Hacktoberfest 2026：メンテナが10月に向けて印を付けた、オープンで初心者向けの issue。 Hacktoberfest の issue を見る

Comment on the key moments of an interview, in the report and the replay

オープン
  • 何か問題がありますか？
#258 コメント 2 件 リアクション 0 件 担当者 0 名 GitHub で見る

メンテナーはふだん 1 日以内に返信

まだ誰も着手していません。

評価

難易度
4/5
見積もり時間
3〜5日
初心者へのやさしさ
45/100
issue の種類
機能追加
明瞭さ
おおむね明確
活発さ
活発
技術スタック
rust
領域
backend-api-design, full-stack

調査の方向性

Start from where phase judgments are produced and grounded by quote_is_grounded, then from where the report is assembled from the in-memory session, since the moment rules (hint requested, failed-then-passed test run, long response window, first REACTO/STAR step, large rewrite) must read recorded session events. Then trace the replay page's existing "Moments" list to attach notes to matching entries, using event occurrence times rather than evidence-recording times (see #169). Done means a capped "Key moments" section renders with quotes verified against the session, ungrounded notes dropped, scores unchanged, and replay links only when a recording exists.

索引モデルが issue の本文から書いたものです。

説明

Who is affected

Candidates reviewing an interview afterwards. The report says what went well and what to improve, but in general terms ("relied on hints", "could have planned the dictionary update after the lookup"). It does not say where in the interview each of those happened. The replay page lists every editor snapshot, test run and response window under "Moments", but leaves the candidate to work out which ones matter and why.

Desired outcome

A handful of moments, picked from what the session recorded, each with one or two sentences of coaching tied to its time. They appear in the report as a "Key moments" section, and in the replay as notes on the matching entries under "Moments", so the candidate can open the moment and see the code and transcript as they were.

An example from one of my local test interviews, on Chargeback Pair Match:

  • 04:40, tests failed on [3,2,4], 6. You read the failing case, saw that 3 had matched itself, and moved the insert after the lookup. Next time, list the case where the target is twice one amount before writing the loop, and the bug never appears.

Proposed shape

  1. The platform picks the moments, by rule. The session already records the events that make a moment worth a note, so a model does not need to search for them: a hint requested (given, or withheld), a test run that failed and a later one that passed, a long response window, the first time each REACTO or STAR step was recorded, and a large rewrite of the code. A cap, for example at most six, keeps the list short.
  2. The report model writes the notes, grounded the way the phase judge is. Each note must quote the candidate's words or the code from that moment, and the server keeps a note only when the quote is found there, as quote_is_grounded does for phase judgments. A note with no grounded quote is dropped, not shown with a warning. Notes coach; they do not score, and they change nothing in the report's scores.
  3. The times come from when the event happened, not from when its evidence was recorded, so the notes do not repeat #169.
  4. It works without recording. The report is written from the session in memory, so the "Key moments" section does not need recording to be provisioned. The links into the replay appear only when a recording exists.

Later, and optional: a highlight video

With the moments and their notes in hand, a short video could walk through them: the code at each moment, the transcript as captions, and the note as narration. HyperFrames renders HTML compositions to video and would suit this. It needs Node.js, a headless Chrome and FFmpeg, which the runtime deliberately does without, so it belongs in an offline script that reads a downloaded report or replay, not in the server. It should use only the code, the board and the transcript, never the camera.

Open questions

  • Should the moment rules be fixed, or depend on the interview loop (coding only, coding plus behavioral, whiteboard)?
  • Does the note pass belong in the report call, or in a separate call after it? A separate call adds latency but keeps the report schema unchanged.
  • How many moments are useful before the list becomes another wall of text?

Related: #169 (evidence timestamps), #242 (a review aimed at learning).

主要言語
Rust
スター
160
フォーク
64
平均マージ
1日 9時間
マージ済み PR（30日）
114

環境構築

はじめの一歩

  1. issue を最後まで読み、次にプロジェクトのコントリビューションガイドを読みます。
  2. 着手することを issue にコメントします — 二人が同じ作業をするのを防げます。
  3. リポジトリをフォークし、ブランチを切って変更します。
  4. issue 番号を参照したプルリクエストを送ります。

sysprog21/codetrial のほかの issue

sysprog21/codetrial の issue をすべて見る

似ている issue

Rust の issue をもっと見る

新しい issue をメールで受け取る

初心者向けの GitHub issue を短くまとめたダイジェスト。