7.0 How to use the protocol
The ten tests below are for revision, not composition. Applied during drafting they produce paralysis at a scale from which manuscripts do not recover; applied to a completed draft they are the most efficient structural triage available.
Five procedural notes, one more than the other chapters in this book require, because the stakes are larger.
- Run them in order. Tests 1 and 2 are definitional: a thread that is not a chain cannot be assessed for tax, convergence or proportion, because the categories do not apply to it. Tests 3 to 5 are functional. Tests 6 to 8 are quantitative. Tests 9 and 10 are the ones that change books, and they are last because they should be asked only of a thread that has survived the others.
- Record the answer, not the verdict. The output of each test is a sentence describing what the thread does. “Fails test 4” is not usable information; “converges only at the party in chapter 31, and nothing before chapter 27 prepares it” is.
- A failed test is a question. Every test here is failed by at least one novel discussed in this chapter, and §8 supplies the counterexamples deliberately. The test’s job is to make a choice visible so that it can be confirmed as a choice.
- Run the protocol twice, weeks apart. At this scale the first run reports the writer’s intentions and the second reports the text. The discrepancy between the two is itself diagnostic.
- Apply the protocol to every thread, including the main plot. Several of the tests — the kernel test, the page-share audit, the promotion test — only produce information comparatively. A main plot that fails the kernel test is a discovery worth an entire redraft.
7.1 The ten tests
Test 1 — The engine test
Can I state this thread’s own goal, its conflict, and its consequence, in three sentences, without mentioning the protagonist’s goal?
The chain requirement of §2.1, in its operational form. Write the three sentences. If the goal sentence has to be phrased as helps the protagonist to… or makes it harder for the protagonist to…, the thread has no goal of its own and is main-plot machinery distributed across other people — which may be perfectly good writing and is not a subplot.
If there is no goal at all, the diagnosis is presence, not plot (§2.3), and the remedy is usually cheap: the scenes exist; supplying a want and a reversal converts inert page-share into a functioning thread.
Interpretation. This test has three outcomes rather than two. Chain present proceeds to test 2. Chain absent, scenes valuable proceeds to a decision about whether to build a chain or reclassify the material as texture. Chain absent, scenes not valuable proceeds directly to test 9.
Test 2 — The kernel test
Are this thread’s turning points kernels within its own system — moments that open an alternative — or are they satellites even there?
Chatman’s instrument (§3.4). List the thread’s four or five chief moments. For each, ask: at this point, could the thread have gone another way? A kernel is a branching point. If every listed moment is elaboration — a further illustration of a condition already established, a further instance of a character being what they already were — the thread is a sequence of satellites and has no internal causality, whatever its page-share.
Interpretation. The commonest result in a failing manuscript is a thread whose first moment is a kernel and whose remaining moments are satellites: something happens, and then the thread simply persists in the state that something produced. The remedy is not more incident. It is a second branching point, ideally one that reverses the first.
Test 3 — The tax test
What does this thread supply that the main plot structurally cannot supply on its own?
The chapter’s first principle, asked directly. Name the domain or domains from §4. The word structurally is load-bearing and is the whole difficulty of the test: the answer must be something the main plot could not do by its own nature, not something it merely happens not to do.
Two answers that do not count:
- It shows another side of the protagonist. The main plot can do that, by applying different pressure within itself.
- It’s interesting. Interest is a precondition, not a function.
Interpretation. If the honest answer names only pacing relief, apply §4.8’s first proposition: look for the second job the thread could do in the space it already occupies, before considering a cut. If the honest answer is nothing, the thread has failed the chapter’s central standard and the remaining tests are formalities — with one exception, which is that a book without a single primary investment (§3.8) is exempt from this test and should be assessed at §5.4 instead.
Test 4 — The analogy test
What does this thread say about the main plot, and what would be lost if it were not said?
Levin’s relation (§3.2), asked as a demand for content. The test is not whether a relation exists — one always can be constructed — but whether the relation asserts something. Complete this sentence: Because this thread exists, the reader understands that the main plot’s [situation] is [qualified / inverted / costed / limited] in the following way: …
Three failure signatures:
| The sentence comes out as | Diagnosis |
|---|---|
| …is also true of other people | Duplication, not counterpoint (§8.4) |
| …could have gone differently — with no account of how | Decorative rhyme (§4.1) |
| Any sentence the writer can produce but no reader has | Unmarked analogy: a braiding-rhythm failure (§6.6), not a thematic one |
Interpretation. Empson’s caution (§3.3) applies: the relation may work below the level of paraphrase, and a thread that produces a strong effect the writer cannot formulate is not thereby condemned. But the writer should then be able to name where the effect is produced — which pair of scenes, in what order — and if they cannot, the effect is probably an intention rather than a fact.
Test 5 — The convergence test
At what page would deleting this thread break the main plot’s causal logic — as opposed to merely thinning it?
Chatman again, and the convergence test in its sharpest form. Do not ask where the threads meet. Ask where the main plot would stop making sense without this one.
- If the answer is a page number, that is the convergence point, and everything before it is a loan.
- If the answer is nowhere, the convergence is thematic at best, and test 4 becomes the only justification available.
- If the answer is a page number in the final ten per cent, check test 8 immediately: a very late convergence requires very consistent payment (§8.5).
Interpretation. Writers routinely believe convergence occurs earlier than it does, because they are counting the points at which the threads touch rather than the points at which they depend. A thread whose characters appear in the main plot’s scenes for two hundred pages and whose deletion would break nothing has not converged; it has been visiting.
Test 6 — The proportion test
Is the space this thread occupies proportional to its contribution — and, separately, to its position in the book’s hierarchy?
Two questions in one, and they come apart usefully.
Contribution proportion. Compare the thread’s page-share against the domains it serves. A thread serving one domain and taking a fifth of the book is disproportionate whatever its quality.
Hierarchical proportion. Compare the thread’s page-share against its emotional weight in the reader’s experience. The dangerous configurations are the two mismatches: a thread with high page-share and low weight (padding), and a thread with low page-share and high weight (an under-served co-plot, which is a commoner and more interesting fault).
Interpretation. The second mismatch is the one most worth acting on, because the remedy is expansion rather than deletion, and expansion is what writers in structural revision least often consider.
Test 7 — The page-share audit
What percentage of the book does each thread actually occupy, and does the distribution match the book I think I have written?
The quantitative instrument, run as described at §5.5. Count pages in which each thread is the primary business of the scene; convert to percentages; compare against the bands at §2.2.
Three distributions and what they indicate:
| Distribution | Indicates |
|---|---|
| One thread above 60%, others under 10% | A single-plot novel with minor subplots; the attention tax applies with full force |
| One thread 40–55%, one 20–35% | A double-plot novel; Empson’s and Levin’s questions apply, and §4’s hierarchy is only half-true |
| Three or more threads between 15% and 35% | A multi-plot novel; §5.4’s three duties replace the attention tax |
Interpretation. The value of the count is that it is frequently a surprise. Writers systematically overestimate the page-share of threads they enjoyed writing and underestimate the share of threads that grew during drafting. The number is not a verdict; it is a correction to the writer’s model of their own book.
Test 8 — The dormancy audit
What is the longest interval between consecutive appearances of this thread, and what does the resumption cost?
Run as described at §6.6. Note the maximum interval, not the average. Then look at what the manuscript actually does on resumption: count the words spent re-establishing before the scene begins.
Interpretation. A resumption that requires more than a paragraph of re-establishment has revealed its own dormancy problem, and the re-establishing paragraph is the symptom rather than the remedy. Two remedies, in order of preference: insert a short crossing in the gap, which is cheap; or move the thread’s material so that the gap does not occur, which is not.
A second reading of the same data: check whether the thread was left mid-question. A thread suspended at rest costs more to reinstate than one suspended at an open question (roughly twice, by this chapter’s estimate rather than any measurement), and the fix is free.
Test 9 — The removal test
Cut the thread entirely. What is lost, in what domain, and does the loss exceed the pages recovered?
The most brutal instrument in the chapter. Three procedural requirements make it usable rather than merely painful.
- Perform the cut on the outline, not the prose. Delete the thread’s rows from the chapter chart described at §6.5, and read the resulting book’s shape.
- Name the loss in domain terms. “It would be worse” is not a result. “The book would then contain no instance of a person choosing money over a person, so Lydgate’s choice would carry the whole argument alone” is.
- Then ask the harder question: could the loss be supplied more cheaply? Frequently a thread’s entire contribution is one relation that a single scene elsewhere could carry. This is the most common productive outcome of the test, and it is the outcome writers most resist, because it converts eighty pages into one scene.
Interpretation. If the honest answer is that nothing identifiable is lost, the thread should go, and the writer should expect the cut to be far more painful than the diagnosis suggests. If the answer is that the book’s logic fails — not its richness, its logic — the thread has converged and the test is over.
Test 10 — The promotion test
Would the book be better if this thread were the main plot?
Woloch’s question, asked of a manuscript (§3.5). It is placed last because it is the only test that can invalidate the answers to all the others, and because it should be asked of a thread that has already been shown to work.
The test is answered comparatively, in four parts:
- Which thread has the most at stake for the person inside it?
- Which thread’s ending would cost the reader more to lose?
- Which thread has been growing across drafts?
- Which thread do you re-read?
Interpretation. Three outcomes.
The main plot wins. Proceed; the book’s hierarchy is correct and the subplot’s role is confirmed.
The subplot wins on one or two counts. The likely diagnosis is an under-served co-plot (test 6), and the remedy is proportional rather than structural.
The subplot wins on all four. This is the discovery the protocol exists to produce, and it is uncomfortable. Two responses are legitimate and one is not. Legitimate: promote the thread and restructure the book around it (§6.4, structure 9). Legitimate: recognise that the book is a double-plot novel and stop ranking. Not legitimate: leave the arrangement as it stands and hope the reader’s investment follows the writer’s intention rather than the writer’s interest. It rarely does; the result is §8.3.
7.2 Reading the results
| Failed tests | Likely diagnosis | First move |
|---|---|---|
| 1 alone | Presence, not plot (§2.3) | Supply a goal and a reversal, using scenes that already exist |
| 1 and 2 | Texture wearing a subplot’s page-share | Reclassify as motif or runner; cut to the size of what it is |
| 2 alone | A chain with one branching point and a long tail of consequence | Add a second reversal, ideally inverting the first |
| 3 alone | The thread is competent and functionless | Look for the second domain before cutting (§4.8) |
| 3 and 4 | Filler; the thread neither does nor says anything | Cut, and check whether its one useful relation can be supplied in a scene |
| 4 alone | Relation intended, not delivered | Usually a braiding problem: bring the compared scenes nearer (§6.6) |
| 5 alone | Parallel play (§8.1) | Establish thematic convergence deliberately, or admit the multi-plot structure |
| 5 and 8 | Late convergence unpaid for (§8.5) | Braid the thread into the preceding quarter, or move the convergence earlier |
| 6 and 7 | Proportion mismatch | If share is high and weight low, cut; if share low and weight high, expand |
| 8 alone | Dormancy | Insert a short crossing; leave the thread mid-question next time |
| 9 alone | Nothing is lost by removal | Cut, and expect to grieve |
| 10 | The wrong book has been written | Promote, or re-conceive as a double plot; do not leave as is |
A warning about test 10. When the promotion test returns an uncomfortable answer, writers instinctively defend the existing hierarchy — because the main plot is the more finished object, has been outlined longest, and is what the book was pitched as. In this chapter’s judgement it is usually the wrong instinct. The subplot that keeps growing is information. It is the manuscript telling the writer where its energy is, and the energy is not usually wrong.