{"id":"1464f422-f511-4fd1-b0f6-0630bec5c954","arxiv_id":"2505.10590","paper_version":2,"verdict":"REJECT","confidence":"HIGH","novelty_score":3.0,"correctness_risk":"high","formal_verification":"none","parameter_count":0,"one_line_summary":"A conceptual ratio, CRR, is proposed to separate AI capability from realized business value, and it is applied narratively to firms like OpenAI, NVIDIA, and Adobe.","lead":"This paper proposes a Capability Realization Rate, the fraction of AI potential that has turned into actual business results, to explain why AI stocks get expensive. It applies this lens to the 2023 to 2025 AI boom and argues that low-realization firms are at risk of valuation crashes.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Capability Realization Rate is never measured: the central claim that CRR identifies valuation misalignment is untestable without an operational definition of the ratio in Section 2.","rationale":"We agree with the reader's REJECT verdict. The paper is a clearly written conceptual essay, and the policy recommendations are reasonable, but the central claim is an empirical conditional: if CRR is measurable, then tracking it provides an early-warning signal. The argument never establishes the antecedent. The reader's weakest assumption—that 'Total AI capability potential' is not well-defined—is precisely the load-bearing gap. Our concrete test would force the authors to make the model operational and would reveal whether the proposed construct can be measured and whether it has predictive content. If the test cannot be executed, the case-study classifications remain arbitrary labels applied post hoc. The absence of machine-checked proofs or reproducible data is not by itself a flaw for a conceptual paper, but it reinforces that the central claim is currently unsupported. We therefore leave the verdict unchanged.","tokens_in":10263,"tokens_out":3123,"duration_ms":32027,"concrete_test":"Select one firm from the case studies (e.g., Adobe) and, using only information publicly available on 2024-12-01, define a specific CRR formula from the Section 6 indicators (e.g., CRR = (AI-attributable revenue + AI-attributable cost savings) / total AI R&D spend plus an estimated capability index), compute it for that firm and for NVIDIA and OpenAI on the same date, and pre-register a prediction for each firm's subsequent 3-month abnormal return. If the formula requires subjective inputs, or any of the three CRR values cannot be computed, or the predictions are no better than assigning the sector average return, the model is not operational.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim—that CRR can help identify valuation misalignment risk—requires that CRR be a measurable quantity. Section 2 defines CRR as 'Realized AI-driven performance / Total AI capability potential' with no units, no measurement procedure, and no data source; neither component is defined at the level needed to compute a ratio. No section of the paper actually computes CRR for any firm. The case-study labels (e.g., 'NVIDIA … had a very high CRR', Section 4.3; Adobe's CRR described as 'not yet convincing', Section 4.2) are assigned after observing stock-price reactions, using narrative summaries rather than a pre-specified formula, so they cannot provide independent evidence for the model. Section 6 lists candidate indicators (AI-driven revenue share, user engagement gains, cost savings), but these are never aggregated into CRR, given weights, or validated. Without an operationalization, the model is unfalsifiable: any observed price movement can be rationalized as consistent with some unspecified CRR. The paper therefore supports, at most, a qualitative taxonomy of AI adoption, not the claimed early-warning capability.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper proposes the Capability Realization Rate (CRR) model, defined in Section 2 as the ratio of realized AI-driven performance to total AI capability potential, and argues that CRR can help identify valuation misalignment risk, where stock prices diverge from realized AI-driven value. The hypothesized mechanism is anchoring: during the 2023-2025 generative AI boom, investors anchored on total AI capability and overpriced low-CRR firms. The evidence consists of sector-level market data (Section 3, including self-constructed AI-native vs. Traditional indices and 2023 MSCI sector returns) and six narrative case studies (Section 4: OpenAI, Adobe, NVIDIA, Meta, Microsoft, Goldman Sachs) in which firms are classified as high or low CRR. The paper closes with policy recommendations (Section 5) and conclusions (Section 6) that present CRR as a gauging and risk-management tool.","tokens_in":10488,"tokens_out":10147,"duration_ms":92781,"significance":"If a measurable version of CRR existed, the framework could give investors and regulators a structured way to separate AI-driven earnings from AI-driven hype, and the anchoring mechanism is a plausible behavioral channel for the 2023-2025 valuation dispersion. The paper is clearly written, the case studies are informative, and the authors are transparent in labeling the proposal a hypothesis and the working paper exploratory. However, the paper ships no operational definition, dataset, code, or computed CRR values; no falsifiable prediction is derived or tested; and the empirical sections are descriptive. The conditional contribution is therefore a qualitative taxonomy of AI adoption rather than the early-warning metric claimed in the abstract. Whether the concept can deliver the claimed benefit remains an open research question that this manuscript does not resolve.","major_comments":[{"comment":"The central construct, CRR, is defined only as the ratio of realized AI-driven performance to total AI capability potential, with no units, measurement procedure, or data source assigned to either component. The paper never computes CRR for any firm or sector, and the case-study classifications (e.g., NVIDIA 'had a very high CRR' in Section 4.3; OpenAI's CRR 'arguably very low' in Section 4.1) are narrative judgments not produced by a stated rule. The candidate indicators listed in Section 6 (AI-driven revenue share, engagement gains, cost savings) are never aggregated into a CRR value, weighted, or validated, and the fair-value diagonal in Figure 1 is asserted rather than derived. Because any observed price path can be rationalized by an appropriately chosen but unspecified CRR, the abstract's claim that CRR 'can help identify valuation misalignment risk' is unfalsifiable as stated.","section":"Section 2 (CRR definition)"},{"comment":"The paper's core prediction, that low-CRR firms are prone to mispricing, is close to tautological under the given definitions. Misalignment is defined as prices diverging from realized AI-driven value (Section 2, Figure 1), while CRR is defined as the realized fraction of capability potential. The case studies assign CRR tags after observing the valuation outcomes - for example, Adobe's December 2024 price drop is presented in Section 4.2 as evidence that its CRR was 'not yet convincing' - so the documented pattern of low-CRR firms being corrected is built into the labeling procedure rather than discovered. A non-circular test would compute CRR from pre-specified data and then predict subsequent returns; this paper does not attempt such a test.","section":"Sections 2 and 4"},{"comment":"The empirical analysis does not test the CRR model and is not reproducible as reported. The AI-native and Traditional composite indices in Figure 2 have no disclosed constituent lists or construction details beyond 'market-cap weighted,' and the 2023 sector returns in Figure 3 are quoted from media citations ([4], [5]) without underlying data or uncertainty estimates. The paper itself notes that much of the observed dispersion is 'factors largely unrelated to AI' (Section 3), so the sector returns cannot establish that investors anchor on AI capability in particular. Section 3 therefore documents a well-known market rally rather than providing evidence for the CRR mechanism.","section":"Section 3 (Figures 2 and 3)"},{"comment":"The conclusions overstate what the evidence supports. Section 6 asserts that valuation misalignment risk 'can be gauged and, to some extent, managed,' and Section 5 proposes specific regulatory interventions (standardized AI revenue disclosure, sanctions on misleading AI claims), but no gauge is produced anywhere in the manuscript, and the recommendations rest entirely on the unvalidated CRR construct. The paper's own disclaimer (page 2) calls the work exploratory, which is consistent with the evidence; the definitive policy conclusions in Sections 5 and 6 are not.","section":"Sections 5 and 6"}],"minor_comments":[{"comment":"Sections 1 and 4 state that the paper presents seven case studies, but Section 4 contains only six subsections (4.1-4.6: OpenAI, Adobe, NVIDIA, Meta, Microsoft, Goldman Sachs); the count should be corrected to six.","section":"Sections 1 and 4"},{"comment":"Figure 2 requires a methodology description: constituent lists for both composite indices, data vendor, rebalancing rule, and treatment of companies that entered or left the sample during 2023-2025.","section":"Figure 2"},{"comment":"Several references are incomplete or unverifiable as formatted (e.g., [1], [2], [4], [5], [16] lack author, venue, or date), which matters because all quantitative claims trace back to secondary media sources.","section":"References"},{"comment":"The claim in Section 3 that the three leading sectors 'accounted for essentially all of the above-average performance' should be quantified from the underlying sector data rather than attributed to a headline citation.","section":"Section 3"},{"comment":"Section 4.4 asserts that Meta's CRR improved, but no metric is given; a stated measure (e.g., AI-attributable revenue or documented engagement lift) would make the classification auditable.","section":"Section 4.4"},{"comment":"The statement in Section 4.3 that NVIDIA became 'the first public company in history to surpass $4 trillion in market valuation' should include a date and source in the text itself, given the reference list format.","section":"Section 4.3"}],"recommendation":"reject","confidential_remarks":"This manuscript reads as a position paper or industry commentary rather than a research article: the central variable is never measured, the empirical section is descriptive, and the case studies are assembled to fit the narrative. The reference base is dominated by media and blog sources, and the engagement with the behavioral finance literature on anchoring (only the cited Cen et al. line) is too shallow to derive testable implications. I would suggest the authors develop an operational CRR with real firm-level data and an out-of-sample prediction before resubmitting to a peer-reviewed venue; in its current form the paper is better suited to a workshop or practitioner outlet."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You asked for a candid read on the CRR paper. Here it is: the paper is a clearly written, timely piece of market commentary, but the central construct is never measured, so the main claim that CRR can identify misalignment risk is untestable. The reader's reject verdict is right.\n\nWhat the paper does well: it is well organized and readable, it correctly points to a real phenomenon (investors anchoring on AI potential rather than realized results), and it offers sensible policy suggestions—especially the call for standardized AI revenue disclosure. The case studies are engaging narratives, and the taxonomy of high- vs. low-realization firms is a useful way to frame the 2023–25 AI rally. The authors cite the anchoring literature, so the behavioral premise is credible.\n\nThe soft spot is not minor; it is load-bearing. CRR is defined in Section 2 as realized AI-driven performance divided by total AI capability potential, but neither component is given units, a measurement procedure, or a data source. The paper never computes CRR for any company, never offers a dataset, and never validates the ratio. The case-study labels (e.g., NVIDIA high CRR, Adobe low CRR) are assigned after observing stock price movements, which is selection on the outcome. A low-CRR firm is, by definition, one whose realized value is below its potential—and misalignment is, by definition, price diverging from realized value. So the central prediction is close to tautological, and any price move can be rationalized post hoc. The sector-level data in Section 3 is presented as supporting evidence, but the numbers are drawn from media reports without systematic sourcing or uncertainty bars. The authors' own disclaimer calls the paper exploratory, which is honest.\n\nSo what is actually new? The label CRR and the qualitative map of capability versus realization. That is a reasonable framing device, but it is a restatement of ordinary performance analysis, not a new scientific result. The behavioral anchoring point is already established in the cited literature. The paper's value is as a policy discussion and a cautionary narrative, not as a model.\n\nWho gets value from this? Readers looking for a digestible overview of AI valuation dynamics and regulators thinking about disclosure rules. It could also serve as a teaching example of an unmeasured construct dressed up as a metric. But for a research audience, the lack of operationalization is disqualifying in its current form.\n\nRecommendation: I would not send this to peer review as is. If the authors return with an operationalized CRR—clear components, weights, and a measurement strategy applied to a real dataset—then it would deserve referee time. As submitted, it is a well-intentioned essay, not a research paper.","headline":"A readable, timely essay that never operationalizes its central ratio; the CRR model is a label, not a measurable construct.","tokens_in":10966,"tokens_out":1900,"would_cite":false,"duration_ms":22489,"reading_group":"no","serious_thinker":"yes","would_accept_peer_review":false},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper argues that AI stock prices during the 2023–2025 boom anchored on total AI capability, and introduces a Capability Realization Rate to quantify the gap between promise and realized earnings.","keywords":["capability realization rate","anchoring bias","AI equity valuations","valuation misalignment risk","generative AI boom","market bubbles","AI disclosure policy"],"falsifier":"Construct a panel of listed AI-exposed firms over 2023–2025 using a standardized proxy for CRR, such as AI-attributable revenue divided by total revenue or R&D. If low-CRR firms do not systematically experience larger valuation drawdowns or worse forward returns than high-CRR firms within the same sector and size bucket, the central claim that CRR identifies misalignment risk would be empirically undermined.","tokens_in":10061,"feed_emoji":"📈","tokens_out":11209,"duration_ms":94952,"temperature":0.7,"pith_summary":"The paper argues that during the 2023–2025 generative-AI boom, equity investors anchored stock prices to firms' total AI capability rather than to the fraction of that capability actually generating revenue or profit. To make this precise, it introduces the Capability Realization Rate (CRR), defined as realized AI-driven performance divided by total AI capability potential, and hypothesizes that low-CRR firms carry the largest valuation misalignment risk. On this view, NVIDIA's soaring value was supported by realized chip sales, while OpenAI's $80 billion mark rested on promise rather than earnings. The authors use sector returns and seven company case studies to illustrate the pattern, and they argue that a standardized CRR would give investors and regulators an early-warning signal for AI-related bubbles. The paper is a conceptual framework backed by illustrative cases, not an operationalized metric.","feed_headline":"Capability Realization Rate can flag AI stock mispricing","feed_subtitle":"The model divides realized AI performance by total AI capability potential to separate hype from earnings.","key_machinery":"The central object is the Capability Realization Rate (CRR), a ratio defined as $CRR = \\frac{\\text{realized AI-driven performance}}{\\text{total AI capability potential}}$. It is a conceptual gauge of how much of a firm's AI promise has been converted into market-visible results. The argument treats CRR as the missing mediator between AI capability and valuation: investors anchor on the potential rather than the realized fraction, pushing low-CRR firms into the upper-left danger zone of a capability–premium map. The paper uses this map to organize its case studies, placing OpenAI and Adobe as low-CRR over-anchored names and NVIDIA as a high-CRR justification for a high premium. The ratio itself is not assigned units or a measurement procedure; it functions as an interpretive lens rather than a computed statistic.","core_discovery":"The central claim is that the Capability Realization Rate (CRR) mediates between AI hype and fundamental valuation: markets price companies as if capability equals realized value, so firms whose CRR is low (high potential, little current monetization) trade at premiums that later deflate when realized earnings fail to catch up. The paper defines CRR as realized AI-driven performance over total AI capability potential, and uses the 2023–2025 generative-AI boom as its laboratory, documenting that AI-native firms and sectors with high AI exposure attracted outsized returns, that low-CRR firms such as OpenAI and early Adobe exhibited the widest gaps between valuation and fundamentals, and that high-CRR firms such as NVIDIA supported their premiums with actual earnings. The authors argue that tracking CRR over time can identify misalignment risk before a correction, and they recommend standardized AI revenue and efficiency disclosures so that the behavioral anchor is replaced by measurable realization.","pith_inferences":["Inference: The CRR logic generalizes beyond the 2023–2025 episode; a similar capability-anchoring lens could be applied to earlier technology waves, such as the internet or clean-tech booms, where a small set of anchor firms drove sector-wide valuations.","Inference: The model implicitly predicts a testable cross-section: if a standardized CRR proxy, such as AI-attributable revenue as a share of total revenue, were constructed, low-CRR firms should exhibit lower subsequent realized returns and higher return volatility after a correction, while high-CRR firms should not.","Inference: Because CRR is a ratio, two firms with identical CRR but vastly different absolute scale, for instance $10 million realized on $100 million potential versus $1 billion on $10 billion, occupy the same point in the model, yet mispricing risk likely depends on scale and on the variance of the realization process; the paper does not address this.","Inference: The policy recommendation of mandatory AI-metric disclosure could be tested as a natural experiment if a single regulator adopted it; treated firms' valuation volatility could be compared with that of a control group."],"forward_implications":["If investors tracked CRR, valuation misalignment risk for AI-related equities would be identifiable earlier, since low-CRR firms with high capability anchors are the most prone to sharp corrections.","Companies that publicly report AI-attributable revenue, cost savings, or engagement gains (raising measured CRR) should trade with more stable premiums than those whose realization is opaque.","Sectors with high AI exposure will continue to see outsized returns and subsequent drawdowns whenever narrative anchors, such as flagship model releases, outrun realized monetization.","Regulators that mandate standardized AI disclosures would give investors the inputs needed to compute CRR, converting the framework from a retrospective lens into a forward-looking risk indicator.","Even high-CRR leaders are not immune: NVIDIA's premium carries an assumption of sustained dominance, so any competitive or efficiency shock, such as custom chips or more efficient open-source models, can de-anchor the valuation."],"supporting_citations":[{"why":"Supplies the anchoring-bias result that underpins the paper's behavioral mechanism for valuations anchored on AI capability.","marker":"[17]"},{"why":"Records OpenAI's early $29 billion valuation, the initial capability anchor used to illustrate a low-CRR premium.","marker":"[23]"},{"why":"Documents OpenAI's jump to $80 billion, the over-anchored re-rating at the heart of the OpenAI case study.","marker":"[24]"},{"why":"Provides the NVIDIA earnings-surge evidence that the paper uses to show high CRR justifying a high premium.","marker":"[10]"},{"why":"Reports Adobe's single-day stock plunge on weak AI monetization, the paper's key evidence for a low-CRR correction.","marker":"[9]"},{"why":"Cited for the macro-level projection that generative AI could raise global GDP by 7%, which the paper treats as an economy-wide anchor for valuations.","marker":"[27]"},{"why":"Documents Microsoft's $10 billion OpenAI investment, the paper's example of an incumbent bridging capability and realization.","marker":"[14]"},{"why":"Confirms NVIDIA's record revenue and $4 trillion market cap, the paper's high-realization counterweight to speculative premiums.","marker":"[11]"}],"fun_headline_variants":["Capability Realization Rate flags AI stock mispricing","New model measures AI's value gap via Capability Realization Rate","AI valuation misalignment risk measured by CRR model","Track AI stocks with Capability Realization Rate to catch mispricing","Capability Realization Rate: how much AI promise is realized"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The model assumes that total AI capability potential is a well-defined, measurable quantity that can be compared across firms; if capability potential cannot be quantified, CRR cannot be computed and the paper's classification of firms as high- or low-CRR reduces to case-by-case judgment rather than measurement.","fun_headline_variants_meta":{"raw":{"variants":["Capability Realization Rate flags AI stock mispricing","New model measures AI's value gap via Capability Realization Rate","AI valuation misalignment risk measured by CRR model","Track AI stocks with Capability Realization Rate to catch mispricing","Capability Realization Rate: how much AI promise is realized"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000761,"raw_usage":{"total_tokens":3352,"prompt_tokens":895,"completion_tokens":2457,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":511,"completion_tokens_details":{"reasoning_tokens":2371}},"tokens_in":511,"tokens_out":2457,"duration_ms":19670,"temperature":1.0,"reasoning_tokens":2371,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T21:21:30.886472+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Construct a panel of listed AI-exposed firms over 2023–2025 using a standardized proxy for CRR, such as AI-attributable revenue divided by total revenue or R&D. If low-CRR firms do not systematically experience larger valuation drawdowns or worse forward returns than high-CRR firms within the same sector and size bucket, the central claim that CRR identifies misalignment risk would be empirically undermined.","supporting_citations":[{"cited_title":"The role of anchoring bias in the equity market","cited_arxiv_id":null,"evidence_quote":"Supplies the anchoring-bias result that underpins the paper's behavioral mechanism for valuations anchored on AI capability."},{"cited_title":"From nonprofit to $29 billion valuation - the promise and danger of openai, Jan 2023","cited_arxiv_id":null,"evidence_quote":"Records OpenAI's early $29 billion valuation, the initial capability anchor used to illustrate a low-CRR premium."},{"cited_title":"Openai seals tender offer with a valuation of $80 billion, Feb 2024","cited_arxiv_id":null,"evidence_quote":"Documents OpenAI's jump to $80 billion, the over-anchored re-rating at the heart of the OpenAI case study."},{"cited_title":"Nvidia stock surges off huge ai-focused earnings report, May 2023","cited_arxiv_id":null,"evidence_quote":"Provides the NVIDIA earnings-surge evidence that the paper uses to show high CRR justifying a high premium."},{"cited_title":"Adobe shares plunge amid ai monetization concerns","cited_arxiv_id":null,"evidence_quote":"Reports Adobe's single-day stock plunge on weak AI monetization, the paper's key evidence for a low-CRR correction."},{"cited_title":"The bear case for ai, Jul 2024","cited_arxiv_id":null,"evidence_quote":"Cited for the macro-level projection that generative AI could raise global GDP by 7%, which the paper treats as an economy-wide anchor for valuations."},{"cited_title":"Microsoft and openai extend partnership, Feb 2023","cited_arxiv_id":null,"evidence_quote":"Documents Microsoft's $10 billion OpenAI investment, the paper's example of an incumbent bridging capability and realization."},{"cited_title":"Nvidia becomes first company to surpass $4 trillion in market valuation","cited_arxiv_id":null,"evidence_quote":"Confirms NVIDIA's record revenue and $4 trillion market cap, the paper's high-realization counterweight to speculative premiums."}],"review_version":1}