Consolidate latest OA question batch

[tomtoto757]

Summary

• consolidate the latest verified Amazon, ByteDance, and Microsoft Coding questions into one batch PR
• add Microsoft Roll the String and Minimum Length Good Subarray with their Aug 27, 2026 sighting date
• retain the previously verified Amazon and ByteDance additions/date refreshes already owned by this PR
• preserve newest-first ordering and refresh date badges with the repository freshness script
• keep the Stage 4 change scoped to README.md

Latest Microsoft update

Company	Question	FastPrep route	Updated
Microsoft	Roll the String	https://www.fastprep.io/problems/microsoft-roll-the-string	Aug 27, 2026
Microsoft	Minimum Length Good Subarray	https://www.fastprep.io/problems/microsoft-minimum-length-good-subarray	Aug 27, 2026

Both production problem documents are attributed to Microsoft and expose OA, INTERN, and HackerRank metadata with no public source images.

• Roll the String: public API assertion passed at 2026-08-28T18:43:50Z with lastSeen=["2026-08-27"]; rendered title, Microsoft attribution, problem content, and the canonical 200×200 Microsoft logo passed at 2026-08-28T18:44:04.913Z; the company-logo API passed at 2026-08-28T18:44:42Z
• Minimum Length Good Subarray: public API assertion passed at 2026-08-28T18:41:16.955Z with lastSeen=["2026-08-27"]; rendered title/content passed at 2026-08-28T18:41:26.413Z; the canonical Microsoft logo passed at 2026-08-28T18:41:18.503Z

Exact revision

• base: 3c98ad57c68670b3cfb001a50c7818d470548683
• head: a67b856163143ffcdb0574b4376295ba55d21622
• prior head before this Microsoft update: 007e54fc9472bb118c134e9ae6ad328502d28037

Verification

• exact prior-head audit: 1,991 rows to 1,993 rows, exactly the two Microsoft additions, zero removals, zero existing-row changes, and every prior row preserves its payload and relative order
• exact-base semantic audit: 1,986 rows to 1,993 rows, exactly seven intended additions, zero removals, and exactly four intended Amazon date updates; all other row payloads and relative order are unchanged
• README audit: 1,993 rows and 1,993 unique matching FastPrep routes; no malformed rows, duplicate routes, bad URL pairs, future dates, date-order violations, company-list duplicates, or company-list order violations
• target audit: exactly one row for each Microsoft question, each with six pipe characters, paired identical FastPrep URLs, Coding format, Microsoft attribution, and Aug 27, 2026
• public route checks: both Microsoft problem routes returned HTTP 200 and rendered their exact titles and Microsoft attribution; both public APIs returned the exact canonical IDs, titles, company, metadata, and latest sighting
• python3 scripts/refresh-freshness.py — two exact-head passes each reported 0 marker changes and 0 reordered rows; README SHA-256 remained 249d8dde7ed79a4df459815b410572e01c758654327e7af4c00a8c64c2fba320
• PYTHONDONTWRITEBYTECODE=1 python3 -m unittest -v scripts/test_sync_practice_formats.py — 10/10 passed
• git diff --check 3c98ad57c68670b3cfb001a50c7818d470548683...a67b856163143ffcdb0574b4376295ba55d21622 — passed

Expected runtime impact: none; this changes README catalog metadata and deterministic row ordering only.

Draft status

A single cache-busted production catalog snapshot (SHA-256 68e894aa48c1fc27f643e83994fc82c0caf456c308f14db957b2c8c95ac78bef; Coding 2,679, system design 209, low-level design 36, project coding 26) was used for the exact base/head comparison.

python3 scripts/sync-practice-formats.py --check fails on both exact base and exact head for the same generated targets: README.md, formats/coding.md, formats/system-design.md, formats/low-level-design.md, and formats/ai-coding.md. Both report managed=271, catalog-missing=148, and unclassified=0; the desired Coding count changes from 1,777 at base to 1,784 at head because this batch cumulatively adds seven Coding rows. The generated Coding page/link remains stale for those additions and the four refreshed Amazon dates.

The flat company list remains 215 unique, case-insensitively sorted entries while its inherited generated label remains 214+. The prior exact-head investigation also demonstrated that a full generated sync is not safely idempotent: Nike, Inc. is split at the comma and re-added as duplicate company tokens on a second pass. That inherited serialization/parsing defect must be fixed before a complete generated-output sync is trustworthy.

This PR remains draft until the company-list parsing defect is fixed and all generated outputs pass --check, or a maintainer explicitly accepts the README-only exception and its known generated-view inconsistencies.

Independent regression review

A fresh independent regression review of exact head a67b856163143ffcdb0574b4376295ba55d21622 against exact base 3c98ad57c68670b3cfb001a50c7818d470548683 is pending. The earlier review was bound to the prior head and does not cover the two Microsoft additions. Keep this PR draft until a separate reviewer records the final-SHA findings, verification, and remaining risks here.