Forward Deployed Engineering assessment market landscape
As of: August 31, 2026
Status: dated product research; non-exhaustive; not a vendor rating
Executive finding
Forward Deployed Engineering assessment is no longer an empty category. Public offerings already cover career diagnostics, job-description interpretation, multiple-choice certification, structured training, mastery claims, and verifiable credentials. LockedIn should therefore avoid a category-first claim and compete on a harder proposition:
A continuously versioned, independently governed evidence framework that connects demonstrated work to explicit capabilities, preserves uncertainty and contradiction, separates training from qualification, and makes every authorized conclusion traceable.
That position is distinctive only if the operating evidence supports it. The v0.1 repository is a proposed framework and synthetic foundation, not yet the operating lab.
Observed segments
| Segment | Representative public offering | What the segment makes easy | What remains open for LockedIn to prove |
|---|---|---|---|
| Career diagnostic | FDE Compass | Fast self-assessment, role-shape clarity, job-description analysis, learning plan | Whether a governed field-evidence record produces reliable, fair, useful information beyond self-report |
| Knowledge certification | GSDC CFDE | Accessible syllabus, known exam format, training and credential bundle | Whether actual production contribution and ambiguity handling can be assessed consistently without training-provider conflict |
| Training plus mastery credential | Good Combinator FDE | Deep curriculum, track choice, evidence/mastery language, verifiable credential | Whether an independent method can separate learning completion, demonstrated capability, evidence strength, and readiness |
| Simulated work trial — publicly visible, creator describes as paused/inactive | FDE Arena and creator retrospective | Customer-workspace simulation, transcript-backed report card, and evidence of substantial candidate completion burden | How a controlled demonstration can be learnable, accessible, proportionate, and independently validated without turning a broken or abandoned attempt into low capability |
| Adjacent human/agent task research | METR RE-Bench | Open-ended environments, human baselines, protected solutions, reproducible research artifacts | How to measure a socio-technical human role across contexts without collapsing capability into timed task performance |
| Benchmark research platform | Artificial Analysis | Methods, live data, analysis, archives, and decision tools in one reference product | Whether LockedIn can build comparable research throughput, version discipline, correction practice, and trust on the human side |
Strategic choices
1. Make the unit of truth smaller than the person
The primary record should be an evidence-to-claim-to-capability relationship with provenance, ownership, maturity, verification, recency, contradictions, and review. A person-level profile is a disclosure-filtered projection. This avoids implying that one number is an intrinsic property of a human.
2. Publish a profile, not a human leaderboard
The public product should prioritize domain evidence states, coverage, confidence, limitations, verification class, and version. No named-person ordering, global human intelligence number, or cost-per-person axis should exist. Employer-specific role fit belongs in a private, purpose-bound decision aid with legal and fairness controls, not the public benchmark.
3. Separate research lenses from qualification
Evidence Strength and Recency can be descriptive lenses. Time to Governed Production can be studied at project/cohort level once start/gate definitions, context, censoring, and attribution are defensible. None should silently become a personal ranking feature.
4. Separate training from measurement authority
Training providers may contribute public feedback, but payments, course completion, sponsorship, or advisory access must not change a benchmark release, verification result, appeal, or publication outcome. If LockedIn later offers education, governance and decision personnel need structural separation.
5. Earn research-lab authority in public
Authority comes from operating artifacts, not the phrase “research lab”:
- preregistered questions and analysis plans;
- versioned instruments, rubrics, and evaluator/grader cards;
- representative recruitment and explicit exclusions;
- inter-rater, repeated-run, construct, criterion, fairness, and accessibility evidence;
- null results, failed graders, incident disclosures, and corrections;
- citeable releases with frozen source/data/code availability statements;
- independent methods review, funding/conflict disclosure, and author contribution statements.
6. Expand roles only through validated shared infrastructure
The evidence, consent, provenance, governance, research, and publication systems can become a shared capability-research platform. Role ontologies, tasks, raters, validity evidence, and release decisions must remain role-specific. Expansion to AI engineer, solutions architect, evaluation engineer, or AI product leader should begin only after the FDE program demonstrates repeatable method operations.
Research agenda needed to support market-facing claims
| Claim the market may want | Minimum research before making it | v0.1 posture |
|---|---|---|
| “This evidence supports capability X” | content validity, scoring/rubric evidence, reviewer reliability, known-groups/contradiction analysis | Hypothesis only |
| “This person is ready for role Y” | role study, criterion definition, predictive/decision study, fairness, accommodations, base-rate and transportability analysis | Prohibited |
| “This level is independently verified” | approved verification protocol, independent reviewer qualification/conflict controls, reproducible audit record, appeals | Design target only |
| “This score is comparable over time” | calibrated scale, linking/equating study, drift controls, release stability | Prohibited across unlinked versions |
| “This predicts production success” | preregistered longitudinal outcomes, contextual confound controls, external validation | Unsubstantiated |
| “This is a standard/certification” | operating governance, consensus process, licensed normative sources, scheme rules, impartiality and accreditation strategy | Not claimed |
12-month research operating target
- Stand up the three-person nucleus: Head of Measurement and Validation, FDE Research Director, and Evaluation Research Engineer.
- Establish an external advisory panel with practicing FDEs, industrial-organizational psychology/psychometrics, security, accessibility, privacy/employment law, and affected participants.
- Run qualitative role research across company sizes, industries, and FDE role shapes; publish sampling and disagreement.
- Freeze a narrower pilot ontology and task/evidence rubric; preregister reliability, learnability, and fairness analyses.
- Conduct synthetic and expert-fixture grader validation before any participant evidence.
- Run a controlled consented pilot with no public score and a functioning correction/appeal path.
- Publish the pilot method, exclusions, failures, and uncertainty before deciding whether numeric or credential claims are supportable.
Watchlist
- FDE offerings change quickly; review this landscape quarterly.
- Track adjacent work-trial, skills-credential, professional-certification, and human-plus-AI evaluation products—not only products named “FDE.”
- Record acquisitions, methodology changes, pricing/training conflicts, validity publications, accreditations, and data-use terms as dated facts.
- Do not reproduce competitor content, scoring logic, visual design, or protected taxonomies.
The underlying sources and review rules are recorded in the source ledger.