CPE-20-01 deep review — September 29, 2026

The study guide maps all 26 numbered objectives, grouped 7/8/6/5 under the four blocks. Current CPE-20-01 scope remains the July 22, 2025 syllabus: weights 28/28/24/20 and item counts 9/8/7/6. This same-context AI review retains human review as pending.

Source evidence

The complete CPE page, certification catalog and exam policies were read through the public web reader after direct timeouts. The objective monitor failed; manual comparison supports retaining the snapshots without claiming a successful automated hash check. Exam delivery remains TestNow, English, 30 questions, 45 minutes plus approximately five for NDA/tutorial, and a 70% cumulative-point threshold. The stated 120-point maximum does not establish equal points per item. Catalog lifetime/version retention and the CLE/CPE seven-day failed-retake rule are distinct from the associate/professional policy's 15 days and Pearson delivery. Current booking UI was not inspected.

There are 29 registered sources, with direct health outcomes {'error': 4, 'ok': 23, 'blocked': 2}. HTTP success is not complete-source or rendered-course review. The operation record describes every reading boundary. Selected Microsoft references and C++17 draft clauses support the repairs; cppreference was read only as a third-party index. Microsoft reference wording on remainder signs, string sentinel constness, main parameters and ignore's count limit is qualified using the pinned C++17 rules. The public exam outline does not specify a language switch; C++17 is an author practice choice.

Teaching and execution

All original guide text and its 40 answers were reviewed; the answers were retained and improved. The Product example's assignment now appears inside a complete main. Hosted-main arguments, conversion timing, grouping versus sequencing, whole-line parsing versus numeric prefixes, stream recovery, both named pointer casts, vector reserve/resize and invalidation, allocation pairing, and the C++17 string sentinel are explained with original traces. Three scenario sketches now have complete programs and explicit contracts.

Ten public C++ blocks compiled and ran, with three fragments supplied clearly documented calling/function contexts. Compiler Explorer GCC 14.2 and Clang 19.1 each passed 67 workbook checks and 60 scenario boundary checks. The factorial fragment also passed all 13 inputs from 0 through 12; the parameter fragment passed four caller-state checks. Complete-program stdout, exit status and fragment allocation results were verified. Thirteen final build/run receipts use -std=c++17 -Wall -Wextra -pedantic. A first Clang pass flagged an intentionally unused changed parameter; the final workbook returns that value and verifies both it and unchanged caller state. Remote rate limiting during rerun was treated as an execution delay, not a code result.

Inventory uses bounded integer cents, rejects duplicate catalogs and preserves the total on failure. Tests cover empty/first/last/missing IDs, quantity and price limits, malformed lines and extra suffixes. Sensor tests cover fixed arrays, vector ownership, empty/null input, optional outputs, negative/range/count boundaries and reallocation without using stale pointers. Menu tests cover bounded sum/product, factorial base/endpoints, EOF/quit, missing operands, unknown choices and suffixes. The scenario harness preserves the published function bodies, renames their entry points and redirects in-memory streams. These are selected checks, not all platforms or inputs. No undefined behavior, sanitizer run or allocation failure was deliberately executed.

No local C++ toolchain was installed or used. Compiler Explorer is a remote public sandbox; only original public code was submitted, with no account or saved-code shortener. The catalog's live-partial value describes this remote code execution, not completed infrastructure labs. Eight broader learner activities remain proposed.

Learning resources and remaining limits

Edube's full public landing lists 42 hours and modules 0–4 but includes inconsistent C/CLE labels and obsolete standards prose. Those errors do not replace the canonical CPE scope. Pluralsight's full public listing contains 13 courses/44 hours; the first two total 319 + 348 = 667 minutes (11h07m). No paid lessons were entered or watched. Selected Microsoft calculator sections were read; its sample input loop is not presented as a complete failed-input/EOF solution. OReilly and Udemy returned HTTP403, so old durations remain qualified. Cisco and YouTube yielded shells/title text; no transcript or playback was reviewed. Author planning estimates are labeled.

The operation record contains objective mappings, source receipts, exact public code/hashes and remote execution evidence. Repository, full unit, catalog and strict generated-site gates are recorded after completion. The outcome is reviewed-with-blockers for inaccessible/paid contents, unresolved vendor copy errors, current account/booking verification and human review. No account, purchase, enrollment, exam attempt, external message or runtime/service change occurred.