Why Your Script Keeps Getting Passes: A Diagnostic Guide

Published by StoryNotes. We sell AI screenplay coverage, so we have an obvious interest in you buying a diagnosis. Every number below is sourced to the company that published it, including the ones about our own category.

Three passes in and the pattern is always the same: a verdict, a paragraph of summary, a few lines of notes that could describe half the scripts written this year, and no answer to the only question that matters. Not is it good. Where does it break, and on what page.

The answer to why my script keeps getting passes is almost never inside the coverage that passed on it, and that is a structural fact about the product rather than bad luck. Here is the reframe that makes the rest of this useful. A pass is not a measurement of your script. It is the default output of a system that passes on nearly everything, delivered by a reader whose job is to decide whether to advance a script rather than to explain what is wrong with it. Those are different tasks, and coverage only ever pays for the first one.

TL;DR: Pass is 83% of all coverage, and true recommends effectively do not exist. Across 128,020 Black List evaluations, the mean score is 5.74 and 83.3% of scripts land on a 5, 6, or 7 — so a single pass tells you almost nothing about where you sit. Stop reading verdicts and start mining for repeats: the note two readers both make, converted into a page number, is your real diagnosis. Most of the time it resolves to one of five structural patterns, and all five have tells you can go find in your draft tonight.

It means the reader is not advancing your script. That is the entire content of the word.

The distribution is the part nobody shows writers. Coverage Ink, which has been producing coverage for 22 years, published its own breakdown: 83% pass, 15% consider with reservations, 2–3% consider, and a recommend rate of effectively zero — the company says it can count the true recommends it has seen across those 22 years on three hands (Coverage Ink). Screenplay Readers reports the same order of magnitude from the other side of the desk: two recommends given company-wide in a year (Screenplay Readers).

Readers also have a professional reason to withhold the top verdict that has nothing to do with your pages. A recommend puts the reader's judgment on the line with a supervisor who may disagree, and as Coverage Ink puts it, readers learn quickly not to stick their neck out. The scale is asymmetric by design: pass is free, recommend is expensive.

The numeric scales behave identically. In March 2026 The Black List published the distribution across 128,020 evaluations — 92,613 of them film — and the shape is a narrow hill sitting just above the middle:

The mean film score is 5.74, the median is 6, and 83.3% of everything evaluated lands on a 5, 6, or 7. An 8 or above is 3.8% of all film evaluations; a 9 is 0.3% (The Black List, March 2026). Getting a 6 does not put you in the bottom of anything. It puts you in the largest bucket in the industry, one point below the score that starts opening doors — which is worth knowing before you spend $100 on another evaluation hoping the number moves.

One concrete fix: Take the verdict line off the page. Print your coverage, strike through the PASS and any overall score, and reread what is left. Everything of value in a coverage report is underneath the verdict.

It means one reader, on one day, did not advance it. That is a weaker claim than it feels like, and the industry has now published the numbers to prove it.

The Black List analyzed 212,673 film evaluation pairs — the same script, two evaluators — and found the two matched exactly 30% of the time. They landed within one point of each other 73.8% of the time, which means roughly a quarter of the time two professional evaluators differed by two points or more on the identical draft. The reported reliability coefficient for film is 0.290, which the company notes sits at or above the average for published academic peer review (The Black List, February 2026).

Two things follow, and only one of them is comforting.

The first is that a single low score is genuinely noisy, and rebuilding a script on one reader's reaction is a bad trade. The second is that the noise is not random — when two evaluators diverged by two points or more, a third reader's score fell inside their range 83.5% of the time. The spread is real, but it is a spread around something. Your script has a range, and the reads are samples from it.

There is one more detail in that data with direct diagnostic value: agreement was strongest on overall quality and weakest on dialogue, plot, and character. Readers converge on whether a script works and diverge on why. Which is exactly the wrong way around for a writer trying to fix one, and exactly why the repeated note matters more than any individual one.

One concrete fix: Never rewrite structurally off a single read. Get a second opinion — another reader, a peer swap, or a structural pass you can rerun — and treat only the overlap as confirmed.

Once you strip out the "well written, not for us" passes, most of what remains traces to a small number of structural failures. They are ordered here by how much damage they do downstream: a problem in pattern one poisons every page after it, while pattern five costs you goodwill scene by scene.

Each one below has a page-level tell — something you can go look for in your own draft in the next hour — and one fix.

The reader cannot tell what kind of movie they are holding, so they cannot tell whether it is doing well at being one.

This is the most expensive failure in the list because it corrupts every judgment that follows. A reader evaluating your grounded family drama as a thriller will find act two slack and the stakes low; the same pages, read as what they are, might be paced fine. When a script does not signal its genre and tone early, the reader supplies a guess, and then measures you against it.

The page-level tell: open your draft to page ten and read only the scene headings and the first line of each scene. If someone who had read nothing else could not name the genre and the approximate tone from that alone, the declaration is missing or late. A second tell: your opening image and your climax belong to different movies.

The fix is not to bolt on a genre-signaling scene. It is to move an existing one. Almost every script that fails this test already contains the right opening — it is sitting on page 22, after the setup the writer thought had to come first.

One concrete fix: Find the earliest scene in your draft that could only exist in this genre, and see what breaks if it becomes the opening. Usually less than you think.

If the goal cannot be failed at, the story has no tension, and a reader feels that as nothing is happening long before they can name why.

"Find herself," "get closure," "prove he belongs" — these are conditions, not objectives. An objective can be blocked by a specific person on a specific page, and the audience can tell whether it was achieved. A condition cannot be blocked; it can only be described. Scripts built on conditions read as episodic no matter how well the individual scenes are written, because nothing an antagonist does can actually threaten the goal.

The page-level tell: write down what your protagonist is trying to get, as a sentence with a direct object, and then find the page where they could plausibly lose it forever. If you cannot cite a page number, the goal is a condition.

The confusion is that scripts on conditions often have excellent scenes. The writer feels the quality of the scenes and cannot reconcile it with the pass. The scenes are fine. The spine is missing.

One concrete fix: Convert the condition into an object the character can hold, lose, or be denied — the job, the letter, the custody hearing, the house — and let the internal thing be what the object is standing in for. Then put the moment it can be lost on a specific page.

The middle of the script consists of things that happen in an order, rather than things that happen because of each other.

This is the single most common structural note in coverage, and the reason act two is where most scripts lose their reader. The scenes are causally connected — this leads to that — but nothing gets harder. The protagonist's options are as good on page 70 as they were on page 45, and every setback is absorbed without cost.

The page-level tell: the "and then" test. Summarize your act two in one sentence per scene, then read the summaries aloud, inserting the correct connective between each. If you find yourself saying "and then" more often than "so" or "but," the middle is a sequence. A second tell: you can reorder three consecutive scenes without breaking anything.

The underlying cause is usually that the protagonist has not been made to pay for anything. Escalation is not a matter of raising the volume; it is a matter of narrowing options. Each act-two beat should close a door — an ally leaves, a resource is spent, a lie becomes impossible to maintain.

One concrete fix: List every act-two scene and write next to each one what the protagonist can no longer do afterward. Any scene with a blank next to it is a scene that costs nothing, and it is either cut or it takes something away.

The reader has a memory, and the script keeps making promises it does not keep.

Every deliberate detail in the first act is a promise: the injury, the estranged brother, the object on the shelf, the rule established in the opening scene. Readers track these unconsciously, which is why unpaid setups do not produce a specific complaint — they produce a general feeling of a script being loose. The mirror failure is worse: a payoff arriving without a setup reads as a cheat, and a reader who feels cheated once stops extending trust for the rest of the script.

The page-level tell: two columns. In the left, every setup in your first 30 pages, with its page number. In the right, the page where each is paid. Any left-column row with an empty right column is a broken promise. Then do it in reverse — every consequential turn after page 80, and the page that made it possible.

One concrete fix: For each unpaid setup, choose deliberately: pay it or cut it. Do not leave it in as texture. Texture is what unpaid setups feel like to the writer and loose ends is what they feel like to a reader.

Every scene runs a half-page longer than it needs to at both ends, and the cumulative effect is a script that reads slow regardless of what happens in it.

This is the one pattern readers name out loud, usually as "the pace drags," and it is also the one writers most consistently underestimate. It is arithmetic: forty scenes, each carrying a quarter-page of arrival and a quarter-page of departure, is twenty pages of nothing distributed so evenly that no single scene looks like the problem.

The page-level tell: open ten scenes at random and mark the line where the scene's actual business begins and the line after which the outcome is settled. If the material outside those marks routinely runs more than a few lines, the pattern is systemic.

One concrete fix: Enter every scene at the latest possible moment and leave at the earliest — cut on the line that changes something, not the reaction to it. Run the pass on ten scenes first and check whether anything was actually lost. It usually is not.

By treating your reads as a dataset instead of a series of judgments. Three steps.

Collect. Put every piece of feedback you have on the script in one document — paid coverage, contest notes, peer reads, the friend who trailed off. Ignore all verdicts and scores.

Mark the repeats. Any observation made independently by two or more readers is your signal, and everything said once is a hypothesis. This is the practical use of that reliability data: the industry's own numbers say individual scores wobble, and a note that survives two independent readers has cleared the noise floor.

Convert to page numbers. This is the step that turns feedback into work. "The second act drags" is unactionable; "pages 55–72 contain no scene that removes an option from the protagonist" is a rewrite plan. Take each repeated note and find the pages it lives on. If you cannot locate it, you do not yet understand the note well enough to fix it, and that is worth knowing before you start rewriting.

What usually stops writers here is that they only have one read. The whole method depends on comparison, and human coverage is priced for one purchase, not three — which is why the order you buy reads in matters more than which service you pick.

One concrete fix: Build the document today with whatever feedback you already have. Most writers discover they have three reads' worth of notes sitting in three different inboxes and have never put them side by side.

Then buy the diagnosis separately, because it is a different product from the verdict, and only one of them helps you rewrite. That is the whole case for the cheap structural tier, laid out at length in the honest guide to AI script coverage.

This is the part we sell, so here is the specific claim. StoryNotes returns a six-stage structural read of the full script in about five minutes for $20, one-time — letter grades per stage, page citations in the format "p.48 · INT. KITCHEN," and two to three concrete rewrites per stage. It covers exactly the ground the five patterns above occupy: genre and tone, premise, character, plot structure, theme, and scene-level craft. Because the same read can be run again after the rewrite, the second report answers the question a one-off human read structurally cannot — whether the fix held, or moved the problem thirty pages downstream. Judge it on the output rather than the description: the full sample reports are public, run on scripts you can read yourself.

One instruction that applies to every tool in this category, ours included: ignore the verdict line. If a report prints PASS, CONSIDER, or a headline score, skip it. That is the part with no information in it — which is the whole argument of this article, applied to us. The value is the page numbers.

Where an AI read does not help: it will not tell you whether a script is worth making, whether your voice is distinctive, or how the material sits in the market. Those are judgment calls, and the honest accounting of what AI coverage does and does not do is a separate piece.

One concrete fix: Run a structural pass before your next paid human read, fix what it finds, and spend the expensive read on the questions no rubric can answer.

Sometimes it is genuinely not the script, and a diagnostic article owes you that caveat.

A meaningful share of passes are dispositional. A production company with two coming-of-age films in development passes on your third one. A contest reader who does not like the genre reads it anyway. Coverage Ink is explicit that a pass often just means "not for us" rather than a judgment on quality — and there is no rewrite that fixes a slate conflict.

The way to tell them apart is the same method as before, and it is why the repeats matter. Dispositional passes are uncorrelated: five readers who pass for five different reasons, with no note in common. Structural passes converge, even when the language differs — "I never knew what she wanted" and "the middle felt aimless" and "I put it down around page 60" are three descriptions of the same defect.

If your notes converge, you have a rewrite. If they scatter, you may have a positioning problem instead of a craft problem, and the answer is a different submission list, not a different draft. Both of those are real answers. "Pass" is not.

One concrete fix: Lay your passes side by side and ask one question — do these readers disagree about what is wrong? Convergence means rewrite. Scatter means submit elsewhere.