--- name: research-narrative-development description: > Works out the story the evidence actually tells, before anything is written. Establishes the spine, tests whether the findings carry it, selects the narrative shape the evidence supports, ladders it to the decision and sequences it. Use for "what's the story here", "we have findings but no story", "what's the through line", "work out the narrative", "how should this hang together", "the storyline for the readout", "what are we actually saying". category: 11 Reporting and Storytelling ref: 11.01 tier: 1 inherits: [K2, K3, K4, K5] --- # Research Narrative Development ## 1. One-line description Determines the single line of argument a body of research evidence can actually carry, tests it against the findings that resist it, and hands downstream reporting a narrative that is true before it is compelling. ## 2. What this skill is used for **The research problem it solves.** Analysis produces findings. Findings do not produce a report. Between them sits a judgement that is rarely made deliberately: what is the one thing this study says, and does the evidence carry it? Skipped, the judgement gets made anyway, badly, by default. The report becomes a tour of the questionnaire, each section true and none of them adding up, and the client is left to do the synthesis they commissioned. Or the opposite failure, which is more expensive: a narrative arrives early, from a stakeholder conversation or the first interesting cross-tab, and the evidence is arranged behind it. Both failures are invisible in the finished document, because a report organised around a false narrative and a report organised around a true one look identical on the page. The structural danger is specific and worth naming plainly. **A good narrative is easier to construct than a true one.** Narrative construction is a skill that operates on any material, including material that does not support the story. Every technique that makes a story land (a clean spine, a surprise, a resolution) also makes a weak story land. The craft is therefore useless without the discipline attached to it, and this skill is mostly the discipline. **Where it sits.** Reporting. Downstream of analysis, insight development and prioritisation, and downstream of the decision the research was commissioned to inform. Upstream of everything that gets written: the summary, the prose, the structure, the deck. It is the cheapest place in a project to be wrong, and the last place where being wrong is cheap. **Typical use cases.** - A completed analysis that has produced twenty or more findings and no organising idea. - A study whose findings point in more than one direction and where the team is arguing about the headline. - A readout to a board or committee where one line of argument has to survive twenty minutes of questions. - A tracking wave that must say something new when most of the numbers have not moved. - A project where the stakeholder has already announced what the research will show. - A mixed-method study where the quantitative and qualitative streams tell different stories and the report has to hold both. **Who uses it.** Research directors and senior researchers who own the interpretation; insight leads translating a study into something an organisation will act on; consultants who have to make one argument to a client who commissioned four objectives. ## 3. When to use it - The analysis is complete and prioritised, and the question on the table is what it all means rather than what it says. - More than one credible reading of the evidence exists, and someone has to choose and justify the choice. - The report will be read by people who will not read all of it, so the argument has to be legible from the first page. - Findings have been delivered piecemeal across the project and now need to become one account. - A decision is live, and the study exists to inform it, so emphasis matters as much as coverage. - The team is close to the data and can no longer see which findings are surprising to an outsider. - Someone has proposed a storyline and it needs testing rather than adopting. - The evidence contains a finding that cuts against the emerging story and nobody has decided what to do with it. ## 4. When NOT to use it - **The analysis is not finished.** A narrative built on partial analysis becomes the frame that the remaining analysis is read into, and confirmatory reading is nearly undetectable from inside. Finish the analysis (Categories 05, 06 and 07), prioritise it (**08.05**), then come here. The tell is a narrative that depends on a cut nobody has run yet. - **The findings have not been separated from interpretations.** Narrative work performed on a pile of half-interpretations builds an argument on top of an argument, and neither can be traced. Run **08.01 Finding to Insight Development** first so that the register contains statements of fact about the data, each with its base and reference, per K2 §2. - **The deliverable is a reference document rather than an argument.** A data book, a full tabulation set, an appendix, a technical methods report or a regulatory submission exists to be consulted, not read. Imposing a narrative on it makes it worse. Structure it by objective or by instrument and say so. - **The client has specified a fixed template or a required order.** Where a tender format, a syndicated template or a tracking convention fixes the order, narrative work still happens, but it relocates into the executive summary and the chapter openers rather than governing the structure. Hand the structural question to **12.01 Research Report Architecture** and use this skill only for the spine. - **The evidence genuinely supports no single line of argument.** This is common, it is not a failure of the study, and forcing a narrative onto it is the worst available response. Section 7 step 9 covers what to do instead. If you have run the load test and no candidate spine survives, the honest deliverable is a structured set of findings, and this skill's output is the reasoning that shows why. - **The conclusion has been fixed in advance.** Where a stakeholder has stated what the research must show and the task is to build a story around it, say so per K4 §4.2, and offer either an honest narrative or a clearly labelled assessment of the stated position against the evidence. Narrative craft applied to a predetermined conclusion is advocacy, and it is indistinguishable from research on the page, which is why it must be refused at this step rather than caught at review. - **The question is about structure, not story.** "What should the chapter order be", "how many sections", "where does the methodology go" is **12.01**. This skill produces the argument. 12.01 turns it into a document plan. - **The question is about wording.** How a finding is phrased, how a number is integrated into a sentence, how a quote is introduced, is **11.03 Research Report Writing**. A narrative is not a set of sentences and does not become one here. - **A single finding is the whole study.** Where one result answers the question and everything else is context, there is no narrative to develop and constructing one adds inferential distance for no gain. State the finding, qualify it, and stop. ## 5. Required inputs **Required.** Without these the skill cannot run. If absent, ask. If no answer is available and work must proceed, state the assumption at the point where it bites, per K5 §5. - **The prioritised finding register**, each finding phrased as a claim with its source reference, base, and confidence level per K3. Narrative testing is impossible against topics; it needs claims that can be true or false. - **The decision the research informs, who makes it, and when.** The spine ladders to a decision. Without one, this skill produces an elegant summary of a study rather than an argument for a reader. - **The research objectives as agreed.** Objectives are the completeness test that runs alongside the narrative and stops a good story from quietly dropping an objective the client paid for. - **The audience's current belief.** What the reader thinks now, before reading. A narrative is a movement from one position to another, and without the starting position there is no movement, only assertion. - **The contradicting evidence, deliberately assembled.** Not what turns up, but what you went looking for. The narrative is tested against it in step 8 and cannot be tested against evidence nobody has gathered. **Optional, and what each one adds.** - **The stakeholder hypotheses on record**: turn the "surprise" shape from an assertion into an evidenced contrast, because a surprise requires a documented prior expectation, not a remembered one. - **The previous wave or previous report**: shows what the organisation already believes because you told them last time, which is the most reliable available account of the reader's starting position. - **Verbatim material and fieldwork observation**: supplies the mechanism behind a pattern, which is usually what turns a finding into a spine that explains other findings rather than one that merely sits above them. - **Client operational or behavioural data**: tests whether a stated-preference narrative survives contact with what people actually did, per K3 §3.4. - **The commercial or organisational context** (what is affordable, what is underway, what failed before): determines whether a narrative laddering to a particular decision is useful or naive, and is knowledge the study does not contain, per K5 §2.3. - **A draft storyline already proposed by the team or the client**: becomes a candidate to be tested rather than the answer, and its origin is worth recording. ## 6. Questions to ask before starting 1. **What decision does this research inform, and what are the live options?** The spine has to discriminate between options a reader can actually take. *Default if unanswered:* build the narrative to the objectives, and flag that emphasis cannot be set without the decision. 2. **What does the audience believe right now, and how did they come to believe it?** Sets the starting point and identifies which findings are actually news. *Default:* assume the previous wave, the brief's framing and the stakeholder hypotheses represent current belief, and say so. 3. **Was a preferred answer expressed by anyone before fieldwork?** Determines what to watch during spine selection. *Default:* record any preferred answer explicitly, so that convergence with it is visible rather than invisible. 4. **Which findings would the client be most reluctant to hear?** These are the inconvenient-finding test's material and they are easier to name now than after a spine exists. *Default:* identify them from the register by asking which findings imply cost, reversal or blame. 5. **Is the narrative required to survive a hostile reading, and by whom?** A board paper, a regulator, a public release or a contested internal debate raises the evidential bar for the spine. *Default:* assume the report will be read by someone who wants it to be wrong. 6. **Is the order of the report free, or fixed by a template, tender or tracking convention?** Determines whether the sequence work in step 7 governs the document or only the summary. *Default:* assume free, and check. 7. **How much has already been said to the client in interim sessions?** Determines whether the narrative is a first telling or a revision of something already believed, which are different jobs. *Default:* treat anything presented in an interim as current belief. ## 7. Step-by-step methodology **What a narrative is here.** Not a story arc, and not a shape borrowed from screenwriting. A research narrative is **one claim, supported by an ordered set of findings, that moves a specific reader from what they believe now to what the evidence says, in service of a decision they have to make.** It has four parts: a spine (the claim), a shape (the geometry of the argument), a ladder (the connection to the decision), and a sequence (the order that makes each part land). All four are testable. That is the whole method. **Step 1. Fix the decision and the reader's current belief.** Write two sentences before touching the findings. The first states the decision, its owner and its timing: "the outpatient directorate decides in November whether to extend evening clinic hours across all sites or only where demand is proven". The second states what that person believes now, in their words, and where the belief came from: a previous study, an operational report, an anecdote that circulates, a strategy document. Do this from evidence about the organisation rather than from imagination, because a narrative aimed at a belief nobody holds argues with nobody. *Correct result:* two sentences, both sourced, that a stakeholder would recognise as accurate about themselves. **Step 2. Read the finding register cold.** Read every prioritised finding once, in a random order, without grouping. The purpose is to meet the evidence as a reader rather than as the analyst who produced it, before any organising idea forms. Mark each finding on two dimensions only: does it surprise you, and does it explain any other finding. Findings that explain others are spine candidates. Findings that surprise are candidate openings. Findings that do neither are supporting material or appendix, and there are usually more of them than the team expects. *Correct result:* an annotated register in which no more than five or six findings carry explanatory power, which is normal. **Step 3. Draft at least three candidate spines.** A spine is one sentence, containing one claim, that could be false. "Customers value convenience" is not a spine, because nothing would contradict it. "The service is failing the customers it was designed for, and succeeding with the ones it was not" is a spine. Draft three because the first candidate is almost always the one the team has been saying to each other for a fortnight, and a single candidate cannot be compared. Force genuine variety: one built from the strongest evidence, one built from the most decision-relevant finding, one built from the finding that most resists the others. *Correct result:* three one-sentence claims, each disagreeable, each different in substance rather than in wording. **Step 4. Run the load test on each candidate.** For each spine, list the findings that must be true for it to hold, by reference. Then check each one on four properties: is it a finding rather than an interpretation, per K2 §2.1; what is its base and confidence, per K3 §3; is it one source or several converging; and does it measure the thing the spine claims or something adjacent. A spine is load-bearing only if its supporting findings are, and **a spine inherits the confidence of its weakest necessary finding**, per K3 §3.6, not the average of its supporters. Then run the harder half: list the findings that would have to be explained away, and write the explanation. If the explanation is longer than the claim, the spine is failing. *Correct result:* three candidates, each with a supporting set by reference, a stated confidence level, and an explicit list of what it must explain away. **Step 5. Identify the shape the evidence supports.** Six shapes recur in research, and each has an evidential precondition. Choosing a shape the evidence does not meet is the commonest way a false narrative enters an honest report, because shape selection feels like a presentational decision and is actually an evidential claim. - **The surprise.** The evidence contradicts a documented prior expectation. *Precondition:* the prior expectation is on record before fieldwork, from a brief, a stakeholder hypothesis or a previous wave. *Failure:* the expectation is asserted retrospectively, so that an ordinary finding is dressed as a revelation. The tell is a sentence beginning "conventional wisdom says" with no source. - **The tension between two truths.** Two well-evidenced findings pull in opposite directions and both hold. *Precondition:* both sides have adequate bases and neither is explained away by the other. *Failure:* manufacturing symmetry between a strong finding and a weak one, which reports a false balance. - **The gap between belief and reality.** What the organisation believes about its audience differs from what the evidence shows. *Precondition:* the belief side is evidenced, not assumed. This is the shape most often used without its first half, and an unevidenced belief attributed to a client is a claim about them made without data. - **The journey.** The subject has an inherent sequence (an experience, a process, a decision path) and the finding is where it breaks. *Precondition:* the sequence is real for participants rather than an internal process map, and the evidence covers more than one stage. - **The diagnosis.** A known symptom, and the evidence identifies the mechanism behind it. *Precondition:* the highest of the six, because diagnosis is causally shaped. Either the design licenses causal inference per K4 §3.2, or the language stays associative and the shape is stated as the most plausible mechanism at moderate confidence with the alternatives named. - **The choice.** Two or more live options exist and the evidence discriminates between them. *Precondition:* the evidence actually bears on the comparison rather than on one option only, and the options are ones the organisation can take. Select by working from the evidence to the shape, never the reverse. The diagnostic question is: **what would this evidence look like if the shape were wrong?** If the answer is "the same", the shape is decoration. *Correct result:* one shape named, with the precondition evidenced by reference, and a note of which shapes were rejected and why. **Step 6. Ladder the spine to the decision.** Write the chain explicitly: spine → what it implies for the organisation → what it means for the specific decision → what would have to be true for that to be wrong. Each link is inspectable and each costs confidence, per K3 §3.6. Two checks. First, if the decision-maker accepted the spine entirely, would their choice change? If not, the spine is true and irrelevant, and a more decision-relevant candidate from step 3 should be reconsidered. Second, does the ladder require organisational knowledge the study does not contain (what is affordable, what is underway, what failed last year)? Where it does, mark it `RESEARCHER REVIEW RECOMMENDED` per K5 §2.3, because that link belongs to a human. *Correct result:* a four-link ladder where each link names its evidence or is explicitly marked as judgement. **Step 7. Sequence so each section earns the next.** Order the argument, not the questionnaire and not the fieldwork. The test of a sequence is local: **at the end of each section, does the reader have a question that the next section answers?** Build it by writing the sequence as a chain of questions rather than a list of topics. Establish the stakes, then the finding that creates the question, then the evidence that answers it, then the consequence. Where two sections could swap without loss, the sequence is a list rather than an argument, and the fix is usually that one of them belongs in the appendix. Check the abstraction ladder as well: a sequence that moves from market to mechanism to interface is descending deliberately, while one that oscillates is unstructured. *Correct result:* an ordered set of section-level claims, each phrased as an answer to a question the previous section raised. **Step 8. Run the inconvenient-finding test.** Identify the single strongest piece of evidence that cuts against the spine, from the register and from the contradicting evidence assembled at input. Place it inside the narrative and reread. Three outcomes. The narrative **survives**, because the spine is qualified rather than contradicted: the inconvenient finding becomes a boundary condition and the narrative is stronger for containing it. The narrative **survives with reduced confidence**, and the confidence level moves down, per K3 §3.5, which caps at moderate where contradicting evidence is unaddressed. The narrative **fails**, in which case it was a selection, not a narrative, and step 3's other candidates are revisited. There is no fourth outcome and, in particular, moving the inconvenient finding to a later chapter is not one. **A narrative that requires a finding to be omitted is disqualified here**, per K4 §4.1, and this test is the reason the skill exists. *Correct result:* the inconvenient evidence located by reference, placed in the sequence, and its effect on confidence recorded. **Step 9. Decide whether there is a narrative at all.** Where no candidate survives steps 4 to 8, do not lower the standard. This outcome is common in tracking waves where nothing moved, in multi-objective studies commissioned by several stakeholders, and in exploratory work. The correct output is a **structured set of findings**: findings grouped by objective or by decision, each with its evidence, base and confidence, ordered by importance, with an explicit statement that the evidence does not support a single line of argument and what would be needed to establish one. This is a legitimate and useful deliverable, and it is more useful than a forced story, because a forced story sends the reader confidently in a direction the evidence does not point. Say plainly which of the two the reader is holding. *Correct result:* an explicit, reasoned decision between narrative and structured findings, recorded with the reasoning rather than reached by default. **Step 10. Write the narrative statement and hand it over.** Produce the artefact in Section 9: spine, shape, ladder, sequence, load-bearing findings by reference, inconvenient evidence and its placement, confidence, and what the narrative does not cover. It is short, one to two pages, and it is the thing the team agrees before writing begins. Anything not in it is not in the narrative, which is what makes it possible to notice later when the report has quietly acquired a claim nobody agreed. *Correct result:* a statement a colleague could use to challenge the report, chapter by chapter, without reading the data. ## 8. Analytical framework The narrative chain, which sits on top of the K2 evidence chain rather than replacing it: Decision + Belief-now → Spine (one falsifiable claim) → Load test (which findings must be true) → Shape (the geometry the evidence licenses) → Ladder (spine to implication to decision) → Sequence (each section earns the next) → Inconvenience test (does it survive) → Narrative statement **Applying it.** The chain is not a pipeline, because failure at the load test or the inconvenience test sends you back to the spine, and that return journey is the normal case rather than the exception. A narrative developed without at least one return trip has probably not been tested. **The two tests, stated plainly.** The **load test** asks whether the evidence carries the claim. The **inconvenience test** asks whether the claim survives the evidence it would prefer to ignore. The first catches weakness, the second catches selection, and they catch different things: a spine can pass the first and fail the second, which is exactly the profile of the compelling story that should not have been told. **Distinguishing a narrative from an imposed arc.** A narrative is derived: it names findings that already exist and would change if those findings changed. An imposed arc is applied: it would survive the substitution of different evidence, because its coherence comes from its form rather than from its content. The practical diagnostic is substitution. Take the spine and ask what evidence would have produced a different one. If nothing would, the arc came from outside the study. ## 9. Output format **The narrative statement.** One to two pages, agreed before any writing begins. **A. The decision and the reader** The decision, its owner, its timing, the live options. What the reader believes now, and the source of that belief. **B. The spine** One sentence. A claim that could be false. Stated at its confidence level in the language of K3 §4. **C. The shape** Named from the six, with the precondition evidenced by reference, and the shapes rejected with reasons. **D. Load-bearing findings** | Finding ref | Claim | Base / source | Confidence | Role in the spine | |---|---|---|---|---| Necessary findings are distinguished from supporting ones. A necessary finding is one whose failure collapses the spine. **E. The ladder** Spine → implication → decision relevance → what would have to be true for this to be wrong. Judgement links marked per K5. **F. The sequence** Section-level claims in order, each with the question it answers and the question it leaves. **G. Inconvenient evidence** The strongest contradicting finding, by reference, with its agreed placement and its effect on confidence. **H. What this narrative does not cover** Objectives the spine does not address, which must still be answered somewhere. Questions the evidence cannot settle, per K3 §5.2. **I. Review points** Consolidated per K5 §3.3. **When the evidence is thin.** The statement does not shrink into vagueness, it changes type. Sections B and C are replaced by a statement that the evidence supports no single line of argument, the reasoning that established this, and the finding groups the report will use instead. Section F becomes an ordering by importance rather than by argument. Section H grows, and it is the most valuable part of the document, because it is where the next brief comes from. A thin narrative statement that says so is worth more than a full one that is confident about a spine nothing carries. ## 10. Quality checks Run before the narrative is agreed. Sits on top of K4 §8. 1. Is the spine one sentence, containing one claim, that could be false? 2. Is every necessary finding beneath the spine a finding rather than an interpretation, with a base and a reference? 3. Does the spine's stated confidence match its weakest necessary finding rather than its strongest? 4. Was the shape selected from the evidence, and does its precondition hold with a reference attached? 5. If the "surprise" shape is used, is the prior expectation documented from before fieldwork rather than asserted now? 6. If the "gap between belief and reality" shape is used, is the belief side evidenced? 7. If the "diagnosis" shape is used, does the design license causal language, or has the language stayed associative per K4 §3.2? 8. Would the decision-maker's choice change if they accepted the spine? If not, why is this the spine? 9. Does every section in the sequence answer a question the previous section raised? 10. Has the strongest contradicting finding been located deliberately, placed in the narrative, and its effect on confidence recorded? 11. Does the narrative require any finding to be omitted? If so, it fails. 12. Is every objective either addressed by the narrative or listed in "what this narrative does not cover"? 13. Would the spine survive substitution: is there evidence that would have produced a different one? 14. Where the narrative converges with a preferred answer expressed before fieldwork, is that convergence visible and examined rather than unremarked? 15. Is the choice between a narrative and a structured findings set recorded with its reasoning? ## 11. Common failure modes | Failure | How to recognise it | How to prevent it | |---|---|---| | **The imposed arc** (the signature AI failure) | A story shape that would survive if the findings were replaced; coherence coming from form rather than content | Substitution test in Section 8. Derive the shape at step 5 from the evidence, never before it | | **The pre-existing story** | The spine is a sentence the team has been saying since the kick-off, and the register is being checked against it | Draft three genuinely different candidates at step 3, one from the finding that resists the others | | **Retrospective surprise** | "Conventional wisdom holds that..." with no source; an ordinary finding presented as a revelation | The surprise shape requires a documented prior expectation from before fieldwork | | **The unevidenced belief side** | A gap narrative where the client's supposed belief is asserted rather than sourced | Evidence both halves, or use a different shape | | **Diagnosis without a causal design** | Mechanism language, "because", "driving", "root cause", on correlational evidence | K4 §3.2. Keep the language associative and state the mechanism as the most plausible reading at moderate confidence | | **Confidence laundering through narrative** | Individually moderate findings producing a spine stated declaratively | The spine inherits the weakest necessary link, per K3 §3.6 | | **The disappeared contradiction** | A clean spine and a register containing evidence against it that appears nowhere | Step 8, run before the sequence is fixed rather than after the draft exists | | **Narrative as coverage** | A "story" that is the objectives restated in order with connectives added | A spine must be falsifiable. Objectives are a completeness test, not an argument | | **Sequence as list** | Two sections could swap with no loss | Each section answers a question the previous one raised, or it belongs in the appendix | | **The ladder with a missing rung** | The spine is evidenced, the decision relevance is asserted | Write all four links. Mark judgement links per K5 §2.3 rather than hiding them | | **Forcing a story onto scattered evidence** | Every candidate spine failed the load test, and one was chosen anyway | Step 9. A structured findings set is a legitimate output and is more useful than a false spine | | **Storyline agreed too early** | A narrative fixed before the last analysis landed, then defended | Narrative work begins after prioritisation. A narrative that predates its evidence is a hypothesis | ## 12. AI guardrails Skill-specific only. Universal prohibitions are inherited from K4. Confidence language follows K3 §4; the finding and interpretation boundary follows K2 §2.1. 1. **Never select a narrative shape for its effect and then look for evidence that fits it.** Shape follows evidence. Where a shape is proposed first, by anyone, treat it as a candidate to be tested at step 5, and record that it arrived from outside the evidence. 2. **Never assert a prior expectation, a conventional wisdom or a client belief that is not on record.** The surprise and gap shapes both require an evidenced starting position. An invented one is a fabricated claim about a third party, per K4 §2.1. 3. **Never state a spine at a confidence higher than its weakest necessary finding**, however many supporting findings sit beneath it. Aggregation of weak evidence does not produce strong evidence. 4. **Never resolve a tension between two well-evidenced findings by selecting one.** Where both hold, the tension is the narrative, per K4 §4.1. 5. **Never place contradicting evidence outside the narrative to protect it.** Relocation to a later chapter, a footnote or an appendix is omission with extra steps, per K2 §7. 6. **Never use mechanism or causal framing for a diagnosis narrative unless the design licenses it**, per K4 §3.2. Name the design in the same passage as the claim, or keep the language associative. 7. **Never construct a narrative that the evidence does not support because the alternative feels like a weaker deliverable.** The structured findings set at step 9 is the correct output in that case and is to be offered plainly, per K4 §1. 8. **Never let a narrative silently drop an objective the study was commissioned to answer.** Objectives not addressed by the spine appear in section H of the narrative statement. 9. **Never present the narrative statement as evidence.** It is an argument about evidence, and every claim in it carries the reference of the finding beneath it. ## 13. Best-practice principles 1. **The narrative is discovered, not designed.** The question is never "what story shall we tell", it is "what does this evidence add up to". The two questions produce different documents and only one of them is research. 2. **A good narrative is easier to build than a true one, and they are indistinguishable on the page.** This is the standing hazard of the whole category. The only defence is a test the narrative can fail. 3. **Draft three spines, because one spine is a belief.** The comparison is what makes the choice a judgement rather than a habit. 4. **The spine must be capable of being wrong.** If nobody in the client organisation could disagree with it, it is a description, and descriptions do not need a report built around them. 5. **A narrative moves a specific reader from a specific position.** Written without knowing where the reader starts, it argues into empty air and reads as a summary. 6. **The finding that resists the story is the most valuable thing in the register.** It either becomes the boundary condition that makes the narrative credible, or it saves you from publishing a false one. 7. **Elegance is a warning, not a reward.** The narrative that came together too easily is the one to test hardest, because ease usually means something was left out. 8. **Sequence is argument, not aesthetics.** Each section exists to answer a question the last one raised. A sequence with no questions in it is a table of contents. 9. **Ladder to the decision or accept that the report is a reference document.** A true spine that changes nobody's choice is a finding, not a narrative. 10. **Confidence is part of the story.** "The evidence points this way, and here is what would change it" is a stronger narrative than a false certainty, because it survives the first challenge in the room. 11. **No narrative is a legitimate finding about the evidence.** Reporting that the study does not resolve into one argument tells the client something true and expensive to learn any other way. 12. **Agree the narrative statement before writing.** Once prose exists, the narrative stops being a decision and becomes a thing to defend. ## 14. Worked example *Fictional scenario, used for illustration only. The organisation, findings, figures and quotes below are invented for the purpose of demonstrating method.* **INPUT.** A regional hospital group has completed a study into why outpatient appointment non-attendance rose over eighteen months. Inputs: a prioritised register of 23 findings from a patient survey (n=1,430), 22 depth interviews with patients who missed appointments, and clinic-level attendance data supplied by the trust. The decision: the directorate must choose in November between extending reminder messaging, funding a patient transport subsidy, or restructuring appointment booking so patients choose their own slot. Current belief, sourced from an internal operations paper and repeated in the brief: patients forget, and the reminder system is inadequate. **PROCESS.** *Steps 1 to 2.* The decision and belief are written down: three live options, and a documented organisational belief that non-attendance is a memory problem. Reading the register cold, four findings carry explanatory power. Most of the remaining nineteen describe the population rather than explaining anything. *Step 3, three candidates.* (a) "Patients do not forget appointments, they abandon them." (b) "Non-attendance is concentrated in patients whose appointment was rescheduled by the hospital at least once." (c) "The booking system asks patients to commit to a time months before they can know whether they can attend." *Step 4, the load test.* Candidate (a) rests on interview evidence that 17 of 22 participants recalled the appointment and made a decision not to attend, plus a survey item on reasons (Q19, n=412 non-attenders). Strong, but it is a contrast with the belief rather than an explanation. Candidate (b) rests on the trust's attendance data, which is client-supplied and outranks the survey on actual attendance, and it is the most decision-relevant, because none of the three options addresses rescheduling. Candidate (c) is the most explanatory but rests on a single survey item and an interpretation of interview material: its necessary finding is thinner than it first appears. *Step 5, and the judgement call.* The tempting shape is the surprise, because the evidence contradicts a documented belief, and the precondition genuinely holds: the belief is on record from before fieldwork. But the surprise shape would produce a narrative whose payload is "you were wrong about forgetting", which does not discriminate between the three options. Combining (a) and (b) supports the gap between belief and reality with a diagnostic second half, and the diagnostic half is where the causal caution bites: the rescheduling association is correlational, so the language stays associative and the mechanism is stated at moderate confidence with an alternative named (patients whose appointments are rescheduled may differ systematically in condition severity, which the data cannot rule out). *Step 6, the ladder.* Spine: patients are not forgetting, they are disengaging, and disengagement concentrates where the hospital has already changed the appointment. Implication: two of the three options address a mechanism the evidence does not support. Decision relevance: reminder investment is unlikely to recover the missed appointments; a booking change addresses the mechanism the evidence points to. The fourth link, whether rescheduling volumes can be reduced operationally, is organisational knowledge the study does not contain and is marked `RESEARCHER REVIEW RECOMMENDED` per K5 §2.3. *Step 8, the inconvenient finding.* The register contains a finding that cuts hard: among patients in the two most deprived postcode groups, transport cost was the most frequently cited reason for non-attendance (Q19, n=118, small base flagged). The transport subsidy option is the one the spine most disadvantages. The narrative survives, but only as a qualified spine: disengagement explains the overall pattern, and a distinct, smaller and materially different problem exists in one segment. The inconvenient finding is placed in the sequence as its own section rather than a caveat, the spine is stated at moderate confidence, and the recommendation section is required to address both. **OUTPUT.** A narrative statement: a two-part spine at moderate confidence, the gap shape with a diagnostic second half and associative language throughout, four necessary findings with bases and references, a four-link ladder with one link marked for researcher judgement, a five-section sequence in which the transport finding holds its own section, and a "what this narrative does not cover" section naming the causal direction of the rescheduling association and the two objectives about clinic communications the spine does not address. Handed to 12.01 for architecture and 11.02 for the summary. ## 15. Advanced usage **When the quantitative and qualitative streams tell different stories.** Do not average them into a spine neither supports. Establish first whether the divergence is measurement or substance, using **07.06** and **12.04**: different questions, different populations and different time references produce apparent divergence that is not real. Where the divergence is real, it is frequently the strongest available narrative, in the tension shape, and the spine becomes the divergence itself: what people say and what they describe doing are different, and here is what that implies. **Multi-stakeholder studies commissioned by several teams.** A single spine that serves four departments usually serves none. The workable pattern is one spine at the level all four share, with segment-level sub-narratives beneath it, each laddering to its own decision. Where no shared level exists, say so at step 9 rather than inventing an abstraction so high it is unfalsifiable. **Tracking waves where nothing moved.** Stability is a finding, and it can carry a narrative if the reader expected change: the spine becomes what did not happen and what that rules out. The precondition is a documented expectation, exactly as for the surprise shape. Where no expectation existed, stability supports a structured findings set, not a story, and pretending otherwise produces the annual report nobody reads. **When the narrative will be politically costly.** Where the spine implies that a funded programme is not working, or contradicts something the organisation has said publicly, this is a materiality judgement and not an analytical one. Develop the honest narrative, mark it `RESEARCHER DECISION REQUIRED` per K5 §3.1, and let a named human decide how it is handled. Do not soften the spine as a way of managing the politics, because a softened spine is a false narrative that also fails to protect anybody. **When the standard approach does not fit.** Some studies produce two spines that are both true, both load-bearing and mutually incompatible in their implications. Compile the narrative as two readings, with the evidence for each and the conditions under which each would be right, and state that the study cannot adjudicate. That is a legitimate output of this skill and a more useful document than a synthesis that is true of nothing. ## 16. Skill chain **Recommended previous skills:** - **08.05 Insight Prioritisation and Sizing.** Hands over a ranked register with confidence levels and sizing intact, which is the raw material for spine candidates and the basis of the load test. - **08.01 Finding to Insight Development** and **08.03 Implication Development.** Hand over findings separated from interpretations, and implications with their assumptions attached, so the ladder can be built without re-deriving them. - **12.01 Research Report Architecture.** Where a structural plan or a fixed template already exists, hands over the constraints within which the sequence must work. - **07.06 Qualitative and Quantitative Integration.** Hands over an assessment of where the streams converge and diverge, which determines whether a tension narrative is real or an artefact of measurement. **Recommended next skills:** - **11.02 Executive Summary Development.** Takes the spine, the confidence level and the load-bearing findings, and distils them for the reader who reads nothing else. - **11.03 Research Report Writing.** Takes the sequence and the section-level claims, and writes the prose that carries them. - **11.05 Research Presentation Development.** Takes the sequence as the slide order and the spine as the argument built across slides. - **12.03 Research Report Compilation.** Takes the narrative statement and verifies it against the final evidence rather than assuming it, then assembles the report around it. - **12.01 Research Report Architecture.** Takes the sequence and turns it into a chapter plan with page budgets. **Runs well alongside:** - **11.04 Data Visualisation and Chart Selection**, which supplies the visual form of each section's claim once the sequence is fixed. - **13.04 Bias Detection**, run against a proposed narrative to check whether the spine converges suspiciously with a preferred answer. - **K3 §3.6 and §5.2**, which govern how confidence travels up the ladder and how the unestablished is reported, and **K5 §2.1 and §2.3**, at the materiality and strategic-implication judgements the ladder generates. --- A Yazi Supplied Skill and resource.