Jamie Gomez · Insight Quantix · November 21, 2026
Document: IQ-METH-PFF-2026-02 · Reference version: 2.0
The question: Where does a pathway match its benchmark, and what could make it lose that advantage?
The finding: DG-PFF first maps where the modeled project cost meets a stated comparison price. It then tests how much of that region survives operating, market, and policy changes.
What could change it: The benchmark, input ranges, operating constraints, and the way scenarios and policy support are calculated.
The implication: Read a classification as the result of stated tests. It does not establish financing, permitting, or investment suitability. The equations below show how the tests work.
The Decision-Grade Parity-Fragility Framework (DG-PFF) organizes techno-economic evidence around two questions: where does a pathway meet a stated economic benchmark, and where does that conclusion fail under operating, market, and policy constraints? This reference formalizes the parity margin, feasible and surviving regions, collapse boundaries, invalidation share, and deterministic classification sequence. It distinguishes scenario coverage from probability, framework conformance from project merit, and analytical results from investment authority. Its contribution is the integration of established analytical methods into a traceable decision protocol. It does not claim invention of levelized cost, sensitivity analysis, scenario analysis, or robust optimization.
This is the definitive technical explanation of DG-PFF published by Insight Quantix at reference version 2.0. It accompanies the public framework standard v1.0 and documents the repository’s deterministic ruleset v1.2. These are separate version identifiers. This publication does not silently upgrade the rules engine, recertify earlier applications, or change historical decisions.
The source baseline is repository commit 987a2d9e2283ceef4a1696c6b1b00cbadfd3ef23; rules, schema, and workflow links in this reference are pinned to that revision. The YAML is a declarative rule specification. The reviewed repository sources include an intake schema and workflow instructions, but this review did not locate a tested evaluator consuming that YAML. Consequently, statements about classification describe the specified rules, not verified execution by a deployed engine. Hand calculations below check the specification only.
The equations below make the framework’s terms precise. Section 6 reproduces the existing v1.2 rule profile; sections 7–10 explain interpretation, evidence requirements, and limitations. A report must name its framework standard, methodology reference, model revision, and rule profile separately. Where a legacy summary conflicts with an explicit equation or rule, disclose the discrepancy and its effect; do not combine incompatible definitions under one version label.
DG-PFF applies to benchmark-based pathway screening, pre-FID analysis, and technical diligence. A parity benchmark alone cannot establish bankability, creditworthiness, financeability, environmental acceptability, or an investment recommendation. The decision owner retains responsibility for those judgments.
Discounted cash flow, minimum selling price, sensitivity analysis, uncertainty propagation, and stress testing precede DG-PFF. NREL’s comparative economics methodology describes minimum selling price and sensitivity analysis; Quantifying breakeven price distributions in stochastic techno-economic analysis addresses breakeven-price distributions in stochastic TEA. Those methods remain the analytical foundation.
The DG-PFF contribution is a specified reporting and decision architecture:
This is an originality claim about the framework’s synthesis and formalization by Insight Quantix. It is not evidence of worldwide priority, patentability, independent validation, or superior predictive accuracy. Establishing those claims would require a systematic prior-art review and comparative empirical testing. Threshold terminology and a named protocol do not by themselves establish mathematical novelty.
Let x denote uncertain or controllable inputs, d a design, and Ω the explicitly bounded admissible input domain. Use a common currency year, functional unit, product specification, delivery point, tax basis, and time horizon for project and incumbent costs. State whether prices are real or nominal and whether credits are included in the cost or benchmark. Count each credit once.
| Symbol | Definition |
|---|---|
C(d,x) |
Project delivered cost or minimum selling price in the declared unit |
B(x) |
Delivered displacement benchmark in the same unit; may vary with the scenario |
M(d,x) = B(x) − C(d,x) |
Absolute parity margin; positive values favor the project |
m(d,x) = M(d,x) / B(x) |
Relative parity margin, defined only when B(x) > 0 |
g_j(d,x) |
Constraint residual; constraint j passes when g_j ≥ 0 |
x₀ |
Documented base case |
Parity holds at equality: M ≥ 0. Do not round a negative margin to zero before classification. Record numerical tolerances, solver residuals, and any boundary ambiguity. A zero or negative benchmark permits an absolute margin but invalidates the relative-margin formula and the v1.2 normalized profile until an alternative profile is specified.
A transparent annualized screening model is:
Q(u) = Q_rated × u
CRF(r,n) = r(1+r)^n / [(1+r)^n − 1] for r > −1, r ≠ 0, n > 0
CRF(0,n) = 1/n for n > 0
C(d,x) = [K(d) × CRF(r,n) + F(d,x)] / Q(u)
+ v(d,x) + L(d,x) − S(d,x)
In plain language: annual fixed costs are divided by the amount produced. Running fewer hours usually raises the fixed cost per unit. Delivery adds cost, while realized support reduces the net cost.
r is the effective annual discount rate and n the project life in years. K is installed capital, F annual fixed expense, Q_rated annual output at full rated operation, u realized utilization, v variable cost per unit, L delivery cost per unit, and S realized support per unit. Require Q(u) > 0; zero production is an operational failure, not a finite unit cost. This approximation assumes annualized steady operation and consistent treatment of output and costs.
For financing-sensitive work, calculate minimum selling price from a documented discounted-cash-flow model: solve NPV(P; d,x) = 0 at the stated return requirement. Include construction schedule, ramp-up, taxes, replacement capital, working capital, financing, and terminal assumptions where material. The annualized expression is not interchangeable with a full project-finance model.
Utilization is not automatically renewable capacity factor. Distinguish resource availability, equipment capacity factor, availability, conversion efficiency, storage losses, and saleable output. Couple electricity prices and achievable operating hours where cheap power depends on restricted dispatch. For continuous synthesis supplied by intermittent hydrogen, annual averages do not establish hourly feasibility.
Fix d and define Product A on a common domain:
V_A = {x ∈ Ω : M_A(d,x) ≥ 0}
V_B = {x ∈ V_A : M_B(d,x) ≥ 0 and g_j(d,x) ≥ 0 for every j}
I_μ = μ(V_A minus V_B) / μ(V_A) when 0 < μ(V_A) < ∞
In plain language: Product A contains the conditions that pass the first cost screen. Product B contains the conditions that still pass after the additional constraints. The invalidation share measures how much of the original passing region is lost.
M_A uses the stated structural screening assumptions; M_B includes the documented degradation and constraints. μ is a declared nonnegative measure, such as grid area, weighted scenario count, or a justified probability measure; require measurable regions and a finite positive denominator for invalidation share. Product B is the portion of the original parity region that survives the specified constraints. If redesign creates new viable conditions outside V_A, report that region separately and rerun Product A.
Never compare areas with different axes, domains, resolutions, or weights as though they were the same denominator. If μ(V_A) = 0, invalidation share is undefined: report “zero viable measure under the selected measure.” This does not establish that V_A is empty: an isolated viable point or boundary can have zero area or probability mass. Report emptiness separately when established. A grid-based invalidation percentage is not a probability unless the grid weights represent a justified distribution. A finite grid can miss narrow surviving or failing regions; refine near boundaries and report resolution.
A conditional economic threshold solves M_B(d,x_i*,x_−i) = 0 with the other inputs held at named values. If the parity margin is strictly decreasing with x_i throughout the declared interval, economic parity fails above the root; if strictly increasing, it fails below it. Monotonic project cost alone is insufficient when the benchmark also changes. With non-strict monotonicity, a zero-margin plateau can remain at parity; identify its endpoints rather than assigning failure at every point beyond one root. Operational constraints can fail before the economic root and must be checked separately. Multiple roots, disconnected viable regions, discontinuous credit tiers, and operating limits require explicit handling; a discontinuous crossing need not have a zero-margin root.
The v1.0 standard’s stronger collapse language, failure regardless of favorable assumptions in other variables, requires a stronger demonstration. For a fixed design and declared admissible domain, economic structural collapse at x_i = t means every constraint-feasible completion has M_B(d,t,z) < 0. A sufficient certificate with a strictly negative uniform margin is:
sup { M_B(d,t,z) : (t,z) ∈ Ω and every g_j(d,t,z) ≥ 0 } < 0
In plain language: this test asks whether even the best allowed combination of the other inputs still fails the cost comparison. The conditions below explain when that conclusion can be established.
A zero supremum is inconclusive unless attainment is established: all feasible margins could be negative while approaching zero. On a nonempty compact feasible set with a continuous margin, the maximum is attained, and a negative maximum is equivalent to economic collapse. If that constraint-feasible set is empty, operational feasibility has already failed. A one-variable sweep at base values does not prove structural collapse. Label it a conditional parity-loss threshold unless the favorable feasible envelope has been checked. If designs are allowed to change, include the admissible designs in the optimization before claiming collapse across designs. A numerical solver’s failure to find a viable solution is not proof that none exists.
For the annualized model, holding other terms fixed, set A = (K×CRF + F)/Q_rated and h = B − v − L + S. When A > 0 and h > 0, parity requires u ≥ u* = A/h. If u* > 1, parity is unattainable within 0 < u ≤ 1 under those assumptions. If h ≤ 0, positive fixed costs prevent parity at every positive utilization.
For a nonzero base value and a specified threshold, report D_i = |x_i* − x_i₀| and D_i / |x_i₀| × 100%, together with units and failure direction. If the base is zero, percentage distance is undefined. If no threshold occurs in the search domain, say so rather than extrapolating. The v1.0 near-degenerate warning applies below 5% distance; it is a reporting trigger, not an empirically calibrated failure probability.
A reproducible one-at-a-time ranking can use Δ_i = max |M(d,x_i,x_−i₀) − M(d,x₀)| over each declared stress range, then share_i = Δ_i / ΣΔ_j. Report all denominator drivers and the top three in the decision summary. If all movements are zero, shares are undefined. These shares depend on selected ranges and omit interactions; they are not causal attribution or variance-based sensitivity indices. Joint stresses must supplement them where variables interact.
Report the five v1.0 comparative metrics: dominant-variable share; distance to collapse; distinct one-variable collapse-driver count; policy durability risk class; and standalone-use warning. “Independent” in the legacy count means distinct drivers, not statistical independence. Count only structural thresholds as structural collapse; list conditional thresholds separately. For policy risk (Low, Medium, High, Severe) and standalone warnings (Green, Amber, Red), publish the assessment rubric and evidence. The public standard does not supply universal numerical cutoffs for those categories.
Separate variability in operations or markets, uncertainty in estimated parameters, uncertainty about model structure, and policy scenarios whose probabilities cannot be defended. Each input record should identify its source, date, geography, units, range, dependence assumptions, and whether it is measured, estimated, or judgmental. The distinction between qualitative confidence and quantified likelihood is consistent with the IPCC uncertainty guidance; DG-PFF does not adopt IPCC probability terminology as its own classification scale.
Deterministic stress testing is the baseline. The v1.2 YAML requires capacity-factor variation, delivered-electricity-price variation, and policy-degradation stress groups. The input schema requires policy-dependent cases to include at least one of partial_loss or full_loss; it does not require both. For cases declared policy-independent, the schema permits none. Utilization and capacity factor are distinct inputs: add utilization stresses where material rather than silently substituting them for the required capacity-factor group. Joint stresses, both policy-loss levels, and feedstock, yield, capital, delivery, storage, and benchmark stresses are recommended extensions where material, not additional requirements already enforced by v1.2. Any claimed inapplicability requires evidence and a separately specified profile; silently omitting a required stress group is non-conformance.
For a finite scenario set S, report:
f_S = number of failed scenarios / number of evaluated scenarios
In plain language: divide the number of failed test cases by the number evaluated. This describes the selected tests, not the probability that a real project will fail.
The YAML specifies the ratio but does not define the scenario failure predicate or scenario-generation algorithm. A complete assessment must supply and version those definitions before computing a classification. A proposed assessment convention is to count economic parity loss or a required operational-constraint breach as failure; this convention is not demonstrated executable v1.2 behavior. Distinguish those outcomes from numerical evaluation failures, which require investigation and cannot silently be counted as successful scenarios or dropped.
Require a nonempty set. State whether the base case is included, how duplicates are handled, which constraints count, and how joint scenarios are constructed. f_S is a scenario failure rate. Adding many benign scenarios can lower it without improving the project. Freeze the scenario design before comparing alternatives and show sensitivity to coverage.
Probabilistic analysis is optional and supplementary. If a defensible joint distribution p(x) exists, estimate Pr(M_B ≥ 0 and all g_j ≥ 0) by sampling or integration. Specify distributions, correlations, truncation, sample size, seed, convergence checks, and sampling uncertainty. For independent identically distributed Monte Carlo draws, the approximate standard error of an estimated probability is sqrt[p̂(1−p̂)/N]; use an appropriate interval near zero or one. This error describes simulation precision, not confidence in the input distributions or model. Correlated, weighted, or structured sampling needs an appropriate estimator.
Do not substitute a Monte Carlo probability for f_S in the existing rule profile without explicitly changing and validating the profile. Where probabilities are indefensible, use bounded scenarios and describe what survives each scenario. “Durable” always refers to the tested domain and evidence.
The public standard uses Durable / Conditional / Fragile to describe persistence. The declarative v1.2 ruleset specifies LOW / MEDIUM / HIGH fragility. These are not interchangeable labels: the former is a qualitative boundary assessment, while the latter follows the numerical rules below. Reports using both must show both, explain disagreements, and identify which controls the decision signal. Do not silently translate Durable to LOW or Conditional to MEDIUM.
Conformance is also separate. NON_CONFORMANT means required evidence or execution is missing or invalid. It does not demonstrate that a project is uneconomic. The ruleset routes it to NO_GO as a process gate: do not approve using this assessment until corrected.
The source rules define required fields and stress groups. Missing required inputs, unexecuted stress groups, a missing capacity-factor operational basis, or electricity bounds that do not satisfy low ≤ base ≤ high trigger non-conformance. Undefined ratios, nonfinite results, inconsistent units, and missing constraints must also be resolved before the equations can be interpreted; this reference does not claim the legacy schema catches every such defect.
For valid inputs, require 0 ≤ f_S ≤ 1 with integer counts satisfying 0 ≤ failed_scenarios ≤ total_scenarios and total_scenarios > 0. A missing or invalid status is not evidence of conformance; although the YAML uses status != NON_CONFORMANT, an assessment must establish a successful validation result before applying the approval branches.
The profile derives:
m₀ = (B₀ − C₀) / B₀
q = realized policy value used in the base case / C₀
C_low = C₀ × p_low / p_base
C_high = C₀ × p_high / p_base
overlap = (C_low ≤ B₀) and (C_high ≥ B₀)
In plain language: the margin measures the base case's cost advantage. The policy ratio measures reliance on support. The legacy electricity band approximates how cost changes with electricity price, and overlap asks whether that band reaches the benchmark.
p is delivered electricity price. The formulas require B₀ > 0, C₀ > 0, and p_base > 0; policy value and C₀ must use the same per-output basis. The ratio q uses the declared baseline project cost denominator; state whether it is net of support. It can exceed one and is not a probability or a gross-cost subsidy fraction.
The input schema permits a zero baseline cost, although q is then undefined. Schema acceptance alone therefore does not establish that classification inputs are valid. The YAML also uses policy_value_used_in_base_case without specifying how to derive it from policy tiers, eligibility, timing, or stacking rules. A case-specific calculation must select eligible support, normalize it to the declared output and time basis, account for realization and stacking restrictions, and reconcile it with support already included in baseline cost. It must identify the chosen gross or net denominator. Until those choices and the scenario failure predicate are recorded, the profile is not a complete reproducible assessment algorithm. These are completion requirements for an assessment, not claims that the legacy schema enforces them.
The electricity band scales the entire project cost with electricity price. This is a screening approximation, not a physical electricity-cost decomposition. It can overstate exposure when electricity is only part of total cost and can misrepresent nonlinear operations, credits, or coupled benchmark changes. A validated cost decomposition or model rerun is preferable, but substituting it changes the profile and must be versioned.
Apply rules in the following order, retaining unrounded values:
| Stage | Condition | Output |
|---|---|---|
| Parity | m₀ ≥ 0; otherwise m₀ < 0 |
HOLDS; otherwise FAILS |
| Fragility, first | Parity FAILS, or f_S ≥ 0.40, or overlap is true |
HIGH |
| Fragility, next | Parity HOLDS, 0.15 ≤ f_S < 0.40, and no overlap |
MEDIUM |
| Fragility, last | Parity HOLDS, f_S < 0.15, and no overlap |
LOW |
| Decision, first | NON_CONFORMANT, or parity FAILS, or fragility HIGH |
NO_GO |
| Decision, next | Conformant, parity HOLDS, fragility MEDIUM |
REWORK |
| Decision, next | Conformant, parity HOLDS, fragility LOW, q > 0.25 |
CAUTION |
| Decision, last | Conformant, parity HOLDS, fragility LOW, q ≤ 0.25 |
PROCEED |
The endpoints matter: with baseline parity and no overlap, 15% failure is MEDIUM; 40% is HIGH regardless of overlap. Exactly 25% policy dependence remains eligible for PROCEED only when conformance, parity, and LOW fragility requirements also pass. Touching the benchmark at either cost-band endpoint sets overlap true. If required data are absent, this reference recommends reporting affected parity or fragility fields as unavailable rather than inventing values (the YAML does not specify null-output handling); the process decision is NO_GO with reason NON_CONFORMANT.
These cutoffs are governance parameters in the existing rule specification. This publication offers no empirical calibration showing that 14.9% is economically safe or 40% predicts real project failure. Decision signals summarize the profile, not expected investment returns. Recalibration requires a new profile, documented rationale, and review of changed classifications.
The v1.0 “parity compliance gap” bands are a separate legacy reporting scheme. They neither replace this sequence nor prove investment suitability. That summary leaves the normalization basis unspecified and its “11–15%” wording leaves fractional values between 10% and 11% ambiguous. Do not use it as an executable v2.0 approval gate without a separately versioned formula and exhaustive interval definitions.
All values below are synthetic and have no pathway or policy forecast meaning. Let annualized fixed cost at rated production be A = $0.90/kg, variable plus delivery cost $1.50/kg, realized support $0.30/kg, and benchmark $3.00/kg. Then:
C(u) = 0.90/u + 1.50 − 0.30
u* = 0.90 / (3.00 − 1.50 + 0.30) = 0.50
At u₀ = 0.75: C₀ = $2.40/kg and m₀ = 20%
In plain language: with these synthetic assumptions, the plant reaches cost parity at 50% utilization. At 75% utilization it costs $2.40 per kilogram, which is 20% below the $3.00 benchmark.
Under these fixed assumptions, parity requires at least 50% utilization. This design does not meet the stated benchmark below 50% utilization. The distance from the base to the threshold is 25 percentage points, or 33.3% of base utilization. This is a conditional threshold, not proof of collapse across all favorable input combinations.
For the classifier, take electricity bounds $45/$50/$55 per MWh. The legacy scaled band is $2.16–$2.64/kg, so overlap is false. If 2 of 20 predefined stress scenarios fail, f_S = 0.10; q = 0.30/2.40 = 0.125. With complete evidence, the profile yields HOLDS / LOW / PROCEED. At 3 failures it yields MEDIUM / REWORK; at 8 failures it yields HIGH / NO_GO. Raising the upper electricity bound to $62.50/MWh gives C_high = $3.00/kg: touching the benchmark alone triggers HIGH / NO_GO under this profile.
At full support loss the example becomes C(u) = 0.90/u + 1.50, requiring at least 60% utilization. This second boundary demonstrates why policy and utilization must be tested jointly.
Separately, if Product A contains 100 equal-weight parity-feasible grid points and Product B retains 65, the invalidation share is 35%. That describes the selected grid, not a 35% probability of project failure. An application must supply its own parity/collapse map, threshold curve, and fragility representation; this synthetic example is not a conformant project assessment.
Every decision package must preserve the decision question, owner, timing, functional unit, benchmark basis, input provenance, model revision, domain, scenario matrix, constraints, solver settings, and output tables. Include a labeled parity or collapse map before secondary sensitivity figures, a threshold curve, the five comparative metrics, and an explicit directive tied to the failing conditions.
Archive the exact rules and model commit with the assessment, rather than relying on a moving main link. Record the reference version separately from the executable profile. Preserve sufficient numerical precision to reproduce equality and boundary decisions. Report failed calculations and excluded scenarios; do not silently remove them from the denominator.
Review must check dimensional consistency, mass and energy balances, benchmark comparability, policy realization, scenario dependence, and boundary convergence. Verify equality, 15% and 40% failure endpoints, 25% policy dependence, zero denominators, non-conformance, and electricity-band contact before using a new implementation. These checks validate rule execution; they do not validate predictive accuracy.
The sign-off workflow separates computed classifications from reviewer authority. Retain the original system output. Any override must preserve the prior and revised classifications, rule reference, rationale, approver identity, and timestamp. A reviewer note cannot silently change the numerical result.
DG-PFF inherits errors in process models, capital estimates, dispatch assumptions, incumbent benchmarks, and policy interpretation. Transparent thresholds can still be wrong. A narrow stress domain can make a weak project appear durable; an implausibly severe domain can do the reverse. One-at-a-time tests miss interactions, finite grids miss small regions, and static cost comparisons omit timing, liquidity, covenant, and counterparty failure.
Parity is only one decision criterion. A parity-positive design can fail permitting, safety, lifecycle-emissions, resource, infrastructure, demand, or financing requirements. Model these as explicit constraints when within scope; otherwise identify them as unresolved diligence. A common economic unit does not make unrelated technologies or scenario sets directly comparable.
No out-of-sample predictive validation, inter-reviewer reliability study, or calibrated error rate is established by this reference. Published application notes demonstrate use of the protocol; they do not by themselves validate the rules statistically. A credible validation program would preregister cases and stress domains, compare classifications against observed outcomes and simpler TEA baselines, measure false approvals and false rejections, test reviewer agreement, and assess stability under alternative scenario weighting and threshold choices.
Version 2.0 supplies a mathematical specification and explicitly distinguishes conditional boundaries from structural collapse, area invalidation from probability, qualitative persistence from executable fragility, process non-conformance from economic failure, and reference version from rule version. It exposes the legacy electricity-scaling approximation, exact rule endpoints, ambiguous compliance-gap bands, and unspecified scenario and policy-aggregation contracts. Proposed assessment conventions and recommended extensions are distinguished from the source rules. It does not retroactively certify v1.0 applications or change v1.2 executable rules. Future changes to formulas, cutoffs, scenario construction, or precedence require a named revision and an impact record.
Gomez, Jamie (2026). DG-PFF Methodology Reference (Version 2.0). Insight Quantix. Canonical reference.
@techreport{gomez2026_dgpff_methodology_v2,
author = {Jamie Gomez},
title = {DG-PFF Methodology Reference},
institution = {Insight Quantix},
year = {2026},
version = {2.0},
url = {https://insightquantix.com/standards/dg-pff-methodology-reference-v2/}
}
Framework sources: DG-PFF public standard v1.0, DG-TEA validation standard, ruleset v1.2, and sign-off workflow. External foundations are linked at the claims they support in sections 2 and 4. No DOI, external certification, or independent peer-review status is asserted.