TranscriptX assessment (2026-10-23)

Sibling repo: /Users/89298/Documents/transcriptx. Package 1.0.1. Tags v1.0.0 and v1.0.1. HEAD was about 13 commits past v1.0.1 (Granite STT + library archive unreleased). Classifier in pyproject.toml still said Beta.

This product is a local audio transcript analysis workbench. Streamlit default port 8501. Transcribe must not depend on it.

Layout (portable vs not)

Package

Role

Copy into Transcribe?

src/transcriptx/core/

Engine: pipeline, ~50 analysis modules, domain transcript

Pattern only (adapters / registry)

src/transcriptx/app/

UI-agnostic controllers

Pattern only

src/transcriptx/web/

Streamlit shell, settings hub, CCv2 bridges

IA / settings chrome pattern

src/transcriptx/io/

Import adapters, tag validation

Adapter registry pattern; not the tag catalog kernel

src/transcriptx/services/

Host STT, watcher, speaker studio

No (audio)

packages/transcriptx_workspaces/

Vitest/Vite CCv2 workspaces

No for v1

Shipped (audio product)

  • BYO import plus in-app STT (WhisperMLX, WhisperX Docker, Granite on HEAD).

  • Diarization ladder, Speaker Identification, speaker profiles, Groups.

  • Library (including archive flag on HEAD), Search, CCv2 karaoke reader, Corrections Studio.

  • Analysis presets Quick / Balanced / Thorough / Custom; large module set including voice/prosody and multi-speaker interaction modules.

  • Workspace backup ZIP (transcriptx.workspace-backup), Settings hub, Diagnostics, schema-epoch gate.

  • Export Overview/Artifacts ZIP, optional EPUB, Library export-by-tag.

Maintainer infrastructure (1.0 gate shape)

Piece

TranscriptX

PR CI

Blocking: compose-config, lint, smoke. Advisory: NLP/RAG tests, docs, release-checks, workspaces

Nightly

nightly.yml: contracts + fast + coverage; integration-core

Coverage badge

coverage-badge.yml

Pages

pages.yml website + Sphinx

Tests

pytest + AppTest GUI acceptance + Playwright + Vitest for CCv2

Release docs

docs/dev/pre_release_roadmap_1_0.md, severity triage, release ops, manual acceptance, install/trust/perf matrices

Scripts

scripts/release/* plus clean_env_audit.sh, image_pip_check.sh, perf envelope helpers

Transcribe should steal nightly offline lanes, issue templates, Docker-in-release-checks, severity/ops/manual-acceptance docs — not TX’s smoke-only PR policy. Transcribe already runs the full offline matrix on every PR.

Portable patterns

  • Nav groups: primary / Workflow / View / System.

  • Settings hub with segmented tabs (content differs by product).

  • Workspace backup ZIP with a typed format id and role roots.

  • Doctor / Diagnostics page.

  • Analysis module registry + presets + result envelopes.

  • Hosted docs (Sphinx + modest website/ + Pages).

  • First-run honesty: docs-first; TX removed Guided/demo/checklist modes after trying them.

Do not copy (audio / speakers / timed media)

STT providers, diarization, karaoke timings, voice feature family, Speaker ID workstation, inbox-watch / recordings mounts, interaction/QA/echoes/contagion modules, Compose “engines outside analysis image”, Playwright/Vitest workspace packs, Dashboard Builder as a 1.0 Transcribe requirement.

Tagging honesty

TX has organisation string tags (io/tag_validation.py, Library filter/export). It does not ship Transcribe’s personal_corpus.tag-catalog kernel. “Shared with TX Theme F” in Transcribe contracts is a future copy boundary, not current TX code.

Maturity line

Public 1.0 is tagged. ROADMAP “Current state” still mentioning a 0.9.x track is stale relative to tags. Post-1.0 work continues (ASR ladder, library archive, Theme E/H). PyPI and RTD hostname go-live were not 1.0 blockers.