Source and citation quality¶
Best-effort principle¶
All repository research, validation, and discrepancy handling is best effort. That means doing the most useful bounded investigation available with the authorized evidence and tools, not claiming that a search is exhaustive or that missing evidence proves a feature does not exist. Record where and when the search was performed, the source classes checked, the strongest supported answer, conflicting or unavailable evidence, confidence, and the specific work that could improve the answer later.
Best effort is not permission to lower the evidence standard. When certainty is unavailable, prefer a useful provisional explanation and a safe validation path over either unsupported certainty or an unexplained omission. Keep the gap visible, preserve failed and contradictory findings, and replace the provisional answer when stronger evidence becomes available.
Living-guide default¶
Research findings should improve the guide during the same bounded repair wave whenever the available evidence supports a useful statement with reasonable confidence. Do not leave a verified concern in a candidate queue merely because the evidence is incomplete or may change. Add a dated note near the affected material, cite the strongest exact sources, distinguish exam scope from adjacent learning, apply the appropriate confidence label, and state what remains unknown and how to recheck it. Promote evaluated sources to the catalog in that same change.
Use review-required only when accepting the source or wording the guidance requires a material
decision that cannot yet be bounded safely. When an official page is unavailable, continue down
the authority hierarchy and document the strongest provisional model that useful vendor-owned,
upstream, named-expert, or independent evidence supports. The official-source check may remain
blocked even after the guide gains useful provisional content. Living guides are expected to
replace or relabel that content when stronger evidence appears; a blocker is a validation state,
not a reason to omit the best answer currently available.
Authority hierarchy¶
| Tier | Source | Permitted use |
|---|---|---|
| 1 | Official exam blueprint | Exam identity, domains, objectives, weighting, and status |
| 2 | Official product documentation | Current technical behavior and configuration |
| 3 | Official training | Vendor explanations, labs, and approved preparation |
| 4 | Structured third-party technical documentation or training | Learning sequence and additional explanation; technical behavior only under the corroborated-use rules below |
| 5 | Named expert or maintained repository | Practical examples, specialist explanation, and reproducible implementation evidence |
| 6 | Independent assessment | Gap detection, never canonical dispute resolution |
The hierarchy expresses authority, not a ban on diversity. A lower-tier source can add useful technical depth when it is attributable, current enough for the claim, legally usable, free of exam-dump signals, and more specific than the available vendor material. It never overrides a Tier 1 exam contract. Prefer direct tests, upstream standards, maintained project documentation, or two genuinely independent accounts over an unattributed blog summary.
Claim confidence labels¶
- Confirmed: supported by the current official blueprint for exam scope or current first-party product documentation for behavior.
- Corroborated — verify current: supported consistently by a maintained upstream or named non-vendor source and another independent source or reproducible observation, while first-party confirmation is missing or incomplete.
- Unconfirmed — validation needed: a useful hypothesis or implementation report with an explicit evidence gap and a safe validation path. Do not present it as required exam behavior, production guidance, or a required lab step.
Put the label next to the affected passage; a note in a distant source list is not enough. When stronger evidence appears, replace or relabel the claim and retain the earlier discrepancy in the review record when it affected an audit or blocker.
Citation requirements¶
- Place a descriptive link near the claim it supports.
- Cite the final authoritative page, not a search-result page.
- Do not cite a course landing page as evidence of product behavior.
- Do not let a model invent URLs. New generated references must come from the registered source set; seed-guide links should be registered as they are reviewed.
- Record publisher, authority class, last check, and supported objectives.
- Mark a dependent page for review when a canonical source changes.
- Treat publication dates and update dates as different fields.
Volatile claims¶
Pricing, discounts, exam duration, delivery vendors, languages, product UI, model lists, quotas, credits, retention, preview status, feature availability, and command syntax must be labeled VERIFY CURRENT unless directly synchronized from a current official source.
Blockers and source discrepancies¶
When first-party documentation is missing, contradictory, or too shallow to teach a published objective, apply the best-effort principle before declaring the topic unresearchable. Check, in order: current vendor documentation; vendor-owned source code, examples, issue trackers, and training; maintained upstream projects; named experts; and independent implementation reports. Search results are leads, not evidence.
Record promising exact URLs in data/source-candidates.json with the proposition each
could support, its age/ownership, conflicts, and what still needs validation. Two sources
repeating the same claim do not make it official, particularly when one copied the other.
Do not promote a candidate merely to remove a blocker.
A guide may summarize a provisional model only when it clearly separates:
- what the blueprint requires;
- what current first-party evidence confirms;
- what supplementary sources consistently suggest;
- what remains unknown, version-sensitive, or contradictory; and
- the safe test and evidence needed to validate the hypothesis.
Mark provisional material VERIFY CURRENT, avoid production-ready wording, and do not turn unverified commands into a required lab. Prefer a disposable, authorized validation matrix that records versions, expected/observed behavior, logs, failure cases, and cleanup. Keep the review blocked until accepted evidence or reproducible validation meets the normal source standard. Recheck the candidate set during later freshness passes so better first-party guidance can replace provisional material without erasing the earlier gap.
Third-party evaluation¶
Evaluate an individual resource rather than an entire marketplace. A non-vendor source may support a technical passage after this evaluation, but cannot establish exam identity, scope, weights, logistics, lifecycle, or scoring. Record:
- author or publisher;
- format and access model;
- publication and update dates;
- objectives covered;
- strengths and known gaps;
- whether explanations cite first-party evidence;
- whether the source contains independent practice or questionable exam claims.
Claims such as “actual questions,” “real exam dump,” or guaranteed passing are exclusion signals.
Practice-assessment boundary¶
Practice assessments are Tier 6 sources because of how they are used, even when the exam sponsor publishes them. They may help identify weak objectives, exercise time management, and provide explanations or first-party references. They cannot establish exam scope, technical truth, scoring rules, or the contents of a live exam. Prefer the free Microsoft assessment as the first baseline when one is available, then use a paid provider only when another question style or explanation set would add value.
Link an exact exam-specific product only after checking its publisher, visible objective map, question count or format when published, update signals, access model, and exam-integrity language. Record contradictory or stale public copy as a caveat. Estimate consumption time for an initial timed attempt, targeted practice, and explanation review rather than implying that every learner should memorize or exhaust the bank.
When a source declares a permissive license but its linked license file is missing, record the declaration and missing notice in THIRD-PARTY-NOTICES.md, attribute the author and source, prefer independently written synthesis, and add the exact notice if it becomes available. Do not infer that publicly readable material is in the public domain.
Seed training sources¶
The initial catalog may include Microsoft Learn, GitHub Docs, O'Reilly, Pluralsight, Whizlabs, Udemy, John Savill's Technical Training, and Timothy Warner's public repositories. Inclusion is not endorsement; each resource receives its own metadata and coverage review.