Strategic Research Framework India: No-Default Design from Taxonomy to Field

Classify the study first — study type constrains which designs are legitimate. STUDY_TYPE_TAXONOMY declares 17 types and is injected into the research guide and goal formulation prompts in Hercules. Classic commercial cores are U&A (category funnel + usage occasions + brand repertoire + attitudes), Brand Health/Tracking (salience → unaided before aided awareness → familiarity → consideration → usage → loyalty + imagery batteries), Concept Screening (1–N ideas with appeal/uniqueness/relevance/believability/purchase intent top-2-box + verbatims where monadic designs live), Product Testing, Packaging/Design, Advertising Pre-test, Campaign Tracking post, Pricing Research, Prioritization, CSAT/CX/NPS, Churn, Segmentation, Sizing, Shopper Path-to-Purchase, plus Public Opinion/Social, Employee/Internal and Academic Exploratory. A study may combine 2–3 types as Mixed Methods, but modules are allocated per objective — never bolting a product frame onto a non-product study. See brief creation at survey research brief goals India and pipeline at hercules survey creation process India.

Why public opinion and academic studies must stay out of product frames. Public-opinion, civic, academic or general studies get balanced attitude statements, knowledge checks and demographic cross-cuts — never purchase intent, concept appeal or pack cells. That is why a "tea vs coffee preference among Tier 1 18–24" prompt correctly lands on a Linear structured questionnaire with semantic differentials and randomised batteries, while "compare two biscuit packs" is eligible for Paired Comparison/Forced Choice. Each module names construct, method, who should answer, routing need and question-budget consumed, so business_flow.questionnaire_sections with intents and objective maps trace directly to study type. Qualitative extensions (FGDs 8–10 participants, IDIs, Ethnography & IHUTs, Online Communities, Co-creation Workshops) are available via the enterprise team when quant alone cannot reach the "why." Quant self-serve on SuperJ via SuperJ app — 20M+ ZK-verified Indians at 60-90%+ for rewards — includes Product Testing, Brand Health, U&A, Ad Testing, CSAT, Pricing, Concept Screening and Shopper Path-to-Purchase with full geography and NCCS targeting.

How this shows up in your chat. When you type "test 3 new masala biscuit concepts among mothers 24–45 in Hyderabad and Lucknow," the research guide first classifies Concept/Idea Screening (with pricing overlay), then RESEARCH_DESIGN_CATALOG enumerates stimulus vs pricing vs descriptive options, and only then does METHODOLOGY_DECISION_FRAMEWORK count 3 discrete stimuli to shortlist Sequential Monadic/Proto-Monadic/Pure Monadic/Side-by-Side with rationale and two alternatives for override. The formulated_goals that follow (e.g., "Measure purchase intent top-2-box among category-aware mothers") inherit that classification, so business_flow later places pack stimuli via image_question/image_multiple_choice with forced-exposure timers. This Layer-1-first discipline prevents the costliest mistake: choosing a sophisticated method for the wrong question.

Stimulus-evaluation (cell) designs — only when there is a stimulus to show. The catalog treats all designs as equal citizens; none is default and selection comes only from the decision framework. Cell designs require at least one discrete stimulus (concept, pack, ad, name, logo, price point, product variant) and are grouped by burden and need. Single Monadic Exposure is one stimulus with a full KPI battery (cleanest absolute read; n 150–300 directional, 300+ validated). Pure Monadic (between-subjects) gives each respondent exactly one of N stimuli — the gold standard when contamination must be eliminated (taste, fragrance, price points, final ad validation) but multiplies sample (150–200+ per cell). Sequential Monadic (within-subjects) shows 2–4 stimuli one-at-a time with the full battery after each, order rotated via Latin Square for 3+ (SD-008), sample-efficient absolute + comparative read with watch on fatigue and battery length.

Stimulus-evaluation (cell) designs — only when there is a stimulus to show. Proto-Monadic is monadic first then side-by-side preference at the end when both an uncontaminated absolute and an explicit preference are required (common for HPC packs). Reverse Sequential shows all stimuli briefly (shelf/menu overview) then evaluates each — shelf/assortment realism. Paired Comparison/Forced Choice (SD-005, SD-007) shows exactly two stimuli simultaneously and forces a pick with no neutral — fast winner call, no absolute diagnostics. Side-by-Side Comparative (SD-006) shows 3–4 simultaneously with comparative ratings — fastest, weakest absolute. These map directly to framework_designs.md SD-001→007, SD-043 and rotation families SD-008→011, and each is traceable to a detection trigger (e.g., "A/B test" → Paired Comparison, "Latin square" → rotation control). See routing implications for cells and filters at survey business flow routing India.

Trade-off, prioritisation and pricing — ratio-scaled, not vibes. When the objective is to rank many items, the catalog offers ratio-scaled designs: MaxDiff/BWS (SD-015) repeated best/worst picks from balanced subsets yielding interval-level utilities — far better than rating grids or ranking for 8–30 features/claims/flavours; CBC (SD-019) choose among attribute-level profiles for utilities and market simulation (needs experimental design and higher burden); Kano (SD-016) functional/dysfunctional pairs classifying must-be/performance/delighter; TURF (SD-020) reach optimization across portfolios; Constant Sum/Chip Allocation for intensity; simple Ranking only for 3–7 items (beyond that, use MaxDiff). Pricing families: Van Westendorp PSM (SD-012) four open price questions (too cheap/cheap/expensive/too expensive) on numeric keyboards mapping acceptable range; Gabor-Granger (SD-013) sequential price ladder yielding demand curve and revenue-optimal price among tested points; Brand-Price Trade-Off/CBC with price; and Monadic price cells (purest but most expensive). These overlay the primary cell design in Step 3 of the decision framework — e.g., a 3-concept Sequential Monadic core plus Van Westendorp or Gabor-Granger module when pricing is a decision.

Descriptive (non-cell) designs — the majority of Indian surveys never show a stimulus. Most studies on Hercules are descriptive, with zero stimuli to evaluate: Linear structured questionnaire (screener → warm-up → core modules → closing) for U&A, CSAT/CX, opinion polls, employee and sizing; Modular funnel (general-to-specific, unaided before aided, behaviour before attitude, attitude before intent) for brand funnels and batteries; Tracking wave design (identical core across waves, rotate non-core) for brand tracking; Semantic differential / attribute batteries (bipolar scales, randomised statement order, balanced positive/negative) for imagery. These are linear or modular, not monadic, because with ZERO stimuli cell-based designs are off the table by rule.

Step 1 — count discrete stimuli (0 / 1 / 2–4 / 5+). The framework reasons in order, never skipping to a favourite. With ZERO stimuli (no discrete stimulus to show or describe), monadic, sequential monadic, A/B, paired comparison or split cells are prohibited — the design must be linear/modular/descriptive driven by study type (U&A modules, brand funnel, satisfaction drivers, opinion batteries). Most public-opinion, satisfaction, U&A, tracking, sizing and employee studies land here; that is intended. With ONE stimulus, choose Single Monadic Exposure (with pre-exposure baseline when before/after helps). With TWO TO FOUR stimuli, jump to Step 2. With FIVE OR MORE evaluable items, stop showing full batteries per item; instead use MaxDiff (features/claims/messages), a screening round (short battery, top 2–3 to deeper test), or CBC — the only humane way to rank 12 flavours without a 45-minute survey that dies on mobile. This single count prevents the two costliest mistakes: fielding a 12-item rating grid that straight-lines at scale, and proposing a split-cell A/B when there is nothing to split. On Hercules Works in Hyderabad, this count is computed from your brief's stimulus list, not guessed from keywords.

Step 2 — for 2–4 stimuli, pick on need + budget. Absolute read + affordable cells + contamination risk (taste, smell, price, final validation) → Pure Monadic. Absolute read + limited sample → Sequential Monadic with rotation (Latin Square/balanced). Absolute AND explicit preference → Proto-Monadic. Relative read only, fast winner call → Paired Comparison/Forced Choice for 2, Side-by-Side for 3–4. Shelf/assortment realism → Reverse Sequential. Absolute here means scores comparable to norms/usable standalone; relative means just a winner. Budget explicitly means per-cell n and total n — e.g., 3 packs × 200 per cell = 600 total for Pure Monadic vs ~200 with rotation for Sequential — and the framework requires stating that in the brief so stakeholders see cost before fielding. This transparent costing is why Hercules can field via the SuperJ app at 60-90%+ completion among 20M+ ZK-verified Indians for rewards without overbuying sample: the design that fits the question and the budget is chosen jointly, not sequentially. Verification docs cite SD-001→007 and SD-043 directly, so no invented price elasticity numbers appear in the recommendation.

Step 3 and Step 4 — overlays and sanity checks before committing. Step 3 overlays objective-specific modules from study type and pricing/trade-off designs where demand exists: a concept test with pricing → monadic core + Van Westendorp or Gabor-Granger module; a U&A with feature prioritisation → U&A modules + one MaxDiff exercise. Mixed-method rule is strict: combine modules only when objectives genuinely require different evidence types, keep each module tied to explicit objectives, and explain why they belong together. Step 4 gates commitment: does the design answer every stated business decision (if not, add module or flag gap), is burden 8–12 minutes (15–25 questions, 20-minute ceiling — cut modules before cutting quality), are per-cell n and total n stated, and are 2–3 credible alternatives presented with one-line rejection reasons inviting override? Only then does the system write business_flow (questionnaire_sections, branch_blocks, dynamic_lists, rejoin) and the abstract routing_logic (rules, display_rules, dynamic_filters) in flow_key terms for downstream compilation. Anti-defaults are hard rules throughout: never default to Sequential Monadic, never A/B without stimuli, never force product frames onto non-product studies, never conjoin for <5 items.

Acquiescence and positive-end scale use — Indian response styles are not Western. INDIA_SURVEY_DESIGN_RULES warns that Indian respondents show elevated acquiescence (yes-saying) and positive-end scale use, amplified by courtesy bias. Hercules therefore prefers fully labeled 5-point unipolar scales over agree/disagree batteries, validates attitudinal claims with behavioural and forced-choice questions ("How many times did you…"), and avoids leading phrasing ("Don't you think…"). Semantic-differential batteries randomise statement order and balance positive/negative statements, and lists of 20+ items use MaxDiff rather than rating grids that invite straight-lining (SD-034). This is not cosmetic — Poseidon's downstream statistical primitives (top-2-box, Borda counts for ranking, mean/median with IQR, chi-square for grids, correlation/regression for drivers) are calibrated for these scales and would be biased if agree-disagree were used naively. Survey logic also enforces at least 25–30% advanced insight techniques — behavioural anchoring, projective, hypothetical, trade-off, aspiration gap, friction surfacing, word-association — so attitudinal claims are triangulated with behaviour before you stake budget on them. Explore cross-tabs by NCCS and tier in tier 2 tier 3 consumer research India and survey data verification India.

NCCS, geography and incidence — sample like Bharat, not Bangalore. Socio-economic classification uses NCCS A1–C1 on this platform (A1, A2, A3, B1, B2, B3, C1) with per-class audience counts — never Western income bands, never caste or religion unless explicitly required and asked, and income in INR bands with optional-feeling tone. Geography thinks metro/Tier 1/Tier 2/Tier 3 and region (North/South/East/West) rather than urban/suburban; Tier 1 is 7 metros (Delhi, Mumbai, Bangalore, Chennai, Kolkata, Hyderabad, Pune), Tier 2 is 60+ cities, Tier 3 is 30+ towns, and category behaviour differs sharply by tier — premium D2C spreads metro-heavy, FMCG staples spread broadly. Audience_payload enforces this via total in [100,400] (typical) with per-class/per-city counts and behavioural attributes (e.g., "Quick Commerce User":100), while General Population uses checkCity:false. Device reality seals the design: respondents answer on smartphones on the SuperJ app for rewards, so instruments are 8–12 minutes, short screens, max 5 rows visible in grids, tap-friendly image_multiple_choice, video/audio with forced timers where comprehension matters, and break screens for 21+/31+/46+ questions. Household purchase joint-decision framing, festive/seasonal timing notes, and nuclear-family-single-earner neutrality are baked into wording — so insights transfer from Hyderabad to Hubballi without distortion.

Language, stimulus and data quality — Indic first, then quality gates. Phasing and content follow Indian execution guidance: write in simple direct English under 15 words for Tier-2/3 skews, free of Western idioms, with Indian brand/currency/festival/category references and media stimuli in Hindi/Indic where needed (PDF/DOCX/PPTX/XLSX/image extraction via PyMuPDF/docx/pptx/openpyxl/Pillow, document classifier routing). Stimulus management (SD-036→039 concept card/pack shot, shelf/planogram, video stimulus, forced exposure timer) and quota families (SD-026→029 hard/interlocking/soft/dynamic quota rebalancing) plus screener/panel logic (SD-030→031) are configurable per study but declared in the framework, not bolted on during routing. Data quality is not an afterthought: attention checks (SD-032 instructional manipulation), speeder detection at 40–50% of median LOI (SD-033), straight-lining/pattern detection (SD-034), open-end gibberish AI scoring (SD-035), and ZK panel deduplication with configurable exclusion windows on SuperJ (SD-044, blockchain identity) are armed via the rubric — compute_survey_quality_scorecard scores them before fielding. See the full rubric at survey quality rubric linter India and the analytics that later verify numbers at statistical significance testing India.

GLOBAL_RESEARCH_STRATEGY plus DISTRIBUTION_CONTEXT — the constraints you design against. On top of the three layers, Hercules injects GLOBAL_RESEARCH_STRATEGY (start from the decision to support, the evidence gap, and the respondent experience needed to answer it credibly — never select by keyword or hardcoded default; treat selected modules as first-class artefacts naming construct/method/who/routing/question-budget; mixed-method only when evidence types differ; global quality standards: behaviour before attitude, unaided before aided, general before specific, absolute evaluation before forced comparison unless explicitly required; build in quality controls and translation discipline) and DISTRIBUTION_CONTEXT (surveys deploy to verified Indians on the SuperJ app and web; targeting is enforced before the survey starts so the questionnaire must not re-collect demographics; media stimuli — images, video, audio — plus routing via skip/branch/display/terminate/filter, rotation via randomization, quotas via audience system, attention checks, and break screens are natively supported). These are fixed platform facts — stated, never invented — and they shape burden, language and sampling realities before a single question is written. The FULL_METHODOLOGY_BLOCK concatenates these six constants (GLOBAL + Layer1 + Layer2 + Layer3 + INDIA + DISTRIBUTION) into the f-string/format prompts that Gemini reads, ensuring every intent handler sees the same canon.

What distribution on SuperJ actually means for your study. Because targeting (age, gender, city, NCCS) is enforced by the audience system before the survey starts, routed screeners are for qualification (e.g., category usage) rather than quota fields — a discipline that prevents duplicate NCCS asks and saves instrument time for real diagnostics. Media stimuli (images, video, audio) are first-class on the SuperJ app with rewards, and quotas (hard/interlocking/soft/dynamic rebalancing SD-026→029) plus screener/panel logic (SD-030→031) run platform-side, while the questionnaire handles substance. That routing, however, is expressed first as business_flow and routing_logic in flow_keys, then compiled via route_compiler to top-level qId fields (next/showIf/filterFrom…) that SuperJ's runtime executes — the very lineage Poseidon later reconstructs for path-aware analytics (see survey business flow routing India). Combined with DuckDB-on-Parquet (10–20× over Pandas), semantic cache 78% (0.88 threshold), three-layer verification at 99.1% and the 18-node report graph on Poseidon, the strategic framework is not a slide — it is a pipeline that knows Bharat's tiers, languages and devices and ships insights in hours, not weeks, from Hyderabad.

Your next step — reason, then route, then field. Start with a business decision, let Hercules classify study type, count stimuli, choose absolute vs relative, overlay pricing/trade-off if needed, apply India corrections, and propose the framework with alternatives. Confirm → business_flow + routing_logic in flow_keys → questionnaire with HOOK/CORE/CLOSE → route_compiler to qIds with deterministic validation → Quality Linter → SuperJ deploy via POST to SUPERJ_CREATE_SURVEY_AI_URL → completed → Poseidon ingest POST /api/v1/ingest/ (X-Secret-Key) → Parquet + 5-phase analytics + 18-node report. Every stage is traceable to methodology_catalog.py, framework_designs.md SD-001→044, or analytics_process.md. No defaults. No templates. Just reasoning that scales from a Tier 3 किराना conversation to a board-ready narrative — built in Hyderabad, priced from ₹0/month, rewarded for real Indians on the SuperJ app.