Hacktoberfest 2026: le issue che i maintainer hanno segnato per ottobre, aperte e adatte ai principianti. Sfoglia le issue Hacktoberfest

Treat low-evidence candidate speech as unassessed

Aperta
  • Qualcosa non va?
#553 1 commento 0 reazioni 1 assegnatario Vedi su GitHub

I maintainer di solito rispondono entro 1 giorno

@Danielncku ci sta già lavorando.

Dal 30/10/2026.

Valutazione

Difficoltà
5/5
Tempo stimato
Più di una settimana
Idoneità per principianti
35/100
Tipo di issue
Funzionalità
Chiarezza
Abbastanza chiara
Stato di attività
Attiva
Stack tecnologico
rust
Ambito
backend

Direzione di ricerca

Read PR #423 and follow-up issues #123 and #553 to understand the existing communication eligibility flow. Locate where candidate turns determine assessment eligibility and review how transcript data is recorded and used. Agree on a deterministic evidence threshold before implementation; done means the listed transcript cases leave communication unassessed when evidence is insufficient, with coverage for each case.

Scritto dal modello di indicizzazione a partire dal testo della issue.

Descrizione

Background

PR #423 fixed the zero-candidate-turn case by leaving communication
unassessed when no candidate speech was recorded.

(guys can check out pr#423 first before fixing this issue)

However, the current assessment eligibility is still mainly determined by
whether at least one candidate turn exists. A transcript segment containing
a cough, background noise, a filler word, or a stray ASR result may therefore
enable communication scoring even when there is not enough meaningful
evidence to evaluate the candidate.

This is a follow-up to #123 and #423.

Current behavior

A session may be treated as communication-assessable as soon as one
candidate turn is recorded.

For example, the following may potentially count as evidence:

  • A cough or background noise transcribed as text
  • Filler-only speech such as "uh" or "um"
  • A short acknowledgement such as "yes"
  • A stray or malformed ASR transcript
  • Interviewer audio captured again through microphone echo

This may cause the report to generate a communication score and strengths
that are not supported by meaningful candidate speech.

Expected behavior

Communication should only be assessed when the transcript contains enough
meaningful candidate speech to support a reliable evaluation.

Sessions should distinguish between at least these states:

  • No candidate speech captured
  • Candidate speech captured, but evidence is insufficient
  • Sufficient evidence for communication assessment

For the first two states:

  • communicationScore should remain null
  • Communication should be displayed as unassessed
  • Communication strengths should not be invented
  • Coding activity, test runs, and code comments should not be treated as
    spoken communication

Proposed direction

Introduce a deterministic communication-evidence classification instead of
using only candidate_turns > 0.

The exact threshold should be discussed before implementation. Possible
signals include:

  • The number of substantive candidate turns
  • The amount of meaningful transcript content
  • Filtering filler-only and noise-like transcripts
  • Speaker attribution and ASR confidence, when available

The implementation should avoid relying on a model alone to decide whether
the available evidence is sufficient.

Validation

Add coverage for:

  • Zero candidate turns
  • Noise-only or cough-like transcripts
  • Filler-only speech
  • A single short acknowledgement
  • One substantive explanation
  • Multiple meaningful candidate responses
  • Mixed noise and meaningful speech
  • Empty or malformed ASR results
  • Interviewer echo incorrectly captured as candidate speech

Out of scope

This issue does not require implementing general-purpose audio denoising or
speaker recognition. Its scope is deciding whether the recorded transcript
contains enough reliable candidate evidence for communication assessment.

Lingua principale
Rust
Stelle
215
Fork
141
Merge medio
1g 12h
PR unite (30g)
230

Preparare l'ambiente

Come iniziare

  1. Leggi tutta la issue e poi la guida ai contributi del progetto.
  2. Commenta sulla issue per dire che te ne occupi tu — evita che due persone facciano lo stesso lavoro.
  3. Fai un fork del repository e lavora su un branch.
  4. Apri una pull request che faccia riferimento al numero della issue.

Altre issue di sysprog21/codetrial

Tutte le issue di sysprog21/codetrial

Issue simili

Altre issue su Rust

Ricevi le nuove issue nella tua casella

Un breve riepilogo di issue GitHub adatte ai principianti.