Grading screenplays and short films

Two tools, both built for narrative work (feature screenplays, short films, web series episodes, sketches, music video treatments):

• Script Grade — a quality score. Letter grade, 0 to 100, Pass/Consider/Recommend (features only), six narrative axes scored 0 to 10 each (Opening Image, Inciting Incident, Structure, Pacing, Character Arc, Ending), plus specific strengths and improvements with page numbers.
• Reader / Coverage — a deeper structural diagnosis. Maps your script against both Comedic Structure (redemption arc) and Tragic Structure (failed or sealed arc), runs five rule checks per mode, and issues a verdict on which mode the script actually operates as. Includes Prospects: genre tags, comparable films, budget tier, distribution targets, talent references.

From the chat, you have two ways to submit a script:

• Paste the text — drop the script directly into the chat input and send. Works well for shorts and any text you already have copied.
• Attach a PDF — click the paperclip icon in the chat input, pick your PDF (up to 25MB, around 250 pages), and the text is extracted client-side. A file chip appears with the filename and page count. Add a message like "Grade this" or just send and the agent prompts the grader automatically.

The agent detects screenplay format automatically from INT./EXT. headings, character cues in caps, FADE IN markers. Based on length and shape it picks the right rubric: short_film for content under 25 pages, feature_screenplay for full features, web_series_episode for serialized work, sketch_comedy for premise-driven shorts, music_video_treatment for concept docs.

The grade card surfaces, from top to bottom:

• Pass / Consider / Recommend (feature screenplays only) — the industry-standard professional reader designation. Mapped to the score: 90+ Recommend, 80-89 Consider, below 80 Pass. Calibrated to industry distributions: roughly 3% Recommend, 20% Consider, 77% Pass.
• Letter grade and 0 to 100 score — A/B/C/D/F. We do not inflate scores. Most amateur features land in the 50 to 65 range and earn a Pass. That is the industry default, not a failure.
• Six axis bars — Opening Image, Inciting Incident, Structure, Pacing, Character Arc, Ending. Each scored 0 to 10.
• Strengths and Improvements — expanded by default for narrative grades. Each entry is anchored to a specific page or scene.

After a grade lands, you can run Reader for a deeper structural diagnosis. Click "Run Reader" on the grade card, or type "give me coverage on this" in the chat. Reader takes 60 to 120 seconds and produces a Coverage report with:

• Verdict at the top — comedic_structure, tragic_structure, hybrid, or unresolved, with a confidence percentage and 2 to 3 sentences of reasoning.
• Dual structural reads — your script interpreted twice, once as Comedy (Flaw to Virtue arc) and once as Tragedy (failed-arc or locked-identity). Each mode identifies eight dramatic slots (Flaw, Virtue, Want, Point of No Return, False Win or False Triumph, Crisis, Defining Choice, Resolution), runs five rule checks, and scores how well the script fits that mode.
• Overall notes — three to five highest-leverage observations.
• Prospects (Market Positioning) — genre tags, comparable films (with years and why-comps), budget tier, distribution targets, talent references, market summary.

After a grade, you can keep talking. Ask for deeper analysis on a specific axis. Ask the agent to walk through the midpoint beat by beat. Ask why the opening image scored a 4. The agent responds with written craft notes in plain language, in the chat.

What it will not do after a narrative grade: generate a new opening, draft alternative dialogue, write a revised scene, produce a polished rewrite, or output a beat sheet as a deliverable. The agent gives written analysis you can use, ignore, or argue with. The rewrite stays your work.

The Coverage card has three export options for sharing with collaborators, managers, or writers' rooms:

• Copy as Markdown — copies the full report to your clipboard. Paste into Notion, Final Draft notes, Highland, Obsidian, Google Docs, anywhere markdown is supported.
• Download .md — saves as coverage-{protagonist}-{YYYY-MM-DD}.md. Sortable by date in any project folder.
• Print / PDF — opens a print-friendly view in a new tab and auto-triggers the print dialog. Use your browser's "Save as PDF" option to generate a clean PDF for sharing.

Most AI screenplay tools try to rewrite the script for you. You hand them a draft and they hand back a "polished" version that is no longer your work. The voice flattens. The judgment calls disappear.

That is not what a script reader does, and it is not what we built. A reader points at problems. The writer solves them. When the reader identifies that the Crisis lands at page 95 instead of page 75, that is information about the script as it stands, not a prescription for what to write next. When it identifies the script as Tragic when you intended Comedy, that is a diagnostic, not a verdict on your taste.

Every observation is anchored to a specific page or scene. Not "the second act sags" but "pages 40 to 80 drift through episodic speed dating without escalating the protagonist's internal conflict." Not "weaken your antagonist" but "the False Win on page 28 is conceptually strong but underplayed in the protagonist's subjective experience."

From there you do what writers do. Consider the note. Decide whether you agree. Figure out the fix, in your voice, with your judgment. The reader does not get to make those calls.