Collect
Interfaces, requirements, tests, decisions, changes, and defects.
Methodology · evidence to decision
Evidence you can inspect.
Trace each finding to evidence and each estimate to its assumptions. Challenge the recommendation before deciding.
Work directly withDr Zak Ouzzif
The complete chain
Five steps. No black box.
Measurement and judgment stay separate so your team can challenge the estimate without losing the evidence.
Connect
Show where dependencies cross system and ownership boundaries.
Quantify
Use your change history to estimate excess effort and delay.
Stress-test
Vary assumptions and expose what moves the result.
Decide
Fund, measure, defer, or leave alone.
Use the records you already have.
Inputs
Use the records you already have.
Source code sharpens the map when available. It is not the only place system design evidence lives.
Interface records
Boundaries, contracts, and coupling edges.
Core inputRequirements
Allocation and traceability across modules.
Core inputChange history
Effort weighting for the exposure model.
PreferredTest evidence
Where failures and verification effort concentrate.
PreferredDecision records
Known compromises, waivers, and deferred work.
PreferredSource repository
Confirms whether documented boundaries match reality.
OptionalThe evidence behind the method.
Published research
The evidence behind the method.
Two studies, two different questions.
Document evaluation
How well did the classifier perform?
Aerospace documentsClassification test
- Documents
- 141
- F1 score
- 0.82
- Cohen’s κ
- 0.84
Scores describe classification performance and agreement, not cost-model accuracy.
Evaluation: §4.1.1 / Table 4.1, pp. 39–40 ↗Separate practitioner feedback
What did practitioners report?
PractitionersExpert survey
- Survey participants
- 35
The 35-person survey is separate from the document population behind the classifier scores.
Survey: §4.3, pp. 47–56 ↗Financial decisions need case-specific evidence. Classification scores and practitioner feedback support different parts of the assessment.
The published evaluation covers technical-debt classification in aerospace test-and-evaluation documents; organizational and AI-adoption applications require separate validation.
Scope of the published research
Ouzzif, Zakaria (2026), A Technical Debt Management Framework for Aerospace Systems Engineering: An AI-Driven Approach to Test and Evaluation Documentation Analysis, PhD dissertation, Worcester Polytechnic Institute, May 2026.
Page 62 defines the study scope and the evidence needed for broader application.
Constructed exposure model
Change the assumptions. Watch the estimate move.
This illustration starts with 1,200 engineering days at a $650 loaded day rate. Real assessments use your records and publish a range—not a false point estimate.
Illustrative modelNot client data
Effort versus a low-coupling baseline
Share of annual change effort
Excess days497
Annual exposure$323k
Schedule5.5 wk
$262kSensitivity band$363k
The coupling multiplier dominates this example. Better module-level change data narrows the band.
Concentration becomes a ranked plan.
What the model produces
Concentration becomes a ranked plan.
Constructed figures demonstrate the format only. They are not a client result, benchmark, or promise.
01 · Locate
AChange effort concentration
71%of effort
Modules18
Cluster4 modules
Interfaces62%
02 · Rank
BReturn per engineering day
- 01Re-partition A–B boundaryHighest exposure removed per dayFund
- 02Extract shared stateCredible intervention and paybackFund
- 03Instrument supplier changesEvidence needs narrowing firstMeasure
- 04Rewrite low-change internalsWeak return despite visible debtLeave
Keep confidence separate from coverage.
Expected
- Interfaces
- 91
- Change history
- 86
- Source checks
- 64
- Review records
- 82
- Effort attribution
- 48
- Interviews
- 78
Limits
The boundary is part of the method.
Open the caveat that matters. Each one is raised during scoping and shown again in the report.
Classification is not cost accuracy
κ 0.84 and F1 0.82 describe repeatability. The cost range depends on the economic assumptions tested separately.
Evidence quality controls confidence
Missing or outdated records widen the range. Source access can confirm whether documented boundaries match the implementation.
Causal evidence is not a controlled experiment
The model compares high- and low-coupling work inside the same organization. It supports a causal argument; it does not claim experimental proof.
The result is point-in-time decision support
System design continues to change. The assessment is not certification evidence and does not replace an auditor, DER, or safety assessor.
Control the evidence before it moves.
Material handling
Control the evidence before it moves.
Transfer, access, retention, and deletion are agreed in writing before the assessment begins.
Minimum necessary
Limit material and people to the agreed system boundary.
Work in place
Scope work inside your environment where transfer is impractical.
Never
No model training, benchmarking, or reuse in another engagement.
Confirmed
Destroy material on the agreed date and confirm it in writing.
Dr Zak Ouzzif
About the practitioner
Dr Zak Ouzzif
Principal architect · ZOYA Solutions
Experience with critical defense systems informs how I examine dependencies, test changes, and plan for continuity. I apply that judgment to business improvement and practical AI.
My doctoral research in systems engineering produced the Technical Debt Management Framework (TDMF), built to bridge the gap between academic rigor and executive decision-making. My approach translates deep systems engineering theory into pragmatic, executable risk assessments.
I founded ZOYA Solutions on a simple structure: advisory and delivery are separate scopes with an explicit investment decision between them. The diagnostic and the business case stand on their own — you can act on the roadmap yourself, hand it to another partner, or ask us to implement it.
When we do implement, it is scoped and quoted separately, after the evidence is already yours.
Have a decision due? Check fit first.
Bring the decision and deadline. The 30-minute fit check is free and needs no preparation.
Request a free workflow review