August Studying Schedule

Aug 11 → Aug 20, pointed at the Google onsite Thu Aug 20. Organised by topic, not by date — work down the list in priority order and aim for 2–3 problems per topic. Weekdays 1–2 problems, weekends 2–3. Systems design is a 12-hour block weighted to the weekend. Progress is saved in this browser.

Priority 1 — most askedPriority 2Priority 3 — only if timeSys DesignInterview

Progress

0 / 0

Tue Aug 112.5 h · 1–2 problemsSD 1h

Wed Aug 123 h · 1–2 problemsSD 1.5h

Thu Aug 13Mock 4–5pm PTSD 1h prep

Fri Aug 142.5 h · 1–2 problemsMock debrief

Sat Aug 155.5 h · 2–3 problemsSD 3.5h

Sun Aug 165.5 h · 2–3 problemsSD 3.5h

Mon Aug 172 h · 1–2 problems

Tue Aug 182 h · 1–2 problems

Wed Aug 192.5 h · light, no new patternsSD 1.5h review

Thu Aug 20ONSITE · warm-up onlyGoogle

How to use this — read first

Topic order is priority order. It's weighted by what 8 recent 1point3acres threads actually show, not by Google's general reputation. Full evidence in google-prep.html.
Each topic has MUST DO and FOLLOW-UPS. Do all the MUST DOs before any follow-up. Then take 1–2 follow-ups from the topics that felt shakiest. Target 2–3 total per priority-1 topic.
The follow-ups matter more than usual here. Almost every Google report describes one problem that escalates, not a fresh problem per follow-up. The follow-up is what's being scored.
DP is deliberately down to 2 problems from 5. Zero DP questions appeared across all 8 threads. Kept at low volume rather than cut, since n=8 is a small sample.
Sys design blocks are 1 + 1.5 + 1 + 3.5 + 3.5 + 1.5 = 12h. Focus is still open. TODO · confirm sys design focus
Log every problem in learning_journal.html the same day. Retry dates matter more than volume.

Intervals + heap scheduling

seen 3× · most askedP1▾

Why first: the single most common shape in the sample — two separate onsites and a phone screen. The recurring frame is "assign overlapping intervals to the fewest resources," and Google asks it with an OOD twist (design a Car class) and a capacity twist (only K cars, else IMPOSSIBLE). Suggested days: Aug 11–12.

Must do

#253 Meeting Rooms II Min-heap of end times. The base pattern for every question in this topic.
Car rental drill — the actual Google question Orders (id, pickup, return) → fewest cars, return orders per car, and design a Car class. See python-class-cheatsheet.html for the class mechanics.
#1094 Car Pooling Sweep line with a capacity ceiling. Bridges #253 into the K-resource variant.

Follow-ups

"Output every window with exactly X concurrent" drill Asked at the Aug 5 onsite. No clean LeetCode equivalent. The difficulty is start/end tie-ordering on the same timestamp, then merging adjacent qualifying windows into maximal intervals.
K-car capacity + IMPOSSIBLE Same drill, but only K cars exist. n ran to 200,000 in the real question.
#986 Interval List Intersections Two pointers, overlap is max(starts) → min(ends). Quick confidence rep.
#435 Non-overlapping Intervalsin journal Greedy by end time. Skim only.

Union find & connected components

seen 2×P1▾

The tell to internalise: when a problem says things get added incrementally, that is a union find signal. A Nov 2025 candidate failed this round by trying to patch their BFS instead of switching structures. Suggested day: Aug 15.
Added Aug 13 — MST sub-track: the "Build bridges between islands" question (flood fill to label islands → straight-line water gaps as edges → Kruskal) makes this a two-skill topic: union find as a connectivity check, and union find as the engine inside minimum spanning tree. #1584 and #1135 cover the second. Solution saved in build_bridges.py.

Must do

#305 Number of Islands II The exact onsite follow-up. Land added one cell at a time, report island count after each. Write find with path compression from scratch.
#684 Redundant Connection The union that fails is the answer. Cleanest possible union-find template.
#721 Accounts Merge Union find keyed by email. A real Google question, and the "map arbitrary keys to indices" step is the part people fumble.
#1584 Min Cost to Connect All Points The purest MST on LeetCode, and the direct rehearsal for the Build bridges between islands question (Aug 13). Same shape minus the flood fill. Kruskal = sort edges, skip any union that fails. Use union by size, not bare parent[j] = i.

Follow-ups

2D clustering drill Points + radius r, connected if distance ≤ r, transitive → count clusters. Write the brute-force O(n²) version first — the real interviewer insisted on that before optimising. Then discuss spatial grid / KD-tree for neighbour lookup.
#547 Number of Provinces Adjacency-matrix version. Fast rep if the others clicked. Attempted Aug 13 — two bugs: rebound parent to an int inside the counting loop, and used parent[j] = i instead of parent[find(j)] = find(i), which silently undercounts. Invariant: only ever write parent[x] when x is a root. Retry from blank.
#1135 Connecting Cities With Minimum Cost Premium. Textbook Kruskal, and it includes the "graph can't be fully connected → return -1" case. Do it after #1584 as the confirmation rep.

Heaps & order statistics

seen 1× · full gapP1▾

Nothing in your journal covers this. #295 appeared with two follow-ups at an onsite and the candidate ran out of time on the second. You did #215 and #621 in July, but two-heaps is a different structure. Suggested day: Aug 16.

Must do

#295 Find Median from Data Stream Max-heap for the lower half, min-heap for the upper, rebalanced so sizes differ by ≤ 1. Python has no max-heap, so negate. Be ready to dry-run and state complexity unprompted.

Follow-ups

Kth smallest instead of median The exact follow-up that ran the candidate out of time. Same two-heap idea with the split at k rather than the midpoint.
Small value range variant If values are bounded, drop the heaps for a counting array. State the trade-off out loud.
#480 Sliding Window Median Two heaps plus lazy deletion. Hard, only if the rest is solid.

Graphs — topological sort, cycles, closure

seen 1×P2▾

The Aug 5 onsite asked transitive closure (which players' ranks are determinable), which is adjacent to but not the same as topological sort. You already have #207 and #743 logged. Suggested day: Aug 14.

Must do

Player rankings drill N players, M "A beat B" results. A rank is determinable if transitive implications settle that player against every other. Floyd–Warshall, or DFS/BFS from each node — say the DFS option first, it's the more natural answer.
#210 Course Schedule II You have #207. Same Kahn's, but return the order; shorter than n means a cycle.

Follow-ups

#269 Alien Dictionary Google staple. The trap is that a prefix word appearing after its extension is invalid input, not an empty answer.
#787 Cheapest Flights Within K Stops Sitting in your journal's Up Next. Plain Dijkstra is wrong because K-stops breaks the greedy invariant — be able to say why, not just that it does. Bellman-Ford, or BFS over (node, stops).

Hash map counting + greedy grouping

seen 1× · full gapP2▾

Cheap to learn, and the Aug 11 candidate got "data structure may not have been optimal enough" as feedback here despite solving both parts. Clean structure is scored separately from correctness. Suggested day: Aug 12 or 17.

Must do

Frequency-partition drill Same value → same partition, no two partitions may have equal size. Collapses to "are all frequencies pairwise distinct." Counter + set, and the whole point is recognising the collapse fast.
#846 Hand of Straights The follow-up: partitions must be consecutive runs. Greedy from the smallest remaining value, consume a run, repeat.

Backtracking

seen 1× · your oldest flagged gapP2▾

#78 and #46 are the same skeleton — subsets passes a start index down, permutations tracks a used set. Do them first or you'll learn the pattern three times. Write the skeleton from memory before looking at any problem. Suggested day: Aug 11.

Must do

#78 Subsets Choose / explore / unchoose, with a start index. The base template.
#47 Permutations II What Google actually asked, so do this instead of #46. Sort first, then skip a value equal to its predecessor when the predecessor is unused at this depth.

Follow-ups

#39 Combination Sum Reuse without advancing the index — the only twist.
#131 Palindrome Partitioning Try every palindromic prefix, recurse on the rest.

Stack, parsing & state machines

seen 2×P2▾

#394 came up in two separate rounds and you already have it logged, so this is a cheap re-fire rather than new learning. The log-parsing question is the same muscle in unfamiliar clothing. Suggested day: Aug 17.

Must do

#394 Decode Stringin journal Re-derive cold, don't reread your notes. Stack of [count, partial] per level.
Quoted-log-split drill Split a log on whitespace, but never inside quotes. Character state machine tracking in-quote. Raise escaped quotes unprompted.

Follow-ups

#394 memory follow-up Decoded string is enormous. Generate lazily instead of materialising.
Log doesn't fit on one machine The real difficulty is that a chunk boundary can land inside a quoted section, so each chunk must carry its in-quote state across.

Dynamic programming

seen 0× · reduced from 5 problemsP3▾

Zero DP questions across all 8 threads. Cut from 5 problems to 2. Kept rather than dropped because n=8 is a small self-selected sample and DP is genuinely part of Google's historical range. Say the recurrence in words before writing any code. Suggested day: Aug 16.

Must do

#198 House Robber Smallest real recurrence — take or skip. Start here even though it's easy.
#322 Coin Change Unbounded, min instead of max. The unreachable sentinel is where people trip.

Follow-ups — only if every P1 topic is complete

#300 Longest Increasing Subsequence O(n²) first and say it out loud. Patience sorting only if that's solid.
#139 Word Break Honestly backtracking + memo, which ties back to that topic.

Trees

seen 0× · you're already strongP3▾

No tree questions in the sample, and you have #98, #102, #199 and #236 logged. One problem, chosen because it's the one classic you don't have. Suggested day: Aug 18.

Must do

#124 Binary Tree Maximum Path Sum The split between what you return to the parent (one branch) and what you record globally (both branches). Naming those two things differently in your code is most of the battle.

Follow-ups

#297 Serialize and Deserialize Binary Tree Preorder with explicit null markers. Good one to narrate.
#437 Path Sum III Prefix sum on a tree — a real Google shape even though it didn't surface in this sample.

Binary search & tries

seen 1× / 0×P3▾

Binary search appeared once, on strings rather than numbers. Tries didn't appear at all and you already have #208 and #212 — skip unless everything above is done. Suggested day: Aug 19 if at all.

Only if time

Binary search over a sorted string array Same template you used for #981, with string comparison. 20 minutes, mostly to prove the template transfers.
#211 Design Add and Search Words You have #208 and #212, so this is only the wildcard delta — at a ., recurse into every child.

Systems design — 12 hours

Round 2 · full 60 minSD▾

Mock is Thu Aug 13, 4–5pm PT. Blocks: Aug 11 (1h) · Aug 12 (1.5h) · Aug 13 (1h prep) · Aug 15 (3.5h) · Aug 16 (3.5h) · Aug 19 (1.5h review). Recent Google design rounds were product-flavoured (token-based rate limiter, anomaly detection), not pure infra. TODO · confirm specific focus

Must do

6-step sequence from memory, out loud, with time budgets sd_playbook_quarter.pdf. Announcing transitions is the single habit the mock exists to test.
Google mock — Thu Aug 13, 4–5pm PT Pick your two default deep dives before the call so you're not choosing them live.
Mock debrief, Aug 14, 30 min Write down the two weakest moments and which step you rushed. This drives the weekend blocks.
Token-based rate limiter drill The Aug 5 onsite question. Token bucket, but the unit is billing tokens. The wrinkle: you can estimate tokens before an LLM call but only know actual usage after, so reserve on the estimate and reconcile on completion.

Follow-ups

Anomaly detection drill Streaming metrics → alerts. Probes: missing/delayed data, seasonality, spike vs drop vs drift, no labels, false-positive suppression, feedback loop.
Aug 19 review — your own worked examples only systems-design/worked-examples/ plus the two quarter sheets. No new material the day before.

Onsite day — Thu Aug 20

in personIV▾

11:15–12:00 DS/A · 12:00–1:00 lunch with a Googler · 1:00–2:00 systems design · 2:00–2:45 DS/A. No new material today.

30-min warm-up, morning One problem you've already solved, purely to get your hands moving. Skim the pattern index in leetcode_cheatsheet.html.
Write Python by hand once before the 20th Google still codes in Google Docs. No autocomplete, no formatter, no run button. Multiple reports mention this.
DS/A rounds: brute force + Big-O before optimising Expect one problem that escalates, not two separate ones. Finish the simple version fast so there's runway for follow-ups.
Design round: requirements before boxes, pick your own dives Never name a technology without the trade-off attached.
Lunch still counts Not rubric-scored, but the Googler's impression can reach the packet.
After: log all three rounds while fresh interview_history.html, plus new shapes in learning_journal.html.

Rules of engagement

Finish all MUST DOs in P1 before starting any P2 follow-up. That's the whole point of dropping dates — breadth across the topics that are actually being asked beats depth in one.
Never trade a weekend sys design block for extra LeetCode. Round 2 is a full hour and 12h is already the floor.
If a problem runs 20+ min past budget, stop, read the solution, write the insight in the journal, schedule a retry. Grinding past the budget teaches the wrong reflex for a 45-minute round.
Aug 19 stays light and Aug 20 gets nothing new. Cramming a fresh pattern the night before has never once helped.
Falling behind? Cut in this order: P3 follow-ups → P3 must-dos → P2 follow-ups. Never cut a P1 must-do or the mock debrief.

Aug 11–20 · topic-ordered by 1point3acres evidence · sys design 12h, mock Aug 13, onsite Thu Aug 20 · evidence in google-prep.html · log sessions in learning_journal.html