Skip to main content

Advanced QA — What It Does (and Doesn’t) Do

Bookcicle’s Advanced QA layer is an optional second pass that runs after a successful conversion or editorial/translation workflow. It uses a carefully prompted LLM to assign a verdict per section:

  • PASS — usable as-is,
  • PATCH — usable, but wants a few surgical fixes,
  • FAIL — fundamentally broken and should be re-run.

This page explains how that works for each service, and what your job is once you get your results back.

1. The Core Model: PASS / PATCH / FAIL​

For every section, Advanced QA:

  1. Compares SOURCE vs CANDIDATE
    • SOURCE = your original manuscript chunk (or original language).
    • CANDIDATE = the edited/converted/translated chunk.
  2. Scores internal quality (0–100) internally.
  3. Chooses:
    • PASS if the text is production-ready and issues are trivial.
    • PATCH if a small number of local fixes (≤ 10 HTML blocks) can clearly improve fidelity or clarity.
    • FAIL if the result is structurally or factually broken and not fixable with local patches.

PASS​

  • The output is safe to ship.
  • There may be tiny style differences or preferences, but:
    • Meaning, structure, and voice are intact.
    • HTML is well-formed.

PATCH​

  • The output is broadly good, but:
    • There are a few high-impact issues worth fixing.
    • Advanced QA returns small HTML patches, each keyed to a data-uid block.
  • Patches are microsurgery, not rewrites:
    • No new scenes or paragraphs out of nowhere.
    • No big structural changes.
    • Only up to 10 blocks per section can be patched.

FAIL​

Reserved for major problems, such as:

  • Whole sections missing, duplicated, or scrambled.
  • Hallucinated large additions (new scenes, new facts, new events) outside the requested edit scope.
  • Broken HTML that will not render safely (unclosed tags, corrupted markers).
  • Severe structural corruption (e.g., a screenplay scene where the story beats are out of order or incomprehensible).
  • Systematic, global voice/tone shift that clearly contradicts the requested rubric.
  • Severe factual corruption relative to the SOURCE.

Important: We explicitly instruct Advanced QA not to use FAIL just because it can imagine many tiny style tweaks. If the text is shippable after a handful of local fixes, it should prefer PATCH, not FAIL.

2. Editorial QA (Copy Editing, Line Editing, etc.)​

When Advanced QA runs on an Editorial task:

  • It uses an edit rubric specific to your chosen Edit Type (e.g., Copy Edit, Line Edit).
  • It checks:
    • Did the edits stay within scope?
    • Is the author’s voice and meaning preserved?
    • Are grammar, clarity, and flow improved (or at least not harmed)?

What PASS looks like (Editorial)​

  • The edited text:
    • Fixes obvious issues (grammar, punctuation, awkward phrasing).
    • Respects your requested depth of edit.
    • Doesn’t introduce new plot points or character changes.

You should still:

  • Skim for subjective voice choices.
  • Decide whether you want to accept all changes or soften them.

What PATCH looks like (Editorial)​

Examples:

  • A few lines where:
    • The editor “tidied” profanity or intensity in a way that changed character voice.
    • A sentence became ambiguous or slightly off in meaning.
  • Advanced QA will:
    • Identify up to 10 specific paragraphs/spans.
    • Suggest revised HTML blocks that restore fidelity while keeping improvements.

Your job:

  • Review the patch explanation.
  • Decide whether to accept patches globally, or manually adjust in your DOCX/HTML editor.

What FAIL looks like (Editorial)​

Examples:

  • Entire scenes rewritten into a new plot.
  • A supposed “Copy Edit” pass that did a full developmental rewrite, changing POV, timeline, or genre.
  • Text riddled with broken HTML tags.

In these cases, the system will:

  • Mark the section as failed.
  • Re-run that section with a new prompt, or surface it as completed_with_issues for manual intervention.

3. Literary Convert QA (Screenplays, TV, Plays, Comics)​

For Literary Convert, Advanced QA has two priorities:

  1. Story fidelity — did we keep the beats, characters, and outcomes you wrote?
  2. Format sanity — does the output look like a screenplay/TV script/play/comic script, not just prose?

What we consider “good enough”​

  • The conversion:
    • Preserves major events and character actions.
    • Delivers a usable script draft in the chosen format.
  • For screenplays:
    • There is at least one clear scene heading (slugline) per sequence.
    • Action lines, character cues, and dialogue are recognizable and consistent.

Where we are intentionally lenient (Screenplays / Scripts)​

Advanced QA does not try to be a hyper-strict film-school grader.

We explicitly do not fail a screenplay section just because:

  • Not every small location shift has its own slugline.
  • A continuous escape or chase sequence runs under one broader scene heading, even if stricter guides would split it.
  • Action paragraphs could be split more aggressively for readability.

If the story is faithful and the script is clearly fixable with a handful of tweaks, Advanced QA should:

  • Return PATCH (with suggested improved sluglines or splits), or
  • PASS if the remaining issues are minor polish.

When Convert QA will FAIL​

Convert QA will use FAIL when:

  • Major beats are missing or invented.
  • Character attributions are badly scrambled (e.g., two characters merged).
  • The order of events is so broken that the scene no longer makes sense.
  • The output isn’t recognizable as the target format at all (e.g., prose paragraphs with no script structure).

Your job post-conversion (Convert + Advanced QA)​

After you receive a converted & QA’d script:

  1. Scan scene headings & acts.
    • Add or adjust sluglines where necessary, especially for major location/time changes.
  2. Tighten for production.
    • Shorten action blocks.
    • Clarify choreography.
    • Add transitions and production notes as needed.
  3. Compare to your key beats.
    • Ensure that your inciting incident, midpoint, climax, and ending are all present and correctly staged.

If you want stricter convention, consult:

  • Professional slugline & scene heading guides, e.g. StudioBinder, ScreenCraft, Backstage, Arc Studio, and BBC Writersroom resources.

These resources show the “gold standard” for scene headings, sluglines, and script layout; your Bookcicle draft is meant to be a strong starting point, not a final locked shooting script.

4. Translation QA (Literary Translation)​

For Literary Translation, Advanced QA behaves more like a bilingual sensitivity reader:

  • It checks:
    • Meaning preservation: no dropped or invented facts.
    • Tone & register: formal vs informal, slang, profanity, etc.
    • Paragraph-level alignment: sections roughly correspond to the source.

PASS (Translation)​

  • The translation sounds natural in the target language.
  • No obvious meaning errors, omissions, or additions.
  • Voice, tone, and register are consistent with the source.

PATCH (Translation)​

  • One or more localized issues, for example:
    • A sentence mistranslated, flattening the emotional stakes.
    • A nuance lost (e.g., sarcasm turned into sincerity).
    • A culturally awkward phrase where a better idiom exists.
  • Advanced QA will propose local fixes only, not retranslate whole chapters.

FAIL (Translation)​

  • Large segments of the source missing or condensed.
  • Entire paragraphs with incorrect meaning (wrong subject, wrong action).
  • The translation reads like a summary, not a sentence-level rendering.

5. How to Use Advanced QA Results as an Author​

Think of Advanced QA as a sharp but pragmatic second pair of eyes:

  • It will not replace your craft judgment.
  • It will:
    • Catch many structural/fidelity issues that are easy to miss in a big project.
    • Give you a prioritized set of fixes (PATCHes) instead of a vague “this is bad” comment.
  1. Start with conversion or editorial/translation.
  2. Add Advanced QA when:
    • You’re nearing a submission draft.
    • You want an extra layer of structural and fidelity checking.
  3. Review the QA summary:
    • PASSED sections: spot-check only.
    • PATCH sections: read the explanation; accept or adapt patches.
    • FAILED sections: decide whether to:
      • Re-run via Bookcicle, or
      • Hand-edit the original and re-queue.

If you ever feel Advanced QA is being too strict (e.g., failing over a formatting nitpick), or too lenient ( missing an issue you care about), tell us: those signals help us adjust the rubrics and make the system behave more like a human pro reader you trust.