Technical Course Comparison Database: A Reproducible Public-Metadata Sample

Quick answer

A useful technical-course database can compare documented attributes only after every row uses the same definitions, checked date, source rules, and unknown handling. Version 1.1 contains 12 official-source records spanning public tutorials, learning paths, labs, career tracks, a certification-prep path, a Professional Certificate, and an online course.

This database is not a ranking and does not tell a learner which course is best. It helps a buyer identify which questions are already answered by official evidence, which attributes use incompatible units, and which gaps require direct inspection before paying or committing study time. No course was completed, and Unknown is not zero.

Evidence and scope

Evidence profile: E1+E3 – official public-source research plus a reproducible, non-participant metadata dataset
Information checked: September 12, 2026
Dataset: Version 1.1; 12 purposively selected technical-learning offerings across 11 platform systems
Direct evaluation: Blocked; no account, course, lab, project, assessment, checkout, support, or export workflow was used
Original-research branch: Complete for the disclosed public-metadata collection and normalization method only
Main limitation: The sample is not representative of the market and cannot establish teaching quality, learner fit, completion, support, accessibility, credential value, or outcomes.

This release refreshes the public metadata snapshot without turning it into a ranking. Current official pages preserve distinctions among course modules, guided projects, labs, paths, certificates, audit access, and temporary environments. The database records what those pages expose on the checked date, keeps provider units intact, and leaves unsupported attributes as unknown. No course was completed and no product or learner outcome was directly evaluated. See How we test for the evidence labels and testing boundaries used here.

Key facts about version 1.1

The sample contains 12 offering records and 20 official-source records checked on September 12, 2026.
Every offering has the same 19 fields, including source IDs and material unknowns.
The sample covers data and AI, cloud and DevOps, and software-development learning jobs.
Offering types include a course, modular course, learning path, tutorial series, tutorial lab, career track, career path, certification path, and Professional Certificate.
Published workload remains in the provider’s own unit. Hours, weeks, modules, courses, projects, and tutorial minutes are not silently converted into one score.
Access model, assessment model, credential surface, and portable-artifact status are separate fields.
Provider marketing about jobs, salary, confidence, or recognition is not promoted into a database conclusion.
The retained JSON file includes a deterministic record fingerprint so later changes can be detected.

Use the database to narrow the next verification step

A comparison database is most useful before a shortlist becomes a recommendation. Start with one capability, artifact, credential, or environment requirement. Then use the fields to remove offerings that clearly use the wrong format or leave a hard requirement unresolved.

1. Define the unit you are actually comparing

Do not compare a single public tutorial with a multi-course subscription path as though both were the same product. The database labels the offering type and records the platform separately. A course, path, lab series, and Professional Certificate can all be useful, but they solve different planning problems.

2. Separate published structure from learner behavior

Course counts, lab counts, module counts, and provider workload estimates are source facts when the page displays them. They do not prove that the sequence is coherent, that a learner will finish on time, or that every component remains accessible under one account. Those questions require direct evaluation or a more specific source.

3. Audit prerequisites and environment costs

A beginner label is not a complete prerequisite record. Microsoft Learn names basic IT familiarity; DataCamp assumes an earlier track; Terraform’s AWS sequence requires authentication; Kubernetes requires a practice cluster; and some guided cloud projects can create account or resource dependencies. Record those conditions before treating content access as the full cost of practice.

4. Match the practice model to the evidence artifact

Browser exercises, temporary labs, guided cloud projects, problem sets, and final projects produce different evidence. A Dockerfile, notebook, infrastructure configuration, cluster manifest, assessment result, or completion certificate should not be treated as interchangeable. The database records only what the official source establishes and leaves portability unknown when export, licensing, or persistence is not documented.

5. Build a two-record inspection plan

Choose two records that differ on one material dimension, such as public tutorial versus subscription path or video-led path versus guided-project path. Verify the unresolved fields on each exact offering page, then inspect an approved free entry point or account only when direct evaluation is separately authorized. A valid outcome may be to combine two narrow resources rather than choose one platform.

Normalized offering records

Scroll horizontally if all six columns are not visible.

Purposive technical-learning sample, official public metadata checked September 12, 2026
Offering	Type and subject	Published structure or workload	Documented practice	Access or credential boundary	Material unknown
Google Machine Learning Crash Course	Modular course; data/AI	Self-contained modules with module-level minute estimates	Browser programming, interactives, checks, and quizzes	Public pages; no credential established in sampled sources	Completion record and artifact retention
Microsoft Introduction to Cloud Infrastructure	Learning paths; cloud/DevOps	Four-part series; sampled paths contain 3 and 8 modules	Concept modules plus step-by-step Azure projects	Public content; project steps use an Azure account route	Resource cost, grading, and artifact retention
Docker Get Started	Tutorial lab; cloud/DevOps	15-minute introduction plus a getting-started lab	Containers, Dockerfile, image build, lifecycle inspection	Public docs; local Docker or documented labspace	Completion record and artifact ownership
Kubernetes Basics	Tutorial series; cloud/DevOps	Six modules	Deploy, expose, scale, update, and debug an app	Public docs; practice cluster required	Environment cost and retained artifacts
Terraform Get Started – AWS	Tutorial series; cloud/DevOps	Six tutorial-specific minute estimates	Configure, plan, apply, change, destroy, remote state	Public tutorials; AWS authentication for provider steps	Cloud cost, credential handling, and completion tracking
DataCamp Data Scientist in Python	Career track; data/AI	Current page displays 9 courses and 11 track items	Interactive exercises and projects described	Public catalog; full entitlement not tested	Project export and assessment enforcement
Dataquest Data Analyst in Python	Career path; data/AI	27 courses, 19 projects, eight months at five hours weekly	Browser coding feedback and guided projects	Free-enrollment entry; full path boundary untested	Project licensing/export and certificate verification
LinkedIn Learning Become a Data Analyst	Learning path; data/AI	12 courses; 41 hours	Video-led sequence; course-level exercises vary	Trial entry displayed; eligibility untested	Assessment model and retained artifact
Pluralsight Python 3	Learning path; software development	16 courses, 21 labs, 53 hours	Courses plus guided code labs	Library license required; full labs excluded from trial	Lab persistence and export
Pluralsight AWS CLF-C02	Certification path; cloud/DevOps	5 courses, 6 labs, 18 hours, practice exam	Courses, cloud labs, and practice exam	Prepares for separate AWS exam	Plan entitlement, lab persistence, exam outcome
IBM Data Science Professional Certificate	Professional Certificate; data/AI	12 courses; four months at 10 hours weekly	Projects, notebooks, and assessments described	Shareable certificate; account and regional inclusion untested	Export/licensing, credit acceptance, grading
Harvard CS50x on edX	Course; software development	11 weeks, self-paced	Lectures, problem sets, and final project	Audit is temporary and ungraded; verified route is separate	Current audit availability and project persistence

The table is a routing aid. It is a refreshed public-metadata snapshot, not a ranking or a claim about teaching quality. A row is comparable only to the extent that its fields are actually populated by official evidence. A larger course count is not automatically more complete, and a longer workload is not automatically more rigorous.

Coverage and field completeness

Version 1.1 keeps the same 12-record stopping rule at 12 records after covering at least 10 platform systems, at least six offering formats, and both data/AI and cloud/DevOps decisions. The sample contains five data/AI records, five cloud/DevOps records, and two software-development records. These counts describe this sample only.

Scroll horizontally if all three columns are not visible.

Nineteen controlled fields used in every database record, version 1.1
Field group	Fields	Why it is separate
Identity	record ID, offering name, provider, platform	Prevents a provider, platform, path, and course from being collapsed into one entity.
Scope	primary subject, offering type, declared level	Keeps subject and format filters independent from quality judgments.
Learning design	prerequisites, structure, practice, assessment	Separates what a learner needs, what is sequenced, what is practiced, and how work is checked.
Commercial and credential	credential surface, access model	Prevents free access, paid access, completion, and third-party certification from becoming one field.
Evidence artifact	portable-artifact status and summary	Records whether an exportable or learner-owned output is established, unknown, or only a candidate.
Volatility and traceability	published workload, checked date, source IDs, material unknowns	Preserves the provider’s unit, dated provenance, and unresolved questions.

All 12 records contain all 19 keys in version 1.1, but key presence is not factual completeness. Unknown values remain visible, and a documented candidate artifact is not upgraded to a tested portable artifact.

Unknowns and limitations

Sampling: The sample is purposive, not representative. It excludes many marketplaces, universities, bootcamps, vendor academies, and independent courses.
Direct behavior: No account, course, lab, project, assessment, trial, checkout, cancellation, support, mobile, offline, or accessibility flow was tested.
Units: Provider hours, weeks, modules, courses, and projects use different definitions and cannot support a single workload score without additional method work.
Access: A public page does not establish complete free access. Subscription libraries, trials, audit tracks, cloud accounts, and provider resources have different boundaries.
Artifacts: A project or notebook mentioned on a page does not prove that the learner can export, license, publish, or retain it after access ends.
Credentials: Completion certificates, Professional Certificates, Skill IQ, practice exams, and vendor certifications are different credential surfaces.
Outcomes: Marketing statements, testimonials, enrollments, ratings, and career claims are not treated as independent evidence of hiring, salary, promotion, exam success, or job readiness.
Freshness: Course counts, lab counts, dates, workloads, and access terms can change. Every record requires recheck before public use.

Methodology

Method version: 1.0. The unit of analysis is one exact learner-facing offering, not an entire provider. An offering was eligible when an official public page exposed enough current metadata to populate identity, format, structure, practice, access, workload, and unknown fields without using a protected account.

Selection: We used purposive maximum-variation sampling. Records were chosen to cover data/AI, cloud/DevOps, and software-development jobs; public documentation and subscription platforms; and multiple formats. The stopping rule was 12 records after the sample reached at least 10 platform systems and six offering formats. This is not a response sample, so response rate and survey weighting are not applicable.

Collection: Twenty official pages were rechecked on September 12, 2026. Facts were transcribed into controlled fields with source IDs. Provider wording was qualified when it described outcomes, recognition, portability, or completion. Conflicts and missing details were recorded as unknowns rather than resolved by assumption.

Reproducibility: The package includes the versioned JSON database, a field dictionary, and a Python builder/validator. The validator checks 12 unique records, 19 exact keys per record, valid source references, controlled statuses, source and subject coverage, and a deterministic SHA-256 fingerprint over the canonical record array.

Privacy and access: The method uses no participant data, account identifiers, credentials, private evidence, or copyrighted course content. Direct evaluation remains a separate blocked E2 branch. The update owner is the Course Evidence Editorial Team, with the next planned review on October 12, 2026.