--- title: study-plan pageLayout: default slug: hinterland/prep/study-plan permalink: https://aarnphm.xyz/hinterland/prep/study-plan.md generator: quartz: v4.6.0 hostedProvider: Cloudflare baseUrl: aarnphm.xyz full: https://aarnphm.xyz/llms-full.txt --- # study plan Two variants. Pick by days remaining. Every solve follows the cold-solve rule in the README: stub first, harness as judge, solutions only after. Keep a redo list of every function you missed; a miss is cleared by re-solving from a clean stub, not by reading the fix. Time boxes are fixed. When a day runs over, drop the stretch targets and keep the mock. ## 5-day plan Daily shape: read the kit’s `notes.md`, cold-solve the listed stubs in order, check failures against `solutions.py`, re-solve misses from clean stubs, finish with the rapid-fire drills in the notes. Mocks run in the evening under a real 60-minute timer and get graded with the rubric in `mock-screens.md` the same night. ### day 1: underpinnings (01-bits + 02-unsigned-alignment) | block | work | | --------- | ------------------------------------------------------------------------------------------------------------------- | | 0:00-0:30 | read `01-bits/notes.md` sections 1-5 | | 0:30-1:45 | cold-solve `pack_rgba`/`unpack_rgba`, `extract_field`/`insert_field`, `next_pow2`, `reverse_bits32`, `sar32` | | 1:45-2:15 | read `02-unsigned-alignment/notes.md` sections 1-2 | | 2:15-3:15 | cold-solve `align_up`/`align_down`/`is_aligned`, `add32`/`sub32`/`mul32`/`shl32`/`shr32`, `to_unsigned`/`to_signed` | | 3:15-3:45 | rapid-fire drills from both notes; misses go on the redo list | Stretch: `popcount_swar`, `struct_layout`. ### day 2: 03-varint, then mock A | block | work | | ------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | 0:00-0:45 | read `03-varint/notes.md` in full; the malformed-classes and canonicality section is the highest-yield prose in the kit | | 0:45-2:30 | cold-solve `encode_uvarint`, `decode_uvarint` (all three error tokens: truncated, too long, overlong), `zigzag_encode`/`zigzag_decode`, `encode_svarint`/`decode_svarint`, `decode_uvarint_seq` | | 2:30-3:00 | re-solve misses from clean stubs; drills section (d) | | evening, 1:15 | mock screen A, then grade it | Stretch: `encode_vlq`/`decode_vlq`. ### day 3: 05-streaming | block | work | | --------- | ------------------------------------------------------------------------------------------------------------------------- | | 0:00-0:40 | read `05-streaming/notes.md`; internalize state = (accumulator, shift) and the all-splits invariance the harness enforces | | 0:40-2:30 | cold-solve `StreamingVarintDecoder`, `FrameDeframer`, `TlvStreamParser` | | 2:30-3:15 | `Utf8StreamValidator`; it is tagged hard, budget the whole block | | 3:15-3:30 | drills | Stretch: `cobs_encode`/`cobs_decode`, `ChunkedDecoder`. ### day 4: 04-byte-streams, the 07/08 screen-core set, then mock B | block | work | | ------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | 0:00-0:40 | read `04-byte-streams/notes.md`; memorize the struct format-char and prefix tables | | 0:40-2:00 | cold-solve `read_u32_le`/`read_u32_be`, `write_i64_le`, `bswap32`, `BinaryReader` | | 2:00-2:45 | `hexdump` (the format is byte-exact), then `float_to_bits`/`bits_to_float` | | 2:45-3:00 | drills | | 3:00-4:15 | cold-solve the 07/08 screen-core set: `MovingAverage`, `StreamingMedian`, `sliding_window_max` in `07-stream-algorithms`, then `TokenBucket`, `mm1_metrics` in `08-queueing` | | evening, 1:15 | mock screen B, then grade it | Stretch: `float32_parts`, `SlidingWindowCounter`. ### day 5: 06-codecs, mock C, consolidation | block | work | | ------------- | ------------------------------------------------------------------------------------------------ | | 0:00-0:40 | read `06-codecs/notes.md` | | 0:40-2:15 | cold-solve `rle_encode`/`rle_decode`, `utf8_encode`, `delta_varint_encode`/`delta_varint_decode` | | 2:15-3:00 | `utf8_decode` (hard) or `b64_encode`/`b64_decode`; pick from your redo list | | 3:00-3:30 | clear the redo list: re-solve every remaining miss of the week from clean stubs | | evening, 1:15 | mock screen C, then grade it | Night before the screen: read `cheatsheet.md` once, slowly. Solve nothing new. If you lose a day, fold day 1 into 15-minute drill blocks attached to the remaining days and keep all three mocks. The mocks predict the screen better than any single kit. ## 1-day cram Five and a half hours. Covers the single most likely question shape (varint ladder plus streaming) and nothing else. | block | work | | --------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | 0:00-0:45 | read `cheatsheet.md`; hand-write the uvarint vector table and the zigzag table from memory until both are exact; skim the sketches and queueing sections once, as discussion ammunition, not solve targets | | 0:45-2:15 | `03-varint` cold: `encode_uvarint`, `decode_uvarint`, `zigzag_encode`/`zigzag_decode`, `decode_uvarint_seq` | | 2:15-3:15 | `05-streaming` cold: `StreamingVarintDecoder`, then `FrameDeframer` as far as the hour allows | | 3:15-3:30 | break, away from the screen | | 3:30-4:30 | mock screen A on a strict timer | | 4:30-5:00 | grade the mock; reread `00-recon/intel.md` sections 3, 4, and 5 (what graders reward, the clarifying questions, the red flags) | | 5:00-5:30 | cheatsheet again; say the three malformed classes and the roundtrip vector set out loud from memory | If the real screen deviates from the varint shape (UTF-8 validation, RLE, buffered writer), the cheatsheet rows plus the intel company table are the fallback briefing; both fit in the last half hour.