The fork
Every novel contains the past. The decision is the register in which it is delivered, and the two errors are symmetrical. A flashback used where backstory would do reads as bloated: the reader is made to live through a moment that needed only reporting. Backstory used where a flashback was needed reads as flattened: a moment the reader should have felt arrives as a fact, and when a later scene depends on its weight the dependency fails silently, so that the writer looks for the fault in the later scene. The second error is harder to find because its symptom appears somewhere other than its cause.
Percy Lubbock’s The Craft of Fiction (1921) framed the decision as scene against picture and assumed scene superior, an assumption later hardened into show, don’t tell; this chapter treats the two as a fork, not a hierarchy. In Genette’s measures of duration backstory sits in the summary band (much story time, little text) and flashback in the scene band (the two roughly equal). The Transition (Chapter 5) asks how much text to spend crossing a gap; this chapter asks how much to spend on the far side.
A flashback is a scene, so The Purposes of a Scene (Chapter 1) applies to it in full: someone wants something, meets resistance, and something is different at the end. A past passage with no want and no resistance has no engine, and a flashback without a change of value is the most expensive form of summary there is. Nor is the fork a spectrum: half-dramatised past, a fragment of dialogue in a stretch of précis, has a scene’s length and a summary’s engagement, so push a doubtful passage to one end and see which version the book prefers.
The one-sentence compression check
The one-sentence compression check settles the fork. Compress the cutaway to one sentence of summary and read the surrounding pages with that sentence in its place. If nothing the reader needed to live through is lost, the material was over-dramatised backstory; if the compression flattens something the later pages depend on, the flashback has earned its place. Write the sentence out, because a writer imagining the compressed version imagines it carrying everything the full version carried. Read it in context, two pages on, not in isolation. And expect a third outcome: some passages compress with nothing lost at all, including any reason for their existence, and the verdict then is to cut, which in first drafts is the commonest result.
Fourteen registers
The fork describes only the degree of dramatisation, and a working writer needs a finer repertoire. The Reference chapter sets out fourteen registers, ways of putting the past on the page, which the table compresses to what each buys and what it costs.
| # | Register | Buys | Costs |
|---|---|---|---|
| 1 | Clause of implication | Density at no cost in momentum | Nothing can be felt |
| 2 | Participial aside | Biography without stopping | Needs a narrator licensed to comment |
| 3 | Summary paragraph | Long extents cheaply | Reads as a report if extended |
| 4 | Iterative habitual | A whole era in a sentence; the normal | Cannot carry a singular event |
| 5 | Dramatised flashback | The reader lives it | The most expensive unit; suspends the present |
| 6 | Embedded oral tale | The teller’s bias as content; a listener | Two narrators to sustain |
| 7 | Found document | Authenticity; another diction | Must be plausible as itself |
| 8 | Involuntary memory | The past as something that happens to a character | Unmotivated by plot |
| 9 | Repeating analepsis | Meaning by accumulation | Every pass must visibly differ |
| 10 | Braided timeline | A sustained parallel argument | Both strands must progress |
| 11 | Reconstructed past | Reconstruction as the reader’s activity | Several accounts to hold |
| 12 | Unreachable past | The gap as subject; maximum curiosity | Must be legible as a refusal |
| 13 | False memory corrected | The reader recruited into a misreading | Needs a specific false belief |
| 14 | Prolepsis | The intervening book as a completion | Can neutralise suspense |
Registers 1 to 4 are the underused half: a typical draft delivers its past almost entirely in registers 3 and 5. A useful discipline is to attempt each piece of past material as a clause first and promote it only when the clause fails. Genette’s analysis of frequency separates the singulative (telling once what happened once), the repeating (telling several times what happened once) and the iterative (telling once what happened many times), and the iterative is the register that establishes what was normal, without which a dramatised exception has nothing to be an exception to. Register 8 complicates the rest: Proust’s madeleine has no plot relation to what it recovers, and a rule requiring a motivated trigger would forbid it. Register 12 works only if the reader can tell she is being refused and why, which The Opening Pages (Chapter 9) states as the question whether a gap is in the character or in the author.
The table’s use is substitution testing: rewrite a failing cutaway in three other registers, one much shorter and one much longer, and judge only when all four exist, in place. The most revealing substitution is into the embedded oral tale, because turning past material into something a character says aloud, to someone, for a reason, forces two questions a flashback lets a writer avoid: why now, and why to this person.
Trigger and return
A trigger is the element in the present scene that motivates the cutaway, turning an authorial decision into an event in the story: an object, a sensation, a phrase, a question asked, a place entered, a silence, a date, a document, a resemblance, or nothing at all. The object is the strongest, because it survives the return: a smell is over when it is over, but a pen is still in the hand on the far side, and the return sentence has something to work with. The requirement of a trigger is a consequence of narrative distance, not an independent rule. Close third-person narration has promised to stay inside a consciousness, and an untriggered departure breaks the promise; confident omniscient narration has made no such promise, which is why Tolstoy can simply go. Over-triggering is a fault too: a stopped clock that recalls the day time stopped shows the machinery, and the best triggers are slightly oblique.
The return is where the cutaway is paid for, and its rule is that the present must have changed by the time the narrative comes back to it, in a way the reader can locate. There are four kinds of change, in rising strength: changed knowledge, in which the reader knows something that recolours the scene; changed understanding, in which the character does; a changed question, in which the present scene now carries a question it did not have; and changed action, in which the character does something other than what she was about to do. The changed question is the most generally available and the least used, since it needs no plot event.
Where the multiplying work lives
Revision looks for multiplicity in a more dramatic memory, but most of the work is done in four humbler places. The trigger object exists in both timelines and can carry plot, character, theme and reader knowledge at once if it is already load-bearing elsewhere in the book. Tense and aspect are the cheapest carriers English has: He had always kept it by the till establishes in eight words a history a scene would need three hundred for, and characterises, since always is somebody’s claim. The first sentence after the return is the only one that can work on both timelines at once, and drafts spend it on re-anchoring. And the sentence not written, the one that would state the cutaway’s significance, is the commonest single improvement available, since a connection the reader makes is held more firmly than one supplied.
Most novels can support only a few fully dramatised past scenes before the reader feels that the present is not carrying the book; call that allowance the flashback budget, and spending part of it on material a clause would have delivered is a structural misallocation. Multiplicity also has a ceiling: when every remembered detail is significant and the return changes the present in the designed way and no other, the cutaway reads as a mechanism. A past needs some surplus, detail that is merely true; Roland Barthes’s “The Reality Effect” (1968) argues that the apparently useless detail signifies reality itself.
See also: the Reference register section, with the fourteen registers and their exemplars and the tables of triggers and returns; the Reference multiplicity section, with the cutaway audit.