A neutral, structural method for auditing any framework, plan, or project: how coherent it is, and how resilient it is to its own stated goal. It generalizes the Resilience Ledger (which measured resilience from a neutral center) and the groundedness scoring from the Agentic AI Governance glossary, raising the same move one level: from measuring terms to measuring whole frameworks.
Most frameworks, plans, policies, and strategies are part structure and part slop. Slop is the share that sounds right but does no work: vague labels, claims that cannot be tested, words that only mean something inside their own story. Slop hides weakness, and weakness is where things break. This method measures the ratio, neutrally, so a project can see which parts carry load and which will fail under their own goal.
A framework is sound to the degree its parts do real, testable, grounded work toward its stated goal, and still hold up after you strip its preferred vocabulary. Measure from a neutral center, function rather than narrative, the way a plumb line measures true vertical. The narrative is not the thing; what each part does is the thing.
Any framework expressed as two inputs:
- The goal. One sentence stating what the framework is for.
- The parts. Its claims, principles, steps, controls, or components, one per line.
Grouped into two questions. Each part is scored on each dimension; the scores roll up to a profile of the whole.
- Function clarity. Is each part defined by what it does, or only named? A part with no action is a label, not a mechanism.
- Internal coherence. Do the parts agree, or do they contradict and repeat each other?
- Goal alignment. Does each part demonstrably serve the stated goal? Orphan parts and quiet goal drift are slop.
- Framing independence. Does the part survive a plain restatement, or collapse the moment its buzzwords are removed? This is the sovereign-zero test, and it is the sharpest slop detector of the set.
- Groundedness. Is the claim backed by independent evidence, a number, or a source, or is it simply asserted?
- Falsifiability. Could the claim be tested or proven wrong, or is it true by construction?
- Failure-awareness. Does it say what happens when the part breaks, with fallback, reversal, or recovery, or does it assume the part never fails?
- Agency clarity. Is it clear who or what acts, and who is accountable when it does not?
Each part scores each dimension from 0 to 100. Average across the parts to get a per-dimension score, and combine those for an overall structural-integrity score. The slop share is the percentage of parts that fail function clarity, falsifiability, or groundedness. Low bars are not opinions. They mark, precisely, where a framework cannot defend itself.
This method is itself a framework, so it must pass its own audit, or it is slop too. Its goal: measure a framework's coherence and resilience neutrally. Its parts are the eight dimensions. Each dimension is defined by what it does (function clarity), can be checked against text (falsifiability), serves the goal of neutral measurement (goal alignment), and survives plain restatement (framing independence). Any dimension that cannot pass its own test is cut. That self-reference is both the honesty check and the proof the method works.
The audit is a red team. Run the profile, then attack the lowest bars first. The parts with the most flags are the weakest points; the lowest dimension is the framework's structural blind spot. Red-teaming a plan is not finding a clever exploit first. It is finding the unfalsifiable, ungrounded, goal-orphaned parts and pressing there until they give.
Audit early, and audit again as the project grows. Parts drift from the goal, vague language creeps in, and failure handling gets assumed rather than stated. Re-running the audit shows drift as it happens. A falling goal-alignment score is scope creep. Rising buzzword density is thinking going soft. A flat failure-awareness score means the project is one surprise away from breaking. The method keeps a project honest with itself over time. This is continuity of coherence, made measurable.
The companion tool scores by word-level heuristics. It flags likely weakness; it does not understand meaning. Treat its output as a thinking aid that points where to look, not a verdict. A part can pass every heuristic and still be wrong, and a blunt, plain sentence can be perfectly sound. The value is in forcing each part to declare what it does, whether it can be tested, and whether it serves the goal. The judgment stays human.