# Narrative Evidence Method — FULL RESEARCH CORPUS Abbreviation: nem Version: v0.1.0 | Generated: unspecified | Files: 1 | Lines: 292 | Payload: 15.3 KiB Source path: /Users/zachshallbetter/Projects/_protocols/narrative-evidence-method Bundle name: llm-full-docs-nem Corpus Payload SHA-256: 830bec0b033d692850e9afb12fd1aa12e9a8d2e062426759d7db6625dddb850c TABLE OF CONTENTS: - NEM.md (292 lines, 15.3 KiB) ======================================================================== ======================================================================== ## FILE: NEM.md (292 lines, 15.3 KiB) ======================================================================== # Narrative Evidence Method **Identifier:** `nem:protocol/narrative-evidence` **Version:** `0.1.0` **Status:** Working Draft **Date:** 2026-08-30 **Author:** Zach Shallbetter **Primary use:** *Distance from Gray* **Companion artifacts:** `app:protocol/authored-prose` (voice and conformance), `eqp:protocol/evidence-qualification` (what the evidence justifies), `Distance-From-Gray/docs/Manuscript/` (the manuscript this governs) --- ## 0. Abstract A method for evidence-heavy trade nonfiction in which claims are developed through concrete observation, unresolved questions, evidence, interpretation, and consequence. Its purpose is to make rigorous evidence read as discovery rather than documentation. It separates two concerns that are usually confused. **Reader experience**, where the argument should feel like an investigation. And **evidence integrity**, where every material factual claim must remain traceable to qualified evidence. Citation machinery may be removed from the reader's immediate path. Provenance may not be removed from the manuscript. The governing rule: *do not begin with the abstraction when the reader can discover it from the evidence.* **MUST** indicates a conformance requirement. **SHOULD** indicates a default that requires a stated reason to break. **MAY** indicates a permitted option. --- ## 1. The canonical movement ```text SCENE → ANOMALY → QUESTION → EVIDENCE → PATTERN → INTERPRETATION → CONSEQUENCE ``` **Scene.** Begin with something concrete: a person, event, object, process, decision, failure, measurement, or moment. The scene gives the reader something observable before asking them to accept an abstraction. **Anomaly.** Identify the detail that does not behave as expected. It should be specific enough that the reader can recognize the discrepancy without being told what it means. **Question.** Convert the anomaly into the question the section will answer. The question should arise from what has already been shown. **Evidence.** Introduce the strongest appropriate evidence. Evidence is not decoration. It must advance the investigation. **Pattern.** Determine whether the observation appears elsewhere. A second domain, independent case or dataset establishes that the opening event is not idiosyncratic. **Interpretation.** Explain the mechanism that best accounts for the evidence. Interpretation must remain distinguishable from the evidence itself. **Consequence.** Show what follows if the interpretation is correct. --- ## 2. Chapter architecture ```text OPENING SCENE Something happens. ANOMALY Something about it does not make sense. QUESTION The obvious explanation becomes insufficient. FIRST EVIDENCE Research or measurement expands the observation. INTERPRETATION A possible mechanism appears. SECOND CASE A different environment exhibits the same mechanism. COUNTEREXAMPLE A case tests the boundary of the explanation. REFINEMENT The claim becomes narrower and more precise. NAME The mechanism receives a durable name, if needed. CONSEQUENCE The mechanism changes how the problem is understood. RETURN The opening scene is revisited with the new explanation. ``` **A1.** Not every chapter requires every element. A chapter **MUST** nevertheless preserve the movement from observation toward explanation, rather than beginning with explanation and accumulating examples afterward. **A2.** A chapter **MUST NOT** state its conclusion in its opening section where the reader could reach it from the evidence. Announcing the argument before the investigation converts discovery into documentation. --- ## 3. Evidence presentation **E1.** Evidence **SHOULD** enter the prose through narrative attribution rather than academic interruption. Avoid: *AI productivity gains can be substantially offset by rework (Workday, 2025).* Prefer: *In 2025, Workday asked 3,200 employees what happened to the time they saved using AI. Nearly forty percent of it came back as rework.* **E2.** The prose **MUST** carry enough context to interpret the finding: who, what was measured, the relevant population or context, and the result. Full bibliographic and methodological detail belongs in the apparatus. **E3.** Research **SHOULD** be described as action rather than authority. Not *a study found*, but *researchers compared*, *the team followed*, *investigators reviewed*, *the experiment measured*. The researcher, archive or dataset becomes part of the investigation rather than a parenthetical endorsement. --- ## 4. Two-layer citation Every factual passage exists in two layers at once. **Reader layer.** Clean narrative attribution. **Evidence layer.** Complete provenance retained in the manuscript: `claim_id`, `claim_lane`, `source`, `author_or_organization`, `publication_date`, `study_date`, `sample`, `method`, `exact_finding`, `qualification`, `url_or_doi`, `access_date`, `relevant_passage`, `notes`. **C1.** The reader does not need to operate the evidence layer. The manuscript does. **C2.** Low citation visibility **MUST NOT** mean low citation rigor. **C3.** Source visibility follows the least intrusive form that preserves confidence and recoverability: body attribution where provenance matters to understanding, endnote for factual recovery, claim ledger for qualification and method, bibliography for canonical publication information. **C4.** Qualifiers that materially change a finding **MUST** appear in the body. Bibliographic details that do not materially affect interpretation **MAY** move to notes. Uncertainty **MUST NOT** be hidden to improve prose. --- ## 5. Evidence, inference, argument ```text EVIDENCE What happened or was measured. INFERENCE What the evidence reasonably supports. ARGUMENT What the author proposes follows from it. ``` **I1.** These **MUST** remain distinguishable in the source apparatus even where narrative transitions make them flow together in the prose. **I2.** An interpretation **MUST NOT** acquire the apparent authority of the study that precedes it. --- ## 6. Statistics as narrative events **S1.** A statistic belongs in the body when it performs at least one of four functions: **reveal** something that cannot otherwise be seen, **compare** so a difference or asymmetry becomes intelligible, **contradict** the expected explanation, or **measure** the magnitude of an already observed phenomenon. **S2.** The preferred sequence is expectation, observation, number, discrepancy, question. A number **SHOULD** cause the argument to move. **S3.** If removing a statistic does not materially change the reader's understanding, it **SHOULD** move to a note. --- ## 7. The anecdote rule > A story can reveal a phenomenon. It cannot establish its prevalence. A documented incident demonstrates that something occurred. It cannot by itself establish how often it occurs, how large the effect is, whether it generalizes, what caused it, or whether another intervention would produce the same result. **N1.** Every narrative example **MUST** carry an internal evidence classification: **illustration** (makes an established concept understandable), **observation** (demonstrates that something occurred), **case evidence** (a documented instance supporting a bounded claim), **experimental evidence** (a controlled or measured result), or **representative evidence** (supports prevalence or generalization through an appropriate sample). **N2.** One class **MUST NOT** silently become another. --- ## 8. Delay the abstraction **D1.** Where practical, the reader **SHOULD** encounter a phenomenon before it is given a name. Show the work. Show the expected saving. Show the subsequent verification. Show the cost. Ask what happened to the saving. Then name the mechanism. **D2.** A coined term **MUST** compress something the reader already understands. It **MUST NOT** create an additional concept the reader must memorize before understanding the phenomenon. **D3.** This applies with particular force to: *gray*, *checking tax*, *transformation seam*, *commitment boundary*, *receipt*, and *V = Δ*. Experience first. Compression second. --- ## 9. Counterevidence **X1.** Where a claim has an important boundary, evidence that tests that boundary **MUST** be included. ```text PATTERN → PROPOSED EXPLANATION → COUNTEREXAMPLE → REFINED EXPLANATION ``` The counterexample is not an interruption. It is part of the investigation. A claim that survives qualification is stronger than a claim protected from qualification. --- ## 10. Evidence distance > The stronger the claim, the shorter the permissible distance between the claim and its evidence. ```text required evidence proximity increases with claim strength × claim consequence ``` This is an editorial relationship, not a statistical formula. A minor descriptive fact tolerates greater source distance. A surprising statistic identifies its source nearby. A causal assertion needs adjacent supporting evidence. A central quantitative claim requires immediate attribution and full documentation. A thesis-bearing claim **MUST NOT** rely on evidence the reader encountered many pages earlier. --- ## 11. Section procedure For each substantial section: identify the claim it must establish; identify the strongest evidence supporting it; determine what that evidence actually establishes; identify any inference required to reach the intended argument; find the concrete observation that exposes the problem; identify the anomaly; form the question it creates; introduce evidence only when the reader wants it; add an independent case where generalization requires one; test the explanation against a counterexample or concession; refine the claim; name the mechanism only where naming improves compression; state the consequence; return to the opening observation where doing so produces a changed understanding; verify every objective statement against the evidence apparatus. --- ## 12. Section test A section is not locked until every line passes. ```text [ ] Something concrete before the abstraction [ ] The opening creates a genuine question [ ] Each major statistic reveals, compares, contradicts or measures [ ] The reader can tell who produced important evidence [ ] Enough methodology is present to interpret the finding [ ] Evidence and author interpretation are distinguishable [ ] No anecdote is asked to prove more than it can [ ] Prevalence is supported by representative evidence [ ] Causal language is supported by causal evidence [ ] Important counterevidence has been addressed [ ] Material qualifications are visible [ ] The coined term arrives after the phenomenon [ ] Evidence is close enough to every strong claim [ ] Every objective statement is recoverable from the apparatus [ ] The section ends somewhere different from where it began ``` --- ## 13. Failure modes | Mode | Symptom | Correction | |---|---|---| | Evidence dump | Several studies appear consecutively without creating movement | Determine what question each source answers | | Anecdote inflation | A compelling case implies prevalence or causality | Narrow the claim or add appropriate evidence | | Citation camouflage | Attribution removed for readability; the sentence now reads more certain | Restore provenance or qualification | | Premature naming | A term appears before the reader understands the phenomenon | Show the phenomenon first | | Statistical decoration | A number sounds impressive and changes nothing | Remove it or make its function explicit | | Authority substitution | A famous name stands in for explaining the evidence | Describe what was measured or observed | | Interpretation smuggling | The author's interpretation is written as the source's finding | Separate evidence, inference and argument | | Narrative overreach | A clean story requires suppressing conflicting evidence | Preserve the uncertainty; the narrative serves the evidence | | Concept repetition | A named mechanism is re-explained after it is established | Coin once, hold the meaning stable, use the term as compression | --- ## 14. Boundary This method governs **presentation**. It does not determine whether evidence is sufficient. A manuscript may contain elegant narrative evidence and still fail its evidence requirements. Sufficiency is decided by `eqp:protocol/evidence-qualification` and by the manuscript's own gates. Voice conformance is decided by `app:protocol/authored-prose`. This document decides only the order in which a qualified claim reaches the reader. ```text SOURCE → QUALIFY → CLASSIFY → CLAIM → NARRATIVE PRESENTATION → READER ``` **B1.** This sequence **MUST NOT** be reversed. Deciding what story is desirable and then searching for evidence that permits it is outside this method and prohibited by it. The narrative is downstream of evidence qualification. --- ## 15. Working template ```markdown ## Claim What must this section establish? ## Scene What concrete event, person, object, process or measurement lets the reader encounter the problem? ## Anomaly What does not behave as expected? ## Question What question should now exist in the reader's mind? ## Evidence Source: / Method: / Finding: / Qualification: / Evidence class: ## Pattern What independent case establishes transfer beyond the opening observation? ## Counterevidence What case most seriously challenges the emerging explanation? ## Interpretation What mechanism best explains the surviving evidence? ## Name Does the mechanism need a name? What does the name compress? ## Consequence What changes if this interpretation is correct? ## Return Can the opening scene now be understood differently? ## Evidence Check Evidence: / Inference: / Argument: / Uncertainty: / Full source location: ``` --- ## 16. Canonical rule > Evidence is maintained at full rigor in the manuscript's source apparatus but presented to the reader through narrative discovery. Chapters move from concrete observation toward abstraction: scene, anomaly, question, evidence, pattern, interpretation, and consequence. Research is attributed naturally in prose; detailed citation information is carried by notes and the claim ledger. Statistics must reveal, compare, contradict, or measure. Anecdotes do not establish prevalence. Evidence, inference, and argument remain distinguishable. Coined concepts are introduced after the phenomenon they describe whenever practical. Claim strength determines required evidence proximity. Narrative clarity may remove citation machinery from the reading experience, but it may never remove provenance from the manuscript. **Short form.** Show before naming. Create the question before supplying the research. Treat statistics as events in the argument. Attribute evidence naturally in prose while retaining complete provenance underneath. Keep evidence, inference, and argument distinct. Never use a story to establish prevalence. Test explanations against counterevidence. Put stronger evidence closer to stronger claims. Let the reader experience an investigation while the manuscript maintains the evidence system. --- ## 17. Status This is a working draft. It has not been reviewed by anyone outside the author, and it has been applied to one manuscript. Its rules are derived from editorial practice rather than from measurement, which distinguishes it from `app:protocol/authored-prose`, whose thresholds come from a measured corpus. A conformance report against Part I of *Distance from Gray* is the first evidence that will exist for it.