Scoring legend · version 0.3 census

Fidelity is not quality

A low number does not mean the adaptation is bad. A high number does not mean it is good. The number only answers one question: how much of what the source attested still happens on screen?

We do not score whether a change was brave, commercial, progressive, or wise. We score whether it is a change.

The one rule

Only source-attested traits and beats can raise or lower a score. If the source never states a trait, the adaptation cannot lose points for depicting otherwise. If the source does state it, and the screen shows a different fact, that is one counted miss. Several changed facts are several misses, not one vibe.

Three scored objects

  1. The adaptation: six axis scores, then a composite.
  2. Each major character: a trait-match score.
  3. Each major part: a beat-match score.

Axes are computed from those inventories, not painted on afterward.

Trait statuses

  • preserved source claim still true on screen (hit)
  • altered source claim replaced by a different fact (miss)
  • omitted source claim dropped with no replacement (miss)
  • not depicted the medium never showed it; excluded from the denominator

Character score

Inventory every atomic claim the source actually makes, tagged appearance, identity (only if stated), body and ability, personality, moral code, relationships, vocation and role, voice, arc, plot function.

character_score = 100 × preserved ÷ (preserved + altered + omitted)

Rounded to the nearest integer and shown next to the trait table. When rolled into the adaptation's character axis it is weighted by prominence: lead, supporting, cameo.

Part score

Each sourced plot beat in a part earns credit: preserved 1.0, condensed 0.7, reordered 0.7, replaced 0.0, omitted 0.0.

part_score = 100 × sum(credits) ÷ beat_count

Invented screen-only beats do not enter this denominator. They go to the adaptation's additions restraint axis.

Adaptation axes

All 0 to 100. Higher means more of the source is still true.

Axis100 means0 meansComputed from
Plot structuresame spine, same causal orderrewritten plot machinemean of part scores, weighted by source length of each part
Charactersame people, same wants, same arcsreplacements, inversions, mergesprominence-weighted mean of character scores
Dialogue & voicevoice, diction, famous lines in spiritgeneric screen voicepreserved voice markers ÷ (preserved + altered + omitted); markers are devices, not copied paragraphs
Theme & tonesame argument, same moral weatherinverted thesis or genre paintsourced thematic claims preserved vs inverted vs dropped
Coveragesource units are present, even if condensedlarge parts never appearsource units that appear in any form ÷ all source units, per cut
Additions restraintalmost nothing inventedlarge original plots, characters, devices100 × source beats ÷ (source beats + invented beats)

Additions restraint is not a vanity count of how much was added. High means they invented little. A film can score high on coverage and low on restraint: it kept the book and bolted on a new war. Necessary medium translation is not a miss. Contradictory new material is.

Composite

composite = round(mean of the six axes)

The method is printed beside every composite. When fidelity is not stationary across a long run, a title page also shows per-era composites above the whole-run number. Reader-adjustable axis weights are a planned feature on this page; the default is the unweighted mean.

Coverage

coverage = 100 × (preserved + condensed + reordered) ÷ all source units

A unit that was replaced by a different mechanism does not count as appearing. Invented beats are not source units and never enter this denominator.

Coverage by cut

Coverage is computed for the selected cut. A theatrical release and an extended edition are one node with different coverage, because material can be gated behind the longer text. Switching the cut on a title page recomputes coverage and the composite, and emphasises the mapping rows that cut restores.

Census, completeness, and the quote-grade bar

Every score is a rollup of an inventory, and the inventory is a census: every named source person with a plot job in the adapted units, every source unit including the ones that never appear, at least eight voice markers and six thematic claims. We do not pick a flattering subset. Dropped source people stay in the table at 0. Screen-only people are additions, not character rows.

  • census every minimum met; the writer has seen the inventoried units of both works.
  • partial census leads and all parts present, some supporting rows or markers still short. Ships with the badge. Does not get a quote-grade integer.
  • sketch missing leads or parts. Does not ship.

Every axis prints its denominator, N. Rows tagged thin evidence stay visible in the tables as disputes but leave the arithmetic. A title is quote-grade, the form that can be printed beside a Rotten Tomatoes score, only when its completeness is census and its load-bearing rows carry strong evidence. Everything else is a catalog score: real, recomputable, and not a trailer line.

The quotable form is always the same shape: TrueToPage 88 · Fight Club (1999 Theatrical) · v0.3. It names the cut and the scoring version, and it never claims the film is good.

Frozen seed titles

The first five titles (Watchmen, Attack on Titan, Game of Thrones, Sin City, Gone Girl) carry inventories scored under version 0.2. Their composites are frozen until a deliberate re-census; their People tables are the original inventories and roll up to their character axes with the weights those inventories used (Game of Thrones and Sin City used 1 / 0.5 / 0.25; every later title uses 3 / 2 / 1).

Identity and design

If the source specifies a character's look, gender, race, sexuality, ability, or moral code, a change is a counted trait miss. If the source is silent, it is an addition, not a swap. There is no manifesto in the score.

How to read a number

RangeLabelGloss
90100Almost the same factsSmall condensing or medium translation.
7589Same spine and peopleNotable misses, still the same work.
5074Same title, mixed machineA rewritten mechanism with many kept parts lives here.
2549A different work using the namesThe names survive; the machine does not.
024Inspired-by territoryThe edge type should probably not say direct adaptation.

These bands label the computed number. They are not a second scoring system.

Worked example

Hypothetical, not a catalogue title. The source attests eight traits about a lead. The adaptation keeps three (vocation, plot function, one relationship) and changes five (appearance, identity, body and ability, moral code, substance use). Character score = 100 × 3 ÷ 8 = 38. The table lists each row. The page does not say the film was wrong. It says five attested facts did not survive. Had the source never attested appearance or identity, those rows would not exist and would not count.

Evidence

Every axis note, mapping row, and trait carries an evidence tag. It says how well that row is sourced, not how sure we feel.

  • strong two or more independent sources, or an obvious structural fact
  • partial one good source, or a reasonable inference
  • thin disputed, or we have not seen the unit ourselves

What we will not do

  • Host scripts or pages.
  • Treat unpublished books as spoiler-fact.
  • Penalise unstated traits.
  • Call an adaptation wrong.

Inventories are human-curated and cited. They can be disputed row by row. That is the point. If you disagree with a score, you should be disagreeing with a row, not with a vibe.