Research Methodology India: Brief as North Star, Evidence as Proof

A Research Brief is a free-text statement of business goals that lives in survey_chats.research_brief and states the bet the research must test, which is why research methodology India on Hercules Works calls it the North Star rather than a field in a form. Example brief: we need to choose concept B versus C for Tier 2 mothers around ₹499 based on purchase intent plus driver gap plus price elasticity, with go criterion intent greater than 60 percent at n equals 400 per concept and driver gap clarity. That sentence names decision, options, segment, price, metrics, threshold and sample, so downstream nodes do not guess; they optimise toward it. Built by Jupiter Meta Labs in Bangalore, Poseidon injects Brief into every major LLM prompt, so Report Planner knows to prioritise concept battery and pricing primitives over demographics, and Insight Generator frames every finding as priced decision evidence, not isolated frequency. Without Brief, a report would allocate equal space to each of 40 questions; with Brief, it allocates space proportional to impact on the bet.

The Brief also curates the report skeleton so methodology serves decision order. A traditional Indian methodology chapter lists sampling frame, weighting, formulae, limitations in textbook sequence; a Brief-driven report opens with Executive Summary that states recommendation plus evidence, then Research Goals plus Decision Context that restates the bet, then Methodology that states how the Brief was operationalised — SuperJ quotas Tier1 slash 2 slash 3 plus NCCS plus language, sample achieved n equals 1,204 versus plan 1,200, weighting post-stratification where needed, and formula choices like Borda for ranking versus top-two-box for Likert — so methodology is traceable to choice, not generic inventory. That traceability is what Government of Karnataka reviewers trust for citizen surveys: the chapter does not merely say we used stratified sampling, it says we sampled wards proportional to census because the Brief asked to represent north Bangalore complaint density, with lift thread to prove it.

Crucially, the Brief isEditable but versioned with the chat, so a methodology change is not hidden. If Meera in Hyderabad revises brief from Tier 2 mothers to Tier 1 to 2 working mothers after pilot, the new brief flows to Pipeline re-plan, recomputation, re-synthesis, with diff visible between Html versions, so report plus audit show which insight shifted because segment shifted, not because analyst forgot. That versioning plus five injection points makes research methodology India on Hercules operational, not declarative, because methodology is not what you wrote before field; it is what you promised, what you did, and whether promise plus proof align, measured via goal traceability. From Free ₹0/month with 10 AI chats to Pro ₹30,000 per quarter, the Brief is therefore the cheapest lever to improve a report, since writing one sharp paragraph steers 18 nodes toward the right charts.

The Brief flows to five pipeline places, which is the execution proof that research methodology India on Hercules is method plus governance, not vibe. One touches Report Planner — which parses questionnaire, identifies question types, maps each research goal to specific survey questions via LLM, and creates ordered task list deciding key metrics versus appendix, prioritising the concept battery and price curves over color preference because the Brief said choose concept at price. Two touches Insight Generator — which extracts per-question plus cross-question insight packs with severity, framing each finding against business goal, so purchase intent 65 percent for B at n equals 412 is severity high for Tier2 ₹499 decision, while intent 52 percent for C is low, with Wilson CI attached. Three touches Narrative Synthesis — which writes goal narratives one per objective, synthesising tasks that advance that goal plus demographic threads plus routing-conditional insights like among Tier2 mothers n equals 340. Four touches Report Drafter — which uses Brief as Research Context in every section prompt, embedding chart tokens {{CHART_task_*}} only where they advance the bet, with structured Html classes. Five touches

Each injection carries a contract. Planner must show goal to question mapping table in appendix so reviewer sees which Brief sentence produced which task; Insight must carry severity tags that Synthesis respects, so high severity insights get executive placement; Synthesis must produce goal narratives that correspond one-to-one with Brief goals, so if Brief promised driver plus elasticity there are two narratives not one; Drafter must keep business implication before evidence per So What rule, enforced by Auditor; Auditor must cite Brief language in its critique — this section claims Tier1 lead but Brief asked Tier2 decision, rewrite with Tier2 evidence — so the rewrite is pinpoint not vague. That contract chain makes research methodology India auditable: an ICICI Prudential risk reader can trace a recommendation choose B to driver gap plus intent plus WTP band plus n plus p plus confidence, each tagged to goal that Brief named, so methodology is not a claim of rigour but a graph of parentage from promise to proof.

The synthesis that emerges is therefore not chapterised by questionnaire order but by goal, which is the single largest readability gain. A Pune agency that once delivered 40 pages in question order now delivers Goal1 Choose Concept for Tier2 mothers as pages 8 to 14 with funnel plus intent by concept plus driver bar plus open theme hidden charges, and Goal2 Price to maximise adoption as pages 15 to 22 with Van Westendorp plus Gabor-Granger plus WTP band, each linked via Goal Traceability coverage map that shows Goal1 92 percent covered, Goal2 88 percent, so missing evidence is known before publish. Built on FastAPI plus LangGraph plus Gemini plus DuckDB plus Parquet with 5-phase plus 18-node duality, research methodology India on Poseidon thus implements what textbooks describe but rarely deliver: Brief in five places turning a questionnaire dump into a decision memo where every chart answers the bet, priced inclusively from Free to Pro alongside SuperJ 20M at 60 to 90 percent.

Every report on Hercules Works stands on five context sources, the material that research methodology India must cite to be credible. Source one is Research Brief — the why, the bet. Source two is Survey Version slash Questionnaire — what was asked, with question text, declared types nps likert_5pt multi_select ranking text versus vertical_ranking grid, options, scales, skip logic showIf next, stored as survey_chats.survey_version JSONB, so Report Planner knows battery composition. Source three is Survey Response Data — what people said, every answer per respondent stored as data/{chat_id}.parquet DuckDB-optimised columnar, with multi_select as semicolon A B C, ranking as ordered semicolon list, so Analysis can UNNEST string_split before counting. Source four is Audience Payload — who was targeted, demographics, quotas, geography via SuperJ, stored as survey_chats.audience_payload, so Respondent Profile section reports achieved versus planned by city tier NCCS language age, with weighting via post-stratification where needed. Source five is Survey Intelligence — auto-discovered patterns at runtime via survey_intelligence_node, including routing pairs via null-correlation where conditional share not base must be reported, plus goal to question mapping, so conditional insight among theatre-goers n equals 40 has

The journey from JSON to PDF respects those sources step by step. SuperJ JSON arrives typed via Question Nodes, is written via CSV audit to Parquet columnar, with DuckDB handling columnar reads ten to twenty times faster than Pandas, and Survey Knowledge Graph built with Question, Column, Formula, Demographic Axis Nodes plus routed_from edges, so Data Analysis knows valid formulas per column type and Demographic Axis for GROUP BY, while Quality Checker flags low-base n equals 26 plus variance plus completeness. Open-Ended Intelligence codes text where present or no-ops if none, Visualisation via chart advisor renders Html interactive plus PNG print identically, Insight plus Synthesis weave the five sources into goal narratives, Drafting writes with tokens, Audit verifies both numbers plus whether Brief plus Questionnaire plus Response plus Audience plus Intelligence all support claim, and Assembly via WeasyPrint finalises MD plus Html plus PDF. That pipeline is why methodology chapter on Hercules can name file paths plus tradeoffs without hedging, because files exist and graph proves denominator correction.

Why five not four matters for Indian relevance. Audience Payload proves Bharat coverage — Tier 1, Tier2, Tier3 plus rural, NCCS A to E, eight plus languages — so a claim about Tier2 mothers has a payload that shows sampling Tier2 quotas plus language Tamil versus Hindi slices, not a national mean masquerading. Survey Intelligence proves skipped logic was not averaged over full base, so churn reasons n equals 112 not 400, hidden via routing edges plus null-correlation discovery that generic BI never runs. Survey Version plus Parquet prove Borda was used for ranking and UNNEST for multi-select, not averaged ranks or respondent count. Those cross-checks are the methodology that reviewers like Kantar demand: not that we sampled well, but that we know for this question what type it was, what base is correct, who answered in which tier, and what discovered pattern we used, all versioned with the chat in Postgres for diff. Research methodology India on Hercules therefore achieves what manual methods promise on slides but forget in execution, with caching at 78 percent hit making re-run cheap after wave add.

The writing rule that makes methodology readable is So What — business implication before evidence — which is why research methodology India on Hercules reads like memo by design, not by luck. Every paragraph the Drafter writes must answer so what does this mean for the business, then show percent plus N, never reverse where evidence precedes implication. Bad 65 percent of respondents prefer digital. Good Digital is the dominant channel — 65 percent prefer it at n equals 612, making mobile-first the highest ROI acquisition bet. Section Auditor enforces this via Self-Critique scoring 0 to 10 for unsupported claims plus causality plus missing N, sending NEEDS_REVISION if implication absent, up to two loops, while Safety Check validates PII plus numeric mismatch plus coverage, so a paragraph that claims Tier2 lead without Brief tie is rewritten before publish. That rule turns methodology from a technical chapter into a decision guide where a busy Brand Manager scanning between meetings learns what to do from the first clause.

Goal traceability ensures promise plus proof align, another governance pillar. Every section is tagged with goal ids it advances, producing a coverage map shown in appendix: Goal1 Choose concept at price — tasks funnel, intent by concept, driver gap, Van Westendorp — coverage 92 percent; Goal2 Price elasticity — Gabor-Granger plus WTP band — 88 percent; Goal3 Channel mix — payment mode multi-select plus theme co-occurrence — 75 percent flags missing theme coding because text questions insufficient. That map tells Meera before signing whether the report can own its recommendations or must hedge with boost sample. Cross-Section Synthesis then hunts for patterns that repeat across two plus sections — hidden charges theme correlates with detractor in NPS plus trust battery low for Tier2 — linking disparate charts into one meta-driver, which is the synthesis that senior analysts sell but junior decks miss. Together So What plus traceability plus synthesis turn research methodology India from per-section reportage into cross-referenced argument.

Writing for Brand Manager not statistician is the voice choice that bundles all above. A statistician wants formula name plus p plus statistic; a Brand Manager wants implication plus p plus n plus CI in plain English. Hercules delivers both via three-persona adaptive voice — executive for summary, analytical for methodology, consulting for recommendations — tuned to section type, with numbers always re-derived within 0.5 and hash muted where n equals 26. So methodology states we used Welch when Levene p less than 0.05 with CI 3.1 to 18.1, but recommendation says Tier2 service gap is real at p equals 0.006, consider phased hiring tied to 100 plus boost test; the first satisfies audit, the second satisfies action. Built by Jupiter Meta Labs in Bangalore with Okabe-Ito paletted visuals plus Inter plus direct labels, research methodology India on Poseidon thus marries proof plus readability plus traceability plus synthesis in one chain, where each paragraph earns its evidence and each goal owns its pages.

Putting it together as methodology you can hand to risk, research methodology India on Hercules Works follows order that textbooks preach but surveys rarely follow. Planning starts with Brief plus Survey Version plus Audience Payload plus Intelligence, creating a task list where Chart advisor assigns Plotly Html plus PNG per shape — horizontal bar for NPS and multi-select beyond 100 plus binary donut plus three pie plus 6 to 8 bar plus cross-tab heatmap plus histogram plus funnel plus Borda MaxDiff — with Okabe-Ito palette plus Inter and {{CHART_task_*}} tokens only where needed, then Data Analysis executes DuckDB on Parquet while Open-Ended Intelligence clusters Hinglish plus Hindi plus Tamil verbatims where text exists, Insight plus Synthesis write goal narratives plus demographic threads like Tier2 kirana over-index 1.8 times, plus gap analysis blame versus act, Drafting writes So What before evidence, Audit plus Safety verify then revise up to two loops each, Assembly renders Markdown to Html to PDF with cover via WeasyPrint, all persisted plus versioned plus cached at 78 percent hit plus streamed with persistence. That is the 18-node plus 5-phase duality in one

Why write for Brand Manager not statistician is the positioning that closes methodology. A statistician audits the appendix where table shows column type to valid formulas — multiple_choice to frequency slash crosstab slash chi-square, multi_select to multi-frequency via UNNEST, ranking to Borda, nps to NPS, number_rating to mean slash correlation, likert to mean slash top-two, text to sentiment slash theme, vertical_ranking to grid heatmap — plus p plus CI plus Levene plus n plus hash where small, so a Kantar peer can verify choice. A Brand Manager reads Executive Summary plus Goal Narratives where those choices are translated to implication — choose B for Tier2 at ₹499 because intent 65 percent n equals 412 plus driver gap trust over discount plus Van Westendorp acceptable 399 to 549 — with chart token footnotes that tie visual to task, so a Hyderabad decision can meter hiring plus media now, not after a statistics workshop. That dual-track — analytical appendix plus executive narrative — is the three-persona voice at work, where methodology serves both and pleases neither by compromise.

Start with a Brief sentence that names the bet, include N plus threshold and segment plus price, target SuperJ 20M-plus verified via WhatsApp at 60 to 90 percent with quotas, wait for ingest where typed JSON shifts via CSV to Parquet and graph builds, watch 18 nodes synthesise plus Audit via 0 to 100 bar that survives refresh, read Markdown plus Html plus PDF where So What plus N precede evidence and goal map shows as 92 percent covered. From Free ₹0/month permanent 10 chats plus 100 users plus 100 free responses month one, Starter ₹1,119 per month ₹895 annual with 20 percent off, Pro ₹30,000 per quarter ₹24,000 annual with 20 percent off, all trusted by Unilever plus Kantar plus Govt Karnataka plus ICICI plus SBI, research methodology India on hercules.works/ai by Jupiter Meta Labs makes method the North Star that flows planner to insight to narrative to drafting to audit, cites five sources from brief to Parquet to payload to intelligence, writes implication before evidence, traces goals to pages, synthesises across sections, and delivers voice for the Manager not the manual.