Skip to content

Limitations

This page documents the known limitations of NMTC Application Builder honestly. Understanding these limitations is necessary for using the tool responsibly.

Alignment score is not win probability

This cannot be overstated. The composite score returned by score_win_probability() applies the CDFI Fund's published CY 2024-2025 Review Process structure with this tool's own sub-criteria, weights and thresholds (Special Targeting, 5 of Community Outcomes' 50 points, is this tool's own criterion, not the Fund's) — it is not the probability of receiving an NMTC allocation.

The distinction matters practically: - A score of 80/100 does not mean "80% chance of winning." It means the application scores well against that structure in most sections. - A score of 40/100 does not mean "40% chance of winning." It means the application has significant gaps against that structure. - An application that scores 90/100 may still not be funded if a competitor scored 95/100 and the program is oversubscribed. An application that scores 55/100 may be funded if it has other strengths the model does not capture.

The CDFI Fund uses a multi-criterion scoring rubric that includes qualitative assessments of management capacity, community engagement, and organizational track record — dimensions that NMTC Application Builder does not score. A high alignment score is a positive indicator, not a guarantee.

Data recency limitations

The round-level statistics embedded in the library (NMTC_AWARD_ROUNDS in historical_awards.py) come from CDFI Fund award announcements. Its winner pattern figures (the WINNER_* dicts) are this tool's own assumed winner patterns, which are unsourced house constants — not measurements of past winners, and not a CDFI Fund publication. This creates two limitations:

  1. Round-level data has a vintage. NMTC_AWARD_ROUNDS runs CY 2020 through CY 2024-2025. The CY2024-2025 row carries the CDFI Fund's own Award Book figures for that double round (awarded 23 Dec 2025); there is no CY2024 row. CY 2026 figures can only be added once that round is awarded.

  2. NOFA criteria change. The CDFI Fund revises scoring criteria between rounds. A criteria change that shifts relative weights — for example, increasing the weight on geographic diversity or adding new bonus categories — would require recalibration of the dimensional weights in WinProbabilityModel. The current library encodes the CY 2024-2025 Allocation Application and Review Process, as a disclosed proxy for CY 2026 (see the round-provenance note on every page and in every generated document).

Practitioners should always verify that the most recent NOFA aligns with the scoring assumptions the library uses.

What the model cannot do

Cannot evaluate qualitative dimensions. The CDFI Fund scores applications on Management Capacity (organizational track record, board composition, CDFi experience) and Business Strategy (narrative coherence, market analysis quality) in ways that require human review. NMTC Application Builder does not assess the quality of your organizational narrative, the strength of your board composition, or the credibility of your market opportunity description.

Cannot assess investor relationships. Having committed investors dramatically improves both the credibility of the application and the probability of deploying capital. The library models award size fit (against a house award-size partition, which is unsourced and is not a measured winner range) but does not model the strength of investor relationships or the certainty of capital deployment.

Cannot model reviewer subjectivity. NOFA review involves human readers who may weigh criteria differently from the mechanical model. The alignment score is a systematic approximation — actual scoring by CDFI Fund reviewers involves judgment that cannot be fully captured.

Cannot predict NOFA changes. The CDFI Fund sometimes introduces new priority categories (Opportunity Zones, Native American areas, climate resilience) mid-program. The model reflects historical criteria but cannot anticipate future NOFA modifications.

Geographic data limitations

The pipeline map (plot_pipeline_map) positions projects at state centroids, not exact addresses. This is a visualization limitation — the intelligence analysis itself uses census tract-level data from nmtc-mapper, which is tract-accurate.

If a pipeline has many projects in the same state, the map will show overlapping dots (with small jitter applied) rather than the actual geographic spread of projects within the state. Do not use the pipeline map as a substitute for a precise geographic analysis.

State centroids used are standard geographic center points of the lower 48 states and DC. Alaska and Hawaii are excluded from the contiguous map view.

Non-public application data means benchmarks are approximate

This library's winner pattern statistics are this tool's own assumed winner patterns, which are unsourced house constants — not measurements of past winners, and not a CDFI Fund publication. They were once described as inferred from CDFI Fund press releases and annual reports; the annual-report series they cited does not exist (see the module docstring of historical_awards.py), and tests/scoring_attribution.txt rules every key HOUSE. The CDFI Fund does not publish application-level data for either winners or non-winners.

Specifically, none of the distress, geographic (states, HHI) or impact figures is derived from a dataset of winner applications, and the percentiles are not computed from any sample. Treat each as a house reference point for reading your own pipeline, not as evidence about where you stand against real applicants.

The optimizer does not guarantee a win

The pipeline optimizer selects the project subset that maximizes alignment score subject to your constraints. This is useful for identifying the best pipeline composition from an available candidate pool — but the selected subset is the optimal answer to the alignment question, not to the funding question.

A pipeline optimized to score 82/100 on alignment does not have a higher probability of winning than a 78/100 pipeline by any quantifiable margin — because we do not know the distribution of competing applications in any given round. The optimizer is a composition tool, not a funding predictor.

Impact projections are unverified

The jobs-per-million-QEI metric, which carries 25% weight in the alignment score, is computed directly from expected_jobs_created values entered in each PipelineProject. The platform does not verify these projections against comparable deals, market conditions, or industry benchmarks.

CDEs are responsible for ensuring that job projections are reasonable and defensible. Overstated projections will produce an inflated impact score. CDFI Fund reviewers have access to historical data and may scrutinize impact claims — the platform's high impact score will not protect an application with unrealistic projections from reviewer skepticism.

Recommendation: always involve a CDFI expert

NMTC Application Builder is a diagnostic and benchmarking tool — a starting point for analysis, not a substitute for practitioner judgment. Before submitting an application, the pipeline, alignment scores, and recommendations should be reviewed by a qualified CDFI/NMTC practitioner who can:

  • Assess qualitative dimensions not captured by the model
  • Verify that job projections and impact claims are defensible
  • Confirm that the application narrative is coherent and compelling
  • Review the deal structure and investor strategy
  • Evaluate the application in the context of the specific round's competitive landscape

The library helps you understand your quantitative position. Allocations are won by practitioners, not algorithms.