Core Engine

Core Engine

This skill is the foundation of the scientific-paper-writer plugin. It is loaded automatically by every writing command and establishes the rules that govern all manuscript output. No section-specific skill may override the guarantees defined here.

---

1. No-Fabrication Guarantee

Absolute rule: Every number, statistic, finding, and factual claim in the generated manuscript text must be traceable to a source material provided by the user (uploaded PDFs, extracted tables, pasted text, figures, or PAPER_CONTEXT.md).

The assistant must never:

• Invent sample sizes, effect sizes, p-values, confidence intervals, or counts.
• Guess the direction or magnitude of an effect.
• Fabricate author names, journal names, years, or DOIs for citations.
• Fill in methodological details (randomisation method, blinding, follow-up duration) that are not stated in or clearly inferable from the source materials.

When information is missing or ambiguous, the assistant must insert the appropriate verification flag (see references/verification-flags.md):

Situation	Flag
Missing procedural or methodological detail	[PLEASE VERIFY: ___]
Data present but ambiguous or contradictory	[UNCLEAR IN SOURCE — please verify]
Citation cannot be verified against known databases	[Citation needed — unable to verify]
Figure that could not be generated or located	[FIGURE NEEDED: description]

The user is responsible for resolving every flag before submission. The assistant must never silently drop a flag or replace it with a guess.

---

2. Prose Conventions

A summary of key rules is provided here. The full reference is in references/prose-conventions.md.

2.1 Tense

• Methods and Results: Past tense ("Participants were randomised...", "The pooled odds ratio was 1.45...").
• Introduction and Discussion (established knowledge): Present tense ("Cardiovascular disease remains the leading cause of death...").
• Discussion (interpretation of current study): Past tense ("Our analysis demonstrated...").

2.2 Voice

• Passive voice is acceptable and standard in scientific writing.
• Active voice is preferred when it improves clarity ("We excluded studies that..." rather than "Studies that were excluded by the authors...").
• Maintain consistency within each section.

2.3 Language

All manuscript output uses UK English by default unless the user specifies otherwise or the target journal requires US English:

UK (default)	US (if requested)
organisation	organization
randomised	randomized
analysed	analyzed
behaviour	behavior
colour	color
centre	center
programme	program
haemoglobin	hemoglobin
oestrogen	estrogen
diarrhoea	diarrhea

The full list of 20+ spelling pairs is in references/prose-conventions.md.

2.4 Formatting

• Full paragraphs only — never output bullet points or numbered lists in manuscript sections. Bullets are acceptable only in non-manuscript output (planning notes, checklists, summaries for the user).
• Subsection headings use ## (Markdown H2) within each major section.
• Cross-reference figures and tables by label: "Figure 1", "Table 2", "Supplementary eFigure 3".

2.5 Precision

• No vague quantifiers ("some", "many", "a few", "several", "most").
• Every finding requires: point estimate + 95% CI + p-value + sample size.
• Percentages: 1 decimal place (e.g., 34.7%).
• Odds ratios, hazard ratios, risk ratios: 2 decimal places (e.g., OR 1.45).
• P-values: 2-3 significant digits (e.g., p = 0.034, p < 0.001).
• Do not write "p = 0.000"; write "p < 0.001".

---

3. Workflow Protocol

The assistant follows this protocol for every writing command:

1. Read source materials first. Before writing any prose, read and digest all uploaded documents, extracted data, figures, and PAPER_CONTEXT.md. Identify gaps early.

2. Confirm study type. Use the routing table (Section 5) to load the correct study-type skill and reporting guideline.

3. Write one subsection at a time. Pause after completing each subsection to allow the user to review, correct, and approve before proceeding.

4. Cross-reference figures and tables. Every figure and table mentioned in the text must exist in the source materials or be flagged with [FIGURE NEEDED: description].

5. No citations in Results. The Results section reports the current study's findings only. Citations belong in Discussion and Introduction.

6. Flag all uncertainties. Insert verification flags immediately when information is missing. Do not leave blank spaces or ellipses.

7. Respect expert writing order. Unless the user explicitly requests a different order, follow Section 7.

---

4. PAPER_CONTEXT.md Integration

When a PAPER_CONTEXT.md file exists in the working folder, the session hook reads it at startup. The core engine uses the following fields:

Field	Purpose
study_type	Routes to correct study-type skill
primary_outcome	Ensures Results and Abstract focus on the right endpoint
target_journal	Loads journal-specific formatting and word limits
citation_style	Configures reference formatting (Vancouver, APA, etc.)
output_format	Sets export target (LaTeX, DOCX, Markdown)
folder_structure	Locates source files, figures, and output directory
integration_preferences	Enables/disables PubMed, CrossRef, Semantic Scholar, bioRxiv
figure_preferences	Configures figure generation style and resolution

---

5. Study-Type Routing

When the user provides a study type (via PAPER_CONTEXT.md or directly), the core engine loads the corresponding study-type skill from skills/study-types/.

Identifier	Display Name	Reporting Guideline	Skill Path
systematic-review	Systematic Review & Meta-Analysis	PRISMA 2020	skills/study-types/systematic-review/
network-meta-analysis	Network Meta-Analysis	PRISMA-NMA	skills/study-types/network-meta-analysis/
ipd-meta-analysis	Individual Participant Data Meta-Analysis	PRISMA-IPD	skills/study-types/ipd-meta-analysis/
dose-response	Dose-Response Meta-Analysis	PRISMA 2020 (adapted)	skills/study-types/dose-response/
bayesian-meta-analysis	Bayesian Meta-Analysis	PRISMA 2020 + BUGS/Stan reporting	skills/study-types/bayesian-meta-analysis/
rct	Randomised Controlled Trial	CONSORT 2010	skills/study-types/rct/
observational	Observational Study (Cohort/Case-Control/Cross-Sectional)	STROBE	skills/study-types/observational/
economic-evaluation	Economic Evaluation	CHEERS 2022	skills/study-types/economic-evaluation/
prediction-model	Prediction Model Study	TRIPOD+AI	skills/study-types/prediction-model/
diagnostic-accuracy	Diagnostic Accuracy Study	STARD 2015	skills/study-types/diagnostic-accuracy/
qualitative	Qualitative Research	COREQ / SRQR	skills/study-types/qualitative/

If the study type is not recognised, the assistant must ask the user to clarify before proceeding, suggesting the closest match from the table above.

---

6. Reporting Guideline Mapping

Each reporting guideline has a canonical reference that must be cited when the guideline is invoked in the Methods section.

Study Type	Guideline	Key Reference
Systematic Review & Meta-Analysis	PRISMA 2020	Page MJ, McKenzie JE, Bossuyt PM, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ. 2021;372:n71.
Network Meta-Analysis	PRISMA-NMA	Hutton B, Salanti G, Caldwell DM, et al. The PRISMA extension statement for reporting of systematic reviews incorporating network meta-analyses. Ann Intern Med. 2015;162(11):777-784.
IPD Meta-Analysis	PRISMA-IPD	Stewart LA, Clarke M, Rovers M, et al. Preferred reporting items for a systematic review and meta-analysis of individual participant data: the PRISMA-IPD statement. JAMA. 2015;313(16):1657-1665.
Dose-Response Meta-Analysis	PRISMA 2020 (adapted)	Page MJ, McKenzie JE, Bossuyt PM, et al. BMJ. 2021;372:n71. (With dose-response-specific extensions as described by Crippa et al.)
Bayesian Meta-Analysis	PRISMA 2020 + Bayesian reporting	Page MJ et al. BMJ. 2021;372:n71. (Supplemented with Bayesian-specific reporting per Rover et al., Res Synth Methods. 2021.)
Randomised Controlled Trial	CONSORT 2010	Schulz KF, Altman DG, Moher D. CONSORT 2010 statement: updated guidelines for reporting parallel group randomised trials. BMJ. 2010;340:c332.
Observational Study	STROBE	von Elm E, Altman DG, Egger M, et al. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement. Lancet. 2007;370(9596):1453-1457.
Economic Evaluation	CHEERS 2022	Husereau D, Drummond M, Augustovski F, et al. Consolidated Health Economic Evaluation Reporting Standards 2022 (CHEERS 2022) statement. BMJ. 2022;376:e067975.
Prediction Model Study	TRIPOD+AI	Collins GS, Reitsma JB, Altman DG, Moons KGM. Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis (TRIPOD). Ann Intern Med. 2015;162(1):55-63.
Diagnostic Accuracy Study	STARD 2015	Bossuyt PM, Reitsma JB, Bruns DE, et al. STARD 2015: an updated list of essential items for reporting diagnostic accuracy studies. BMJ. 2015;351:h5527.
Qualitative Research	COREQ / SRQR	Tong A, Sainsbury P, Craig J. Consolidated criteria for reporting qualitative research (COREQ). Int J Qual Health Care. 2007;19(6):349-357.

---

7. Risk of Bias Tools

The assistant selects the appropriate risk-of-bias tool based on study design. This table is used when writing the Methods section (quality assessment subsection) and when interpreting quality findings in the Discussion.

Tool	Full Name	Applicable Study Designs	Key Reference
RoB 2	Revised Cochrane Risk-of-Bias Tool for Randomised Trials	Randomised trials (parallel, crossover, cluster)	Sterne JAC, Savovic J, Page MJ, et al. BMJ. 2019;366:l4898.
ROBINS-I	Risk Of Bias In Non-randomised Studies of Interventions	Non-randomised comparative studies	Sterne JA, Hernan MA, Reeves BC, et al. BMJ. 2016;355:i4919.
NOS	Newcastle-Ottawa Scale	Cohort and case-control studies	Wells GA, Shea B, O'Connell D, et al. Ottawa Hospital Research Institute; 2000.
QUADAS-2	Quality Assessment of Diagnostic Accuracy Studies, version 2	Diagnostic accuracy studies	Whiting PF, Rutjes AW, Westwood ME, et al. Ann Intern Med. 2011;155(8):529-536.
JBI	Joanna Briggs Institute Critical Appraisal Tools	Qualitative, prevalence, case series, text/opinion	Joanna Briggs Institute. JBI Manual for Evidence Synthesis. 2020.
PROBAST	Prediction model Risk Of Bias ASsessment Tool	Prediction and prognostic model studies	Wolff RF, Moons KGM, Riley RD, et al. Ann Intern Med. 2019;170(1):51-58.

---

8. Expert Writing Order

Sections are written in the following order, which reflects how experienced researchers compose manuscripts. This order maximises internal consistency and minimises rework:

Order	Section	Rationale
1	Methods	Establishes what was done; anchors all other sections.
2	Results	Reports findings that follow directly from Methods.
3	Discussion	Interprets Results in context of existing literature.
4	Abstract	Summarises Methods, Results, and Discussion accurately.
5	Introduction	Frames the study after the full scope of findings is known.
6	Title	Captures the key finding after all sections are drafted.
7	Captions	Finalises figure and table descriptions for consistency.
8	References	Compiles and verifies all cited works as the last step.

The user may request sections in any order. When they do, the assistant follows the request but warns if critical upstream sections have not been written yet (e.g., writing the Abstract before Results are finalised).

---

9. Cross-Cutting Rules

9.1 Abbreviations

• Define each abbreviation at first use in each major section (Methods, Results, Discussion, Abstract, Introduction are treated as independent units).
• Do not abbreviate terms used fewer than three times in a section.
• Standard abbreviations that need no definition: DNA, RNA, HIV, AIDS, BMI, WHO, CI, OR, HR, RR, SD, IQR.

9.2 Numbers

• Spell out numbers zero through nine when used in general text.
• Use numerals for 10 and above, and for all measurements, statistics, and quantities (e.g., "5 mg", "3 studies", "12 participants").
• Never start a sentence with a numeral; rephrase or spell out the number.

9.3 Statistical Notation

• Italicise statistical symbols: p, n, r, t, F, z, I^2^.
• Use the multiplication sign (x) not the letter x for interactions.
• Report exact p-values (p = 0.034) except when p < 0.001.
• Always report 95% confidence intervals in parentheses: (95% CI 1.12 to 1.89).
• Use "to" not a dash for CI ranges in text; dashes are acceptable in tables.

9.4 Supplementary Materials

• Refer to supplementary items with the "e" prefix: eFigure 1, eTable 2, eMethods, eResults.
• Number supplementary items sequentially in order of first mention.

---

10. Error Handling

Error Condition	Action
No source materials provided	Ask the user to upload files before writing.
Study type not specified	Ask the user to specify or provide PAPER_CONTEXT.md.
Contradictory data in sources	Flag with [UNCLEAR IN SOURCE — please verify] and note the discrepancy.
Requested section depends on unwritten upstream section	Warn the user and suggest writing upstream sections first.
Target journal not specified	Use generic academic formatting; note that journal-specific formatting can be applied later.
Citation cannot be verified	Flag with [Citation needed — unable to verify] and retain the user-provided reference.

---

References

• references/prose-conventions.md — Full prose style guide
• references/verification-flags.md — Complete flag definitions and examples