying upfront that the taxonomy comprises two session-decision layers plus a third supporting category would make the formalization consistent with Table 1 and the rule record definition.

XSci (AI review)

November 13, 2026

Summary

ISO/SAE 21434 ends a TARA at cybersecurity goals and IEC 62443 allocates requirements to zones and conduits, but neither obliges anyone to carry a goal through to a configured rule on a named device. The paper proposes a six-step manual procedure that turns goals and threat records into a rule table in which each rule has a trigger, an action, an enforcement point and a backward trace to the threats behind it, with IEC 62443 and ISO/IEC 27001 as a fixed compliance substrate and domain regulation as an overlay. It is applied to an automotive OEM plant network, where 37 threat entries and 24 goals yield 23 rules covering 11 of the 13 in-scope goals, and to a battery storage site under a NERC CIP and ISO/IEC 27019 overlay, yielding six rules. Three rules are reported to have no equivalent in the cited standards; both examples are desk-based and nothing is deployed.

Contribution and Significance

The core contribution is a documented, auditable derivation path from TARA goals to per-identity rules at named enforcement points, with bidirectional traceability as a checkable property rather than an aspiration. Engineers who must justify a firewall or RADIUS policy line to an auditor have no written route from the risk record to the configuration; the rule record and its three closing checks give them one. The second useful idea is the rule-keyed rather than control-keyed compliance mapping, which improves on the control catalogs of PURITY [11] and Djebbar and Sogbade [12].

Relative to the closest prior work, Navarra et al. [6], the delta is narrower than the framing suggests. What holds is organizational-boundary principals (supplier tiers), rules at two enforcement layers rather than one gateway, and a standards overlay; against these, [6] automates the derivation that is manual here. The advance is one of engineering and documentation rather than concept, and its value rests on two properties the paper asserts but does not establish: that the procedure is reproducible, and that it transfers. One omission also matters for positioning. IEC 62443-3-2:2020, whose purpose is to carry a risk assessment through zone and conduit partitioning to target security levels and a cybersecurity requirements specification, including compensating countermeasures for assets that cannot meet a requirement natively, is not cited. The gap claim probably survives, since that part stops short of a rule, but the argument has to be made.

Strengths

  • The full TARA workbook is published rather than summarized: functions (A1), assets (A2), 37 threat entries with STRIDE category, impact ratings and originating goal (A3), and all 24 goals with their disposition (A4). Every count in the body can be checked against the appendix, and the consolidation arithmetic does check out.
  • Derivation 2 and Figure 2 are the clearest part of the paper: FR 1 fails at the legacy PLC, placement moves to the conduit, and the rules that follow (FC 3 and 4 permit and log; FC 5, 6, 15 and 16 deny with a Modbus exception response; any other code denied; connection closed on an unparsable frame) are concrete enough to implement.
  • Negative results are reported with unusual discipline: CG-14 and CG-24 are left unmapped with reasons, CG-17 is declared "only partly discharged" for sessions originating inside the OT zone, and the flag search is stated not to be exhaustive.
  • Table A5 positions the work against eleven groups of prior work with an explicit "limit relative to this paper" column, making the novelty claim testable by a reader.

Recommendation

Major Revision for the SAE International Journal of Transportation Cybersecurity and Privacy.

The contribution is of the kind this venue publishes and the artifact quality is high, but each of the three claimed contributions rests on assertion rather than evidence, and in one case the application contradicts the procedure it defines. All three concerns are repairable within the present approach — the deferral gate made explicit, the energy example reframed, the flag test made well-defined — but none by copy-editing.

Major Concerns

  1. The application contradicts Step 1, and the coverage figure uses an undefined denominator (Methodology, Step 1; Applied TARA Outputs; Table A4). Step 1 states that "Every goal with a risk-reduction treatment enters the procedure." Because attack feasibility is not rated, Applied TARA Outputs treats every threat "as requiring treatment on impact alone", so all 24 goals carry such a treatment and all 24 should enter. In fact 11 are recorded as "Deferred to future work" in Table A4 before the procedure runs, and the criterion for that deferral is stated nowhere. The deferred set is also inconsistent with the method: CG-03 and CG-23 (logging) are deferred although Step 4 requires that "Every decision rule gets an audit rule" and 08e and 02b already log; CG-09 is deferred while the near-identical CG-11 is discharged by 02b and 02e at no extra cost.

    This bears on the claim that "a second analyst given the same TARA can reproduce the rule set": the partition into 13 in-scope and 11 deferred goals determines everything downstream and is the one step a second analyst cannot reproduce. It also makes the headline figure misleading, since 11 of 13 in-scope goals is 11 of 24. State a defensible gate inside the procedure, or run the deferred goals.

  2. The energy example is constructed from the automotive result, so it cannot evidence cross-domain transfer (Worked Example 2; Discussion). The manuscript states that NAC-R-E-06 is "the energy instance of 08d", that NAC-R-E-05 uses "the placement move used for the legacy PLC (Step 3)", and that NAC-R-E-02 is "the automotive tier pattern (08a, 08f, 08b)". The six scenarios were chosen after the automotive rules existed, so the finding that "no new rule category was needed" is largely built in. The example also exercises only part of the procedure: no TARA, no impact ratings, one goal and one rule per scenario, so "consolidation in Step 5 never had anything to merge" and the Step 6 checks are trivial. The Discussion concedes much of this, yet the Abstract still says cross-domain applicability is "demonstrated". Either run the procedure on a BESS risk assessment not derived from the automotive case, or recast the example as an illustration.

  3. The "no standard equivalent" result is an artifact of how the test is specified (Rules Without a Standard Equivalent; Automotive and Energy Overlay). The test requires that "a cited requirement must cover the trigger and the action together". NAC-R-08d is flagged because SR 2.1 "does not quarantine a successfully authenticated principal whose role is undefined, and it does not require an alert" — but the authors chose that action. SR 2.1 RE(2) requires a defined permission-to-role mapping and SR 5.2 RE(1) requires deny-by-default with allow-by-exception, so a principal whose role lies outside the mapping carries no authorization; had the action been written "deny", no flag would fire. The test therefore measures the specificity of the authors' own action vocabulary rather than a protection gap, and the same objection applies to 02e against CM 1.4 and SR 3.4. The search is also admittedly non-exhaustive and omits IEC 62443-3-2 and 62443-4-2. Define the test so its outcome does not turn on the wording of the action, extend the search to those parts, and otherwise state only the weaker claim the body makes.

Minor Comments

  1. Table 4 shows 19 rows while the text claims 23 rules; state the convention (FC-03/04 counts as two, FC-05/06/15/16 as four) in Step 4 and in the caption.
  2. Table 3 binds NAC-R-E-03 to SR 2.1 while Table 6 binds it to SR 2.1 and SR 5.2.
  3. Cite International Electrotechnical Commission, "Security for Industrial Automation and Control Systems - Part 3-2: Security Risk Assessment for System Design", IEC 62443-3-2:2020, June 2020, and position the method against its ZCR workflow.
  4. Cite Rose, S., Borchert, O., Mitchell, S., and Connelly, S., "Zero Trust Architecture", NIST SP 800-207, Aug. 2020; zero trust is the first keyword and CG-14 a ZTNA goal, yet no zero-trust reference appears.
  5. CG-20 should be labelled partly discharged, as CG-17 is, since the protocol rules "place no limit on which registers a permitted read can reach".
  6. Say that the example applies a reduced form of the ISO/SAE 21434 TARA, since neither attack feasibility nor a risk value is determined.
  7. The companion paper is invoked five times and carries CG-14 and all validation, but is never cited.
  8. Discuss the cost of the protocol proxy — latency on the Modbus command path, throughput, and behaviour on proxy failure.
  9. The sentence "Security levels express the target a zone is designed to meet; they do not by themselves select a rule" appears twice in one subsection, and Table 1 has the typo "unparased frames".
  10. Several passages appear transposed (the Introduction's opening sentence, the first paragraph of Rules Without a Standard Equivalent, the Table 2 lead-in); the heading "TARA Methods and ISO/SAE 21434" is empty and line 1 of Algorithm 1 is garbled.
  11. State how the rule set would be evaluated if deployed; the Step 6 checks verify bookkeeping, not that the 23 rules mitigate the 37 threat entries.