{"schema_version":1,"research_id":"eoa_inverse_innovation_exp03_external48_20260801","source_assessment_id":"eoa_inverse_innovation_exp03_opportunity320_20260801","cell_id":"invariant_mode_decomposition_design__innovation_entrepreneurship","selection_stratum":"DEPLOYABLE_PRIORITY","search_queries":["startup metric dashboard decision making KPI conflict metric fixation runway research","software startups metrics empirical study decision making uncertainty metrics startup primary research","startup pivot decision metrics empirical longitudinal study","Strategy Selection Surrogation Strategic Performance Measurement Systems","dynamic mode decomposition business performance metrics","dynamic mode decomposition marketing sales customer","Business Dynamics in KPI Space controllable KPI invariant","dynamic mode decomposition business KPI optimization","dynamic mode decomposition multichannel sales forecasting","dynamic mode decomposition with control Proctor Brunton Kutz","Koopman operator business decision support metrics","online dynamic mode decomposition time-varying systems","dynamic mode decomposition noisy data eigenvalue bias","Amplitude experiment guardrail metrics holdout official","BLS data scientist median annual wage official","NIST Privacy Framework data governance official"],"sources":[{"source_id":"S1","title":"100+ Metrics for Software Startups—A Multi-Vocal Literature Review","publisher":"CEUR Workshop Proceedings / arXiv","url":"https://arxiv.org/abs/1901.04819","source_class":"PRIMARY_RESEARCH","publication_date":"2019-01-15","accessed_at":"2026-08-02","claims_supported":["Software-startup guidance contains a large and context-dependent set of proposed metrics.","The authors found limited startup-specific academic coverage and cautioned that the compiled practitioner recommendations were not empirically verified."]},{"source_id":"S2","title":"Managing Complexity and Unforeseeable Uncertainty in Startup Companies: An Empirical Study","publisher":"INFORMS, Organization Science","url":"https://doi.org/10.1287/orsc.1080.0369","source_class":"PRIMARY_RESEARCH","publication_date":"2008-07-25","accessed_at":"2026-08-02","claims_supported":["A study of 58 startups found that suitable learning and planning approaches depend jointly on uncertainty and complexity.","Under unforeseeable uncertainty, startups may repeatedly change goals and courses of action as information arrives."]},{"source_id":"S3","title":"Strategy Selection, Surrogation, and Strategic Performance Measurement Systems","publisher":"Wiley, Journal of Accounting Research","url":"https://doi.org/10.1111/j.1475-679X.2012.00465.x","source_class":"PRIMARY_RESEARCH","publication_date":"2012-07-14","accessed_at":"2026-08-02","claims_supported":["Experimental evidence shows that managers can treat performance measures as substitutes for the strategic constructs they represent.","Managerial involvement in selecting strategy reduced this surrogation effect in the studied setting."]},{"source_id":"S4","title":"Business Dynamics in KPI Space: Some Thoughts on How Business Analytics Can Benefit from Using Principles of Classical Physics","publisher":"arXiv","url":"https://arxiv.org/abs/1702.01742","source_class":"PRIMARY_RESEARCH","publication_date":"2017-02-06","accessed_at":"2026-08-02","claims_supported":["Prior work explicitly proposed separating volatile externally driven KPIs from relatively stable controllable business KPIs.","It proposed finding dynamical laws and invariants of controllable KPIs for risk indicators, performance indicators, planning, ROI optimization, and business growth."]},{"source_id":"S5","title":"Multifractal Wavelet Dynamic Mode Decomposition Modeling for Marketing Time Series","publisher":"arXiv","url":"https://arxiv.org/abs/2403.13361","source_class":"PRIMARY_RESEARCH","publication_date":"2024-03-20","accessed_at":"2026-08-02","claims_supported":["Dynamic mode decomposition has been applied directly to marketing price and sales-volume time series.","The study used several years of brand data to examine persistence and forecasting rather than startup-wide decision control."]},{"source_id":"S6","title":"A Dynamic Mode Decomposition Approach with Hankel Blocks to Forecast Multi-Channel Temporal Series","publisher":"IEEE Control Systems Letters","url":"https://doi.org/10.1109/LCSYS.2019.2917811","source_class":"PRIMARY_RESEARCH","publication_date":"2019-05-20","accessed_at":"2026-08-02","claims_supported":["DMD was applied to multichannel weekly sales from 2,659 stores.","Hankel augmentation improved aggregate sales forecast fit, but performance remained poor for some stores and depended on measured signals spanning the relevant modes."]},{"source_id":"S7","title":"Dynamic Mode Decomposition with Control","publisher":"SIAM Journal on Applied Dynamical Systems","url":"https://doi.org/10.1137/15M1013857","source_class":"PRIMARY_RESEARCH","publication_date":"2016-01-26","accessed_at":"2026-08-02","claims_supported":["DMD with control separates estimated underlying dynamics from effects associated with observed actuation.","The method provides low-order input-output models, establishing technical prior art for sensitivity or control analysis after modal decomposition."]},{"source_id":"S8","title":"Characterizing and Correcting for the Effect of Sensor Noise in the Dynamic Mode Decomposition","publisher":"arXiv","url":"https://arxiv.org/abs/1507.02264","source_class":"PRIMARY_RESEARCH","publication_date":"2015-07-08","accessed_at":"2026-08-02","claims_supported":["Standard DMD estimates are biased by measurement noise.","The magnitude of the bias depends on dataset size and noise, and modified estimators are needed to mitigate it."] 	} 	] 	,"problem_evidence":{"support":"WEAK","rationale":"The literature supports three adjacent facts: startups operate with scarce resources and context-dependent metrics; startups may repeatedly revise goals under uncertainty; and managers can surrogate strategic objectives with measured KPIs. It does not establish the candidate's specific prevalence claim that coupled acquisition, activation, retention, margin, support, and burn movements routinely cause recurring budget reversals or measurable runway waste in early-stage ventures. The startup-metrics review explicitly notes weak empirical verification, so the exact problem remains a hypothesis requiring decision-log evidence.","source_ids":["S1","S2","S3"]},"stakeholder_evidence":{"support":"WEAK","rationale":"Official product documentation shows a commercial category for startup analytics, multi-metric experimentation, guardrails, holdouts, governance, and forecasting. This establishes category-level interest and recognizable authorizers, but no source shows demand for DMD/Koopman modes, willingness to supply 26–52 comparable weeks, or willingness to act on mode-guided recommendations.","source_ids":["S10","S11"]},"prior_art":{"proximity":"SUBSTANTIAL_COLLISION","closest_analogues":[{"name":"Business Dynamics in KPI Space","similarity":"It already proposes modeling business as a dynamical system of stable versus volatile KPIs, finding business invariants, distinguishing controllable KPIs, and using the results for planning, performance, risk, ROI optimization, and growth. This is the closest conceptual analogue to invariant-mode analysis of coupled venture metrics.","remaining_difference":"It is a conceptual note rather than a validated startup implementation; it does not specify DMD/Koopman estimation, short-window resampling and conditioning gates, a regularized nonmodal comparator, shadow evaluation, or a capped reversible allocation test.","source_ids":["S4"]},{"name":"DMD forecasting of multi-channel retail sales","similarity":"It fits DMD modes to many coupled weekly business time series and uses the estimated dynamics for out-of-sample sales forecasting.","remaining_difference":"The state consists of store sales rather than cross-functional venture-health metrics; the work forecasts rather than recommends budget controls and does not use the candidate's regime, residual, drift, harm, or decision-authority gates.","source_ids":["S6"]},{"name":"Wavelet-DMD marketing time-series modeling","similarity":"It applies dynamic mode decomposition to jointly relevant marketing quantities, including brand sales and prices, to characterize persistence and forecast behavior.","remaining_difference":"It uses long brand histories and does not address startup runway, cross-metric priority reversals, intervention sensitivity, a nonmodal rival, or reversible experiment allocation.","source_ids":["S5"]},{"name":"Dynamic Mode Decomposition with Control","similarity":"It supplies the established technical mechanism for estimating modes while separating observed actuation from underlying dynamics and producing an input-output model.","remaining_difference":"It is domain-general engineering prior art, not evidence that sparse, endogenous venture histories possess stable or useful modes or that mode-guided venture allocations outperform regularized multivariate prediction.","source_ids":["S7"]},{"name":"Amplitude multi-metric experimentation practice","similarity":"Current commercial tooling already combines primary and secondary metrics, guardrail metrics, holdouts, approval workflows, cohort analytics, and controlled experiments for product decisions.","remaining_difference":"It does not decompose a venture-wide transition operator into invariant modes or recommend reallocations from modal sensitivities.","source_ids":["S10","S11"]}],"distinctive_claim_remaining":"The remaining testable distinction is narrow: within a declared locally stable startup regime, do noise-corrected, resample-stable coupled modes add out-of-sample objective prediction and recommendation value beyond isolated-KPI review and an equally regularized nonmodal transition model, while passing prespecified residual, conditioning, gap, drift, privacy, and harm gates? The ingredients and broad business-dynamics concept are prior art; only this gated empirical combination remains potentially differentiating.","confidence":"HIGH"},"implementation_evidence":{"support":"MODERATE","rationale":"Primary research establishes executable DMD, DMD with control, mode-accuracy diagnostics, noise corrections, and online/windowed variants for time-varying systems. Business applications include multichannel sales and marketing series. However, standard DMD can be biased by noisy observations, can miss modes not spanned by measured signals, and a time-invariant model can be inappropriate for time-varying dynamics. Nothing found validates estimation from only 26–52 endogenous startup weeks. Current product tooling makes holdouts, guardrails, permissions, and metric governance feasible, while NIST guidance supports privacy-risk governance. Thus implementation is technically possible, but usable estimation in the proposed data regime is uncertain.","source_ids":["S5","S6","S7","S8","S9","S10","S13","S14"]},"scores":{"meaningful_impact":{"score":3,"rationale":"Better allocation of scarce venture learning resources could matter, and startup research confirms that uncertainty and complexity affect successful learning strategies. The frequency and runway cost of the candidate's exact metric-thrashing problem remain unmeasured.","source_ids":["S1","S2","S3"]},"stakeholder_pull":{"score":2,"rationale":"Commercial analytics and experimentation products explicitly target early-stage startups and offer guardrails, holdouts, and causal-analysis features. This is pull for the surrounding workflow, not demonstrated demand for coupled-mode analysis.","source_ids":["S10","S11"]},"incremental_advantage":{"score":2,"rationale":"DMD can exploit coupled temporal structure, but prior work already applies modes to business forecasting, and established experimentation platforms already handle multiple outcomes and guardrails. No evidence shows an advantage over regularized multivariate models or simpler experiment analysis in sparse venture data.","source_ids":["S4","S5","S6","S7","S10"]},"distinctiveness_plausibility":{"score":2,"rationale":"Business KPI dynamics, dynamical invariants, DMD business forecasting, DMD with control, and guarded multi-metric experimentation all predate the candidate. The remaining distinction is primarily the particular gated startup evaluation protocol rather than a new underlying method.","source_ids":["S4","S5","S6","S7","S10"]},"technical_implementability":{"score":3,"rationale":"The algorithms and diagnostics are implementable with ordinary data-science infrastructure. Noise bias, missing observables, endogenous interventions, short history, time variation, and possible absence of a stable eigengap materially threaten usefulness.","source_ids":["S6","S7","S8","S9","S14"]},"adoption_authority_feasibility":{"score":4,"rationale":"The proposed founders and functional leads match ordinary analytics and experimentation decision roles. Existing products support manager/admin permissions and approval workflows, while the first study can remain retrospective and shadow-only. Cross-functional consent and privacy review are still required.","source_ids":["S10","S11","S13"]},"evidence_readiness":{"score":3,"rationale":"A rolling-origin retrospective comparison and shadow forecasts are concrete and inexpensive relative to deployment. Readiness is reduced because startup-specific metric evidence is weak and no source establishes that 26–52 weeks suffice for stable mode estimation after regime restriction and holdout allocation.","source_ids":["S1","S6","S8","S9","S14"]},"safety_net_benefit":{"score":4,"rationale":"Shadow-only evaluation, preregistration, holdouts, guardrail metrics, residual and drift stops, human approval, and privacy governance create strong opportunities to reject the method before customer-facing action. They cannot prevent misleading interpretation if leaders treat descriptive modes as causal.","source_ids":["S3","S10","S13"]},"scalability":{"score":2,"rationale":"Software computation is inexpensive, but every venture and regime requires metric reconciliation, input/actuation reconstruction, privacy review, stability testing, and refitting after strategic change. Time-varying-system research implies that ongoing window selection and updating are load-bearing rather than optional.","source_ids":["S8","S9","S11","S13"]}},"score_confidence":"MODERATE","costs":{"first_evidence":{"band_2026_usd":"50K_TO_250K","scope":"One venture, retrospective and shadow-only: reconcile metric definitions; document management inputs and decision reversals; prepare and permission 26–52 historical weekly states; implement noise-corrected modal, isolated-KPI, and regularized nonmodal models; preregister rolling-origin comparisons; run four shadow weeks; conduct privacy, security, and stakeholder review; and produce an evaluation record. Includes specialized labor, leadership coordination, data engineering, software/cloud use, compliance review, and evaluation.","confidence":"MODERATE","assumptions":["Existing analytics, finance, support, and experiment logs are accessible without rebuilding the venture's data platform.","Labor dominates cost: the BLS reports a May 2024 data-scientist median wage of $112,590 before benefits and overhead, and the study also requires product, growth, finance, governance, and data-engineering time.","Product-analytics software may be free or discounted for a qualifying startup, but advanced experimentation and governance can require custom-priced tiers.","No live allocation or customer experiment occurs in this phase."],"source_ids":["S10","S11","S12","S13"]},"initial_deployment_startup":{"band_2026_usd":"50K_TO_250K","scope":"If retrospective gates pass, establish a repeatable single-regime pipeline with versioned metric definitions, access controls, input and decision logging, noise and conditioning diagnostics, drift monitoring, approval workflow, rollback documentation, comparator retraining, and independent review. Includes labor, data work, privacy/compliance, coordination, software, cloud infrastructure, and validation.","confidence":"MODERATE","assumptions":["The venture can adapt existing warehouse and experiment infrastructure.","Deployment remains advisory, human-reviewed, and limited to one declared product/segment/pricing/channel regime.","Major instrumentation gaps, identity-resolution work, or external legal review could move the effort above this band."],"source_ids":["S9","S10","S11","S12","S13"]},"operational_launch":{"band_2026_usd":"10K_TO_50K","scope":"One authorized two-week evaluation of a reversible reallocation capped at 5% of an existing experiment budget, only after all prior gates pass. Includes preregistered holdout and rival recommendations, experiment operations, analyst and leadership review, privacy and guardrail monitoring, software/cloud use, evaluation, and rollback readiness; it excludes creating a new product program or automating decisions.","confidence":"LOW","assumptions":["The test reallocates an existing experiment budget and uses existing instrumentation.","No protected-group targeting, employment action, pricing or eligibility change, or additional sensitive-data collection is permitted.","The band excludes the economic opportunity cost of an inconclusive experiment and could be exceeded if adequate power requires a longer test."],"source_ids":["S10","S11","S12","S13"]},"annual_recurring":{"band_2026_usd":"50K_TO_250K","scope":"Operate a small number of declared regimes with periodic refitting, window and mode-stability checks, comparator maintenance, metric-definition governance, access reviews, drift and residual monitoring, human decision review, software/cloud services, privacy/compliance work, incident handling, and evaluation records.","confidence":"LOW","assumptions":["A fractional specialist plus recurring product, growth, finance, data, and governance participation is required.","Advanced product analytics may be custom-priced, although startup discounts and low-cost tiers exist.","Each major strategy, pricing, channel, data-definition, or instrumentation change triggers revalidation rather than automatic portability."],"source_ids":["S9","S11","S12","S13"]}},"verified_pipeline_gates":{"externally_supported_problem":{"status":"UNCERTAIN","reason":"Adjacent evidence supports metric proliferation, strategic surrogation, scarce resources, and iterative strategic change, but no opened source demonstrates recurring cross-metric budget reversals or associated runway waste in early-stage ventures. The exact problem must be established from decision logs before method evaluation.","source_ids":["S1","S2","S3"]},"externally_credible_adopter_or_authorizer":{"status":"YES","reason":"Official first-party documentation shows that startup teams are an explicit market for governed analytics and experimentation and that manager/admin roles and approval workflows are normal. Founders plus accountable product, growth, finance, data, and privacy/experiment owners are credible authorizers for a shadow-only study.","source_ids":["S10","S11","S13"]},"distinct_testable_incremental_claim":{"status":"YES","reason":"Despite substantial prior-art overlap, a narrow falsifiable claim remains: gated modal analysis must outperform both isolated-KPI review and an equally regularized nonmodal transition model on rolling-origin objective prediction and recommendation agreement, not merely produce interpretable modes.","source_ids":["S4","S6","S7","S10"]},"bounded_next_evidence_step":{"status":"YES","reason":"One venture, one declared regime, 26–52 historical weeks, four prospective shadow weeks, fixed comparators, preregistered metrics, and explicit rejection criteria form a bounded non-deployment study.","source_ids":["S6","S8","S9","S10"]},"no_unresolved_safety_or_authority_stop":{"status":"YES","reason":"The next step changes no live budget or product decision. It can use minimized, permissioned data, human review, documented roles, and privacy-risk governance. Any lack of authorization, data-purpose incompatibility, or privacy control is itself a stop condition.","source_ids":["S10","S13"]},"credible_cost_scope_and_range":{"status":"YES","reason":"The four ranges explicitly cover specialized labor, cross-functional coordination, data preparation, software/cloud use, privacy/compliance, experiment operations, monitoring, and evaluation. Official wage and commercial-pricing evidence supports labor-dominant broad bands, although organization-specific costs remain uncertain.","source_ids":["S11","S12","S13"]}},"next_evidence_step":"With one consenting venture, preregister a retrospective plus four-week shadow-only study for one unchanged product/segment/pricing/channel regime. First audit decision logs to test whether at least two recurring cross-metric budget or priority reversals with documented objective or runway consequences occurred; absence falsifies the stated problem for that venture. Using only information available at each forecast date, compare noise-corrected gated DMD/DMD-with-control against (1) isolated-KPI persistence or threshold forecasts and (2) an equally tuned regularized VAR/elastic-net cohort-transition model on rolling-origin prediction of the full next-week state and sustainable cohort contribution or validated-learning yield per runway. Reject progression if no mode survives bootstrap direction-stability, residual-whiteness, conditioning, eigengap, and definition-drift gates, or if the modal method fails to improve preregistered out-of-sample loss beyond uncertainty. Record hypothetical recommendations and guardrail breaches during four shadow weeks, but alter no budget, product, pricing, eligibility, staffing, or customer treatment.","blocking_evidence":["Prevalence and materiality of recurring cross-metric decision reversals in an actual consenting venture.","Availability of 26–52 weeks with stable metric definitions and a genuinely unchanged product, segment, pricing, and channel regime.","Adequate records of prior budget reallocations and management actions needed to separate inputs from endogenous state dynamics.","Bootstrap-stable, well-conditioned modes with an adequate eigengap and unstructured residuals.","Out-of-sample advantage over isolated-KPI forecasts and an equally regularized nonmodal transition model.","Evidence that selected modes are controllable by feasible allocations rather than descriptive correlations.","Reliable, non-gameable measurement of validated learning and cohort contribution per runway alongside customer, employee, cash, and compliance harms.","Decision-maker willingness to use shadow outputs without treating modal coordinates as causal entities.","Portability beyond one venture and one locally stable regime."],"research_disposition":"PARTNERED_RESEARCH_PROGRAM","world_novelty_boundary":"This bounded English-language web and scholarly search found substantial conceptual and technical prior art: dynamical business KPIs and invariants, DMD applied to sales and marketing series, DMD with control, and commercial multi-metric experimentation with guardrails and holdouts. It did not locate a single opened source implementing the candidate's exact regime-bounded startup state, diagnostic gates, regularized nonmodal comparison, shadow study, and capped reversible allocation protocol. That remaining difference is an empirical differentiation hypothesis, not a world-novelty, patent-validity, freedom-to-operate, or exhaustive-literature conclusion; proprietary systems, non-English work, unindexed publications, and unreviewed patent claims may contain closer matches."}