7.0 How to use the protocol
The ten tests below are for revision, not composition. Applied during drafting they produce paralysis and self-conscious dialogue, which is the worst kind. Applied to a finished exchange they are the most efficient triage instrument in this chapter.
Four procedural notes.
- Run them in order. Tests 1–4 are structural, and a failure there invalidates the rest: there is no point tuning the attribution of an exchange that has no engine.
- Record the answer, not the verdict. The output of a test is a sentence describing what the exchange does. Dov holds the upper hand at the start and holds a diminished version of it at the end is useful. Fails test 3 is not.
- A failed test is a question. Every test below is failed by at least one exchange analysed in §9, and in three cases the failure is the reason the exchange is famous.
- Test one exchange at a time. The unit is the exchange as defined at §2.2, not the scene and not the chapter. Applying the protocol to a whole dialogue scene will produce an average, and averages conceal the passage that needs work.
7.1 The ten tests
Test 1 — The removal test
Cut the exchange and replace it with a single line of summary. What is lost that the summary cannot carry?
Write the summary line out. Literally write it: She accused him of lying; he did not deny it. Then ask what the exchange had that the sentence does not. The honest answers fall into a small set — a fact the reader needed to see arrive rather than be told about; a status shift the summary flattens; a voice the reader needs more exposure to; a texture or pace the passage was providing.
If the answer is nothing, the summary is the better version and should replace the exchange. This is not a hypothetical outcome. A substantial proportion of drafted dialogue is a scene the writer wrote in order to find out what happened in it, and summary is the correct final form.
The refinement worth adding: compare the summary against the exchange for length as well as content. An exchange of nine hundred words that a nineteen-word summary fully replaces is not merely redundant; it is expensive, and it has spent the reader’s attention on a transaction that did not require it.
Test 2 — The want test
State, in one sentence each, what each speaker wants from the other in this exchange, right now.
Both speakers. If one sentence will not come, the exchange has one engine and is an interview, an interrogation or a briefing — legitimate forms, but the writer should know which one they have written. If neither will come, there is no exchange, whatever is happening on the page.
Two disciplines make the test sharper. The want must be directed at the other speaker — I want to understand the situation is a want, but it is not a want from a person. And it must be immediate: not the story goal but the thing wanted before this conversation ends. I want him to admit he knew. I want her out of the doorway. I want to leave without answering.
Test 3 — The status test
Name who holds the upper hand at the first line and who holds it at the last.
If the answer does not change, ask the follow-up, which is the part of the test that does the work: was it contested? A status relation that is tested under real pressure and holds is a legitimate outcome, and frequently a more interesting one than a reversal. A status relation that nobody contests is a static-status exchange (§8.4).
The productive form of the test names the currency: who holds the information, who holds the physical position, who holds the right to end the conversation, who is asking and who is answering. These can move in opposite directions inside one exchange, and an exchange in which a character gains one and loses another is doing better relational work than one in which everything moves together.
Test 4 — The cost test
What does each significant disclosure cost the person making it?
Take every fact that passes from one character to another and name the cost of saying it. If a disclosure costs nothing — if the speaker is glad to say it, or indifferent — ask why it is in dialogue rather than in narration.
This is the plot domain’s specific test, and it is the one that catches the As-You-Know-Bob exchange (§8.1) at the level of mechanism rather than surface. A character can be given entirely plausible motivation for a briefing and the passage will still read as a briefing, because plausibility is not the missing ingredient. Cost is.
Test 5 — The silence test
Name what each speaker chooses not to say.
If the answer for either speaker is nothing — if every character says outright what they think and feel — the exchange is over-explaining itself, and the remedy is in §6.4.
The strong form of the test adds a second question: who else is in the room, and are they speaking? A ratified participant present throughout an exchange and silent throughout it is either doing structural work (§4.6) or is a character the writer has forgotten about, and the two are easy to tell apart once the question has been asked.
Test 6 — The implicature test
For each turn, ask: is this exactly what was called for, given what preceded it? If not, what does the deviation imply?
This is Grice (§3.1, §6.2) used as an instrument. Go through the exchange turn by turn with the four maxims in view. Mark each turn as cooperative (the expected contribution, in the expected quantity, on the expected topic, in plain terms) or as a flouting, and where it flouts, write down the implicature in a phrase.
Interpretation is direct. An exchange of twelve turns in which twelve are cooperative is running on one channel. An exchange in which all twelve flout is unreadable. The working ratio in published fiction is usually somewhere between a fifth and a half, and — more important than any ratio — the flouting turns cluster at the exchange’s crises rather than being evenly distributed.
Test 7 — The transcript test
Strip every attribution, tag and action beat. Read the remaining lines aloud in sequence. Can you tell who is speaking?
The most brutal test in the protocol and the most informative. It isolates the speech from everything the narration was doing to prop it up. Three outcomes:
| Outcome | Diagnosis |
|---|---|
| Speakers are immediately distinguishable | The character domain is being served by the speech itself |
| Speakers are distinguishable only by content | Voices are undifferentiated; the reader is tracking who knows what, not who is talking |
| Speakers are not distinguishable at all | The ventriloquist exchange (§8.2) |
The second outcome is the common one and the one writers misread as a pass. Content-based identification means the exchange would collapse if the two characters knew the same things, which is what happens in the second half of most novels.
Reading aloud is not an ornament to this test. Silent reading recovers the attributions from memory; the voice does not.
Test 8 — The substitution test
Take the exchange’s three best lines. Could another named character in the manuscript plausibly say them?
Where test 7 asks whether two speakers differ from each other, test 8 asks whether a speaker differs from the cast. It is the harder standard, and it is the one that catches the novel in which two characters are well differentiated and the other nine are the author.
The practical procedure: type the line, then write beside it the names of the three characters most likely to be able to say it. If the list is long, the line is not doing character work; if the list is empty except for its owner, the line has been earned.
Test 9 — The attribution test
Count the tags. Read them in isolation. What proportion are said or asked, and what are the others doing?
Three sub-questions.
- Density. How many turns carry an attribution? In contemporary practice the answer in a two-hander is usually well under half after the speakers are established, and an exchange tagged on every turn is being slowed for no reason.
- Verb. Every attribution that is not said, asked, replied or a near-neighbour should be justifiable. Expostulated, ventured, hissed and their kin describe the line instead of letting it perform, and their proscription — historically local, per §3.5 — rests on the sound principle that a line requiring an adverb to specify its tone has not been written yet.
- Action beats. Where an action stands in for a tag, ask what else it is doing. An action beat that only identifies the speaker is a tag with extra words. An action beat that identifies the speaker and carries relational or textural value is the most efficient object in an exchange (§5.3).
The test’s most useful output is usually a list of tags to delete rather than to change.
Test 10 — The interruption audit
Mark every point at which one speaker takes the floor from another, and every point at which a speaker yields it.
Then ask three questions. Who interrupts? If the same character interrupts throughout, the exchange has a fixed status structure, which may be correct or may be a failure of test 3. If nobody interrupts, the exchange is a formal alternation, and formal alternation is the rhythm of a committee meeting.
Where do interruptions cluster? They should cluster at the exchange’s crisis. Interruptions distributed evenly across an exchange read as a stylistic habit rather than as pressure.
Are any interruptions structural? An interrupted concession (§6.4, mechanism 10) is a plot event: something was about to be conceded and was not. Marking these reveals how many of a draft’s unresolved threads are held open by a device rather than by a reason.
7.2 Reading the results
| Failed tests | Likely diagnosis | First move |
|---|---|---|
| 1 alone | The exchange is a scene the writer needed and the reader does not | Replace with the summary line already written |
| 2 and 4 | A briefing wearing dialogue’s clothes | Give the facts to someone who does not want to say them |
| 3 alone | Static status; nothing contested | Find what one of them could concede, and make the other reach for it |
| 5 and 6 | On-the-nose; one channel | Apply §6.4 to the three most direct turns |
| 7 and 8 | The ventriloquist exchange | Rewrite each speaker’s turns in a different sentence length and a different vocabulary; do not attempt to make them “distinctive” |
| 6 alone, with 1–5 passing | Competent but flat; the material is there and the surface is transparent | Substitution testing, per §6.4 |
| 9 alone | Over-tagged or over-decorated; the exchange is being narrated at | Delete tags; convert the survivors to action beats that carry a second function |
| 10 alone | Formal alternation; a committee rather than a contest | Let one speaker take the floor from the other at the crisis |
| All of 1–5 | This is not an exchange | Cut it, or find the want that would make it one |
A warning about test 7. When a draft fails the transcript test, the instinct is to make each character more distinctive by adding verbal tics, catchphrases and idiosyncratic vocabulary. This produces the over-delivery described at §4.2 — a label instead of a person. The reliable route to differentiation is structural rather than lexical: give each character a different average sentence length, a different relation to questions, and a different default subtext mechanism. Those three variables produce distinguishable speakers without a single quirk.