The Master Report: Seven Chapters and Twelve Annexures

A finished study has two artifacts, and they answer different questions. The Excel workbook is the data document — the numbers, sheet by sheet, the record a reviewer opens to check a figure. The Word master report is the narrative master document: the structured, chaptered write-up of the benchmark in the form a reviewer, a client or a regulator reads — the engagement framed, the tested party characterised, the method selected and justified, the benchmark documented, the range computed and the conclusion stated. It is a .docx, generated on demand, and its structure is fixed: seven numbered chapters plus twelve lettered annexures, on every study, in every jurisdiction. The same generation run also renders a branded PDF — and it is the same document in another format, not a shorter one (Word and PDF: one document, two renderings).

How it is built: the report engine

The master report is produced by a deterministic report engine in five layers. The generator never calls AI, and no chapter’s content is free-form:

1. The knowledge base — frozen reference data: jurisdiction modules (curated content for 226 jurisdictions — the original six, India, UAE, USA, UK, Singapore and Germany, plus three curated tiers — and an explicitly-labelled OECD-baseline module for the handful without curated coverage), industry modules, the F01–F30 function set, transaction templates, transfer-pricing method definitions, characterization profiles, PLI definitions, the standing glossary, and the report structure itself: the seven-chapter registry plus the twelve annexures.
2. The decision engine — deterministic, defensible rules: the tested party’s characterization (derived from the FAR affinity, never from user free-text), the method selection and its justification, the PLI reconciliation, the benchmarking strategy, and the formula-driven transfer-pricing adjustment where the tested party falls outside the range. A disposal is recorded for every recognised method — selected, considered, rejected, not permitted, or available but not modelled — which is what Annexure D can print.
3. The content source — the only place report language is authored. It walks the structure registry and returns one flat stream of neutral content blocks: chapters, sections, paragraphs, bullets, tables, figures, footnotes, page breaks. It carries no rendering markup.
4. The two renderers — a python-docx builder and a reportlab/platypus builder, each consuming that same stream. Neither renderer owns a sentence, a number or a chapter; they own typography.
5. The validation gate — the consistency check that runs before either renderer, and blocks rather than warns: a method the study’s own profit level indicator cannot compute, a conclusion that disagrees with the margin against the applied range, or an adjustment whose sign, stated direction and conclusion disagree.

The structure is data, not code: the chapters live in a registry (MASTER_REPORT_STRUCTURE), the annexures in ANNEXURE_STRUCTURE, and the content source iterates them. A new section is an entry in the registry plus a builder — never an edit to one renderer, and never a chapter that exists in Word and not in the PDF. Every chapter has a registered builder, so none of them renders as a placeholder — what varies between studies is the data behind a chapter, not whether the chapter exists.

The front matter

Before the chapters, every report carries:

• The cover — “Transfer Pricing Report”, the platform identity line, the tested party’s name, the financial year, and the engagement block: Prepared for, Prepared by, Date of report, Jurisdiction, Reporting currency, Local framework, Document designation, the jurisdiction’s compliance profile, the most appropriate method, the profit level indicator and the report version. Prepared for falls back to the tested party and the date to the day the document was built; Prepared by is the popup’s one required field, so in practice it carries what the author typed (the engine’s own fallback is the platform identity).
• Confidentiality and Caveats — the fixed engagement caveat: exclusive use by the addressee, no reliance by others, no legal or tax advice, conclusions based on the facts as of the report date.
• Table of Contents — over the chapter and section headings. In Word it is a live field, so entries and page numbers appear when the field is updated; in the PDF the contents page is typeset with final page numbers.
• Abbreviations and Acronyms — the terms the report uses, defined up front.
• The running footer — on every page: the generated-by attribution, the product line and the page number, so the numbering follows the document when it is edited.

The seven chapters

#	Chapter	Contents
1	Executive Summary	The engagement in ten numbered sections: 1.1 Taxpayer, 1.2 Group, 1.3 Reporting Period, 1.4 Jurisdiction, 1.5 Controlled Transactions, 1.6 Functional Characterisation, 1.7 Method Selected, 1.8 Benchmarking Result (accepted count, arm’s-length range, comparable median, tested-party margin, conclusion), 1.9 Local Jurisdiction Conclusion, 1.10 Documentation Compliance Status
2	Corporate Background	The Group — structure and business, with the tested party’s position in the value chain; The Tested Party — corporate information from the study parameters and the business description (a narrative slot); Shareholding Pattern — the schedule as supplied at generation (shareholder, relationship, holding, jurisdiction). When nothing was supplied the chapter says the detail is not in the study data rather than printing an empty table
3	Industry Overview	The industry module, reframed for the client: introduction, key industry characteristics, value chain, value drivers, key risks, competitive landscape, regulatory environment, typical business models
4	Functions, Assets and Risk Analysis (“FAR Analysis”)	Section 4.1 takes the heading of the matched transaction template and opens with the functional-analysis framing and the FAR narrative (a narrative slot), then the FAR matrix as a table (category, item, allocation between local entity and associated enterprise) through Functional Analysis, Assets Employed, Risks Assumed and a Summary of FAR Analysis, closing with Characterization of Entities: the decision engine’s characterization, its justification and the criteria from the matched profile — never the user’s free text
5	Economic Analysis	Selection of Most Appropriate Method — each recognised method (CUP, RPM, CPM, TNMM, PSM) assessed and disposed of, then the conclusion on selection; Application of MAM — choice of tested party, profit level indicator and its justification, the search for comparable uncontrolled data (the engagement search details, each line only when supplied), multiple-year data and the range concept, the Arm’s-Length Range for the band actually applied (a percentile band prescribed by the jurisdiction’s rule where one governs, otherwise the interquartile range, or an arithmetic-mean point estimate where the rule prescribes one), comparability adjustments, points of comparison, and the conclusion on the transaction stated twice — 5.2.8.1 Global transfer-pricing conclusion, where a required adjustment is tabled with the engine’s formula narrative, and 5.2.8.2 Jurisdictional conclusion; Limitations and Assumptions
6	Jurisdictional Compliance Statement	What this engagement must file: the basis of the statement, the compliance profile, documentation and filing obligations, an explicit list of positions not taken, the rules considered and excluded, and the statement itself — each rule carrying its source, effective window and verification status, or an explicit statement that no rule set was consulted
7	Annexures	The twelve supporting schedules behind the analysis, lettered A–L, below

The annexures (A–L)

Each annexure is a heading followed by the schedule itself, built from the study’s persisted data — not a pointer to material that lives elsewhere. Schedules that cannot be populated from the record say so plainly instead of printing an empty table.

Annexure	What it carries
A	Transfer Pricing Regulations — the jurisdiction’s regulations, in the sections a reader expects: framework, domestic legislation, associated enterprises, international and specified domestic transactions, ALP computation, safe harbours, penal provisions, APA, MMY data and the range concept
B	Documents and Information Relied Upon — what the report was prepared on
C	Schedule of Controlled Transactions — the transactions in the study’s own terms
D	Method Selection Matrix — every recognised method and how it was disposed of, including the ones rejected and the ones the study’s PLI cannot compute
E	Accept–Reject Matrix — the comparable set with each company’s disposition, in the sample’s three-column form
F	Profit Level Indicator Computation of the Tested Party — the indicator worked through
G	Profit Level Indicator Computation of Comparable Companies — the same for each accepted comparable
H	Business Description of Comparable Companies — what each accepted comparable actually does
I	Regulatory Source Register — every rule consulted, with its provenance and verification status
J	Documentation Compliance Checklist — the fillable checklist for the engagement
K	Glossary of Terms Used — the standing report glossary
L	Index of Principal Documents — where the supporting documents sit

Narrative slots vs. deterministic content

The split is what makes the report defensible and regeneration possible:

• Deterministic, from the persisted study data: every number — the distribution statistics and the arm’s-length range in Chapter 5, the comparable sets in Chapters 1 and 5 and in Annexures E to G, the adjustment — plus the regulatory, industry, transaction, method and glossary content matched from the knowledge base and the decision engine.
• Narrative slots, supplied in the popup: exactly four — the purpose of the study (Chapter 1, section 1.1 Taxpayer), the company overview (Chapter 2, The Tested Party), the transaction description (Chapter 1, section 1.5 Controlled Transactions, where it leads the section and the matched transaction template still governs its heading) and the FAR narrative (Chapter 4). Each is capped at 4,000 characters, the same limit the API enforces. Where a slot is left empty, the generator falls back to a deterministic sentence — the chapter is never blank.
• A fifth slot nobody types: the context also carries adjustment_formula, and Chapter 5 reads it — but the popup does not collect it. The decision engine fills it at build time from the selected PLI, the computed adjustment and the margin type, so the adjustment narrative cannot be steered away from the arithmetic.

One record, two artifacts: the same statistics the results dashboard shows are the ones in Chapter 5, and the same dispositions the grid holds are the ones in Annexure E. Before the document is built, the generation applies the accept/reject overrides recorded in review to a build-only copy of the results — which is what keeps this report and the Excel workbook telling the same story — and the persisted analysis data is left as it was, so the review workspace still shows the live lists. When two artifacts disagree, the study’s persisted data is the authority and the older artifact is the one to regenerate.

Word and PDF: one document, two renderings

The same on-demand call produces the Word master report and a branded PDF. Both render the one content stream, so they cannot disagree on a figure, a range methodology, a heading or an annexure — a parity test renders one engagement through both and compares the two documents paragraph by paragraph, heading level by heading level and table row by table row.

What differs is mechanics, not content:

	Word master report	Branded PDF
Structure	7 chapters + Annexures A–L	The same 7 chapters + Annexures A–L
Contents	A live TOC field; update it in Word for entries and page numbers	Typeset, with real page numbers (the document is built twice so the numbers and the page count are final)
Footnotes	python-docx has no footnote API, so the citation rides in the body as a small-italic N. <citation> line under its paragraph, with the same sequential marker	Drawn in the page’s footnote gutter
Charts	The PDF’s drawing rasterised to PNG and embedded; a rasterisation failure degrades to caption + note rather than failing the report	Drawn natively, caption and artwork kept on one page
Identity	Quartyl brand tokens, frozen	Quartyl brand tokens, frozen
Purpose	The editable narrative master document	The locked, client-facing rendering

Neither generator is re-themed by white-label branding: the deliverables carry the platform identity even in a white-label tenant’s app.

Generating the document

Both files are generated on demand, never inside the analysis pipeline — the partner or firm admin triggers them from the study with the engagement details, and the engine rebuilds the report from the persisted record. The Word leg is the primary deliverable: it is built and stored first, and the PDF is a best-effort second leg, so a PDF failure leaves the Word report intact with no PDF on the study and the failure recorded on the job.

Each study has a generation budget of five builds, counted from the job ledger and printed on the action itself, so a partner knows what is left before clicking rather than after. The count includes builds that failed — dispatching one is what spends it. Refusals happen before any job row exists: the endpoint rejects a study that is out of budget, has an active job, or whose retained data has expired, and the consistency gate rejects a report that would contradict itself, naming the method, the indicator and the permitted set. Entries typed into the popup are kept as a per-study draft in the browser, so a half-filled form that is closed by mistake can be resumed; dispatching a build clears it. The workflow state follows the deliverables: a study at REPORT_GENERATED moves to DOCUMENT_GENERATED, and one still at IN_REVIEW takes the approval leg in the same run (the workbook is built there too), while a rebuild after sign-off moves nothing. The full flow is in Generating the Final Document.

FAQ

My jurisdiction has no curated module in the knowledge base — what renders? Every chapter built from the study’s own data renders in full. The regulatory material in Annexure A then speaks from the OECD Transfer Pricing Guidelines baseline and opens with a printed notice that the statements are not verified against that jurisdiction’s domestic legislation — local prescribed ranges, safe harbours, thresholds and penalties must be confirmed before the report is relied on. It never substitutes another country’s legislation. The industry chapter states plainly that no industry content is available for the engagement, and Chapter 4 does the same when no transaction template matches; the rest of the report is unaffected.

The contents page is empty when I open the file — a bug? Expected in Word. The generator writes a live TOC field, and Word fills it on update (right-click → Update Field, or F9 — the same applies to the page numbers in the footer). The PDF by contrast ships with the contents already typeset and numbered.

Can I hand-edit the generated .docx? You can open it, but the document of record is the generated one — regenerate with corrected engagement details or narrative overrides rather than editing the artifact, so the record and the file stay the same thing. Each study has five generations.

Does the PDF carry all seven chapters? Yes — the same seven chapters and the same Annexures A–L. It is the same document in another format; only the contents page, footnote placement and chart handling differ. See Word and PDF: one document, two renderings.

How does this relate to the Local File? The master report is the benchmark’s narrative write-up; the documentation tiers it feeds into — Local File, Master File, CbCR — are covered in the TP documentation guide.