Operating method inside the SERPIFAI Query Sample Adequacy Framework
STRATA-6 Query Sampling Method
STRATA-6 is the six-control operating sequence for turning one decision into a reproducible query sample. It does not replace the parent adequacy framework; it makes the sampling work visible, reviewable and repeatable.
STRATA means Scope, Target, Resolve, Assemble, Test and Audit—the six controls that move a decision from an undefined keyword universe to a frozen, reviewable sample.
| Stage | Control | What happens | Output |
|---|---|---|---|
| S | Scope Scope the decision and required strata. | Translate the decision into required coverage across buyer language, intent, stage, geography, device and result environment. | Sampling frame. |
| T | Target Target candidate language from multiple sources. | Combine independent vocabulary sources rather than allowing one provider export to define the market. | Candidate pool. |
| R | Resolve Resolve semantic duplicates. | Remove repeated intent without erasing meaningful variants, locations or specialist questions. | Distinct intent set. |
| A | Assemble Assemble and weight the sample. | Choose queries by decision relevance and coverage, with every inclusion and exclusion recorded. | Documented sample. |
| T | Test Test coverage through pilot observations. | Check whether the selected strata behave as expected and expose missing or incomparable categories. | Coverage test. |
| A | Audit Audit, freeze and record the boundary. | Preserve the final version, exclusions, residual gaps, observation conditions and rerun triggers before analysis. | Reproducible frozen sample. |
Building a sample for a multi-location dental group
The business wants to compare visibility for implants, emergency care and routine dentistry across three cities. A defensible sample balances service lines, intent, location and commercial importance instead of copying the largest keyword export.
| Element | Observed condition | Interpretation |
|---|---|---|
| Emergency intent | “emergency dentist Leeds” | Urgent local conversion |
| Treatment evaluation | “dental implants Leeds cost” | High-consideration commercial |
| Provider comparison | “best implant dentist Leeds” | Choice and trust |
| Informational | “how long do implants last” | Education |
| Brand alternative | “alternative to clinic X” | Substitution |
| Repeat across cities | Leeds, York, Sheffield | Location comparability |
A defensible sample is designed from the decision outward and documented well enough for another reviewer to reproduce.
False positive to avoid: Removing zero-volume or low-volume terms can erase strategically important specialist and local questions.
Founder observation
While building SERPIFAI, I repeatedly encountered teams treating a keyword export as if it were a neutral sample. It is not. The export already reflects the provider’s vocabulary, filtering rules, database coverage and row limits. A defensible sample starts with the decision, defines the required strata and only then pulls language from multiple sources.
A defensible query sample is a reproducible set selected to represent buyer language, intent, funnel stage, geography, device and result environments relevant to one decision. It differs from a keyword export because every inclusion and omission has a reason tied to the sampling frame. The sample can support bounded conclusions without pretending to represent all search demand. A defensible sample is one input to SEO competitor analysis; competitor qualification and evidence review still determine what the comparison can support.
Status: Founder reviewed by Pero Baldjiev. Founder reviewed by Pero Baldjiev.
Review Controls
These controls let a second reviewer reconstruct the sampling frame, challenge the selection rule and decide whether the frozen sample can support the stated claim.
| Review control | Question or test | Required output |
|---|---|---|
| Sampling frame | Which buyer, problem, offer and stage must be represented? | Coverage blueprint. |
| Candidate sources | Which first-party, customer, sales, competitor and tool inputs contribute language? | Diverse pool. |
| Stratification | How will intent, stage, geography and device be balanced? | Representative structure. |
| Selection rule | Why is each query included or excluded? | Reproducibility. |
| Coverage gap | Which known needs remain unobserved? | Limitation register. |
Mnemonic check: A strong Assemble stage cannot repair a weak Scope. If Target is narrow, downstream Audit can record the bias, but it cannot remove it.
Where execution commonly fails
| Failure mode | Why it matters |
|---|---|
| Volume sorting | High-volume terms define the sample. |
| Single-source vocabulary | One tool or expert language dominates. |
| Duplicate intent | Many near-identical queries create false sample size. |
| Hidden exclusion | Commercially important low-volume terms disappear without record. |
Composite failure scenario — not a client case
When “low volume” removes the language that matters
In this composite scenario, a multi-location dental team exports several thousand keywords, sorts them by volume and removes 412 low-volume questions before analysis. The sample becomes cleaner and faster to process.
Six months later, a manual review of consultation notes shows that specialist implant questions had been omitted from the research frame. The problem is not that those phrases suddenly became high volume. The problem is that the sample never represented the decision it was meant to support: which content should serve high-consideration implant consultations?
Where the method failed: Scope never declared specialist demand as a required stratum. Target depended too heavily on one provider. Resolve treated rare variants as disposable. Assemble rewarded volume. Test never asked whether consultation-stage language remained covered. Audit had no exclusion rationale to preserve.
STRATA-6 response: reopen Scope, add specialist and substitution strata, Target first-party consultation language, rerun Test, and freeze the revised sample with exclusions and residual gaps visible.
The numbers and sequence above are illustrative. This composite scenario combines common sampling failures for teaching and does not report a verified SERPIFAI client outcome.
STRATA-6 coverage funnel
The model begins with the decision—not the export. STRATA makes every narrowing step explicit: Scope, Target, Resolve, Assemble, Test and Audit.
Decision input: state what the sample must allow you to decide.
- Scope: scope the decision and required strata.
- Target: target candidate language from multiple sources.
- Resolve: resolve semantic duplicates.
- Assemble: assemble and weight the sample.
- Test: test coverage through pilot observations.
- Audit: audit, freeze and record the boundary.
Export-led versus decision-led sampling
A keyword export can supply candidates, but it cannot define the sampling frame. The distinction changes both the sample and the conclusion it is permitted to support.
| Control | Export-led | STRATA-6 decision-led |
|---|---|---|
| Starting point | Keyword export | Decision to support |
| Coverage | Accidental and provider-shaped | Designed across required strata |
| Missing queries | Hidden or discovered late | Recorded as residual gaps |
| Bias | Unknown or implicit | Documented by source and selection rule |
| Reproducibility | Weak; the export can change without a stable frame | Strong; the frame, rules, exclusions and version are frozen |
Export-led
- Keyword export
- Filter
- Analysis
- Decision
The decision inherits the export’s vocabulary, omissions and row limits.
STRATA-6 decision-led
- Decision
- STRATA-6
- Frozen sample
- Analysis
- Bounded decision
The conclusion is limited to documented coverage, exclusions and residual gaps.
- Without STRATA-6
- Export defines the sample. Provider vocabulary and high-volume service terms dominate, exclusions remain invisible and the conclusion is easily overgeneralised.
- With STRATA-6
- Decision defines the strata. Multiple sources contribute language, missing coverage stays visible and the final conclusion remains bounded by the frozen frame.
Old recommendation: use the largest export as the market view. New evidence: urgent, informational, comparison and substitution questions were missing. New recommendation: expand the frame, rerun the pilot and limit the decision to the strata actually observed.
Conditions for valid use
You need a query sample tied to a defined decision, documented selection rules and visible coverage gaps.
It is not proof of causality, a universal reliability score or permission to extend evidence beyond its source, date and market context.
Source lineage, collection date, transformation, market context or reproducibility cannot be reconstructed.
Stage discipline: Scope defines the claim; Target finds the language; Resolve protects meaning; Assemble balances representation; Test challenges coverage; Audit preserves the boundary and the rerun trigger.
Additional method controls
Rerun triggers
- New first-party query data exposes a missing intent.
- Observed results reveal that selected queries are not comparable.
- The buyer, offer or market changes.
- A new answer format or local result environment becomes material.
- Coverage review shows one stratum dominates the conclusion.
Retain the previous conclusion, the changed evidence, the reviewer and the affected decision.
Completion checks
- Purpose: Resolve the decision described on this page rather than every adjacent research question.
- Application: Apply the six controls and the worked example; neither the branded name nor one score is sufficient.
- Boundary: Keep unknowns visible and refuse any conclusion that exceeds the observed evidence.
- Review: Reopen the sample when the scope, source record or execution conditions materially change.
A deliberately small sample can be stronger than thousands of keywords when it covers the decision strata and preserves observation quality.
Questions before execution
What is the practical output of STRATA-6?
The output is a frozen, reproducible query sample and a bounded interpretation for the stated decision—not a universal score or automatic recommendation.
Which mistake is most likely to distort the result?
The most important false positive is this: removing zero-volume or low-volume terms can erase strategically important specialist and local questions. That is a Scope and Target failure; a later Audit can document the omission but cannot retroactively restore the missing observations.
What should trigger a new review?
Review the frozen sample again when the buyer, offer, market, evidence record or execution conditions change enough to alter the decision.
Where STRATA-6 fits
STRATA-6 is the operating method. The SERPIFAI Query Sample Adequacy Framework remains the governing framework that determines whether the frozen sample is sufficient for the decision.
Coverage test: what changed when the sample frame expanded
The dental example can be stress-tested without pretending it is a permission-backed client result. The initial frame covers service and city combinations. The expanded frame adds urgent, cost, comparison, informational and substitution intent. That changes what the sample is allowed to support: from a narrow local-service comparison to an intent-stratified visibility review. The before and after sample record is a controlled comparison of coverage, not an outcome study.
| Sample state | Included evidence | Bias and coverage finding | Decision effect |
|---|---|---|---|
| Before expansion | Service × city terms only | Commercial local intent dominates; educational and provider-comparison needs are under-covered. | Do not generalise beyond service-location visibility. |
| After expansion | Emergency, treatment cost, provider comparison, informational and alternative intent across Leeds, York and Sheffield | Coverage is broader across urgency, evaluation, education and substitution, while the three-city boundary remains explicit. | The decision changed: an intent-stratified comparison becomes supportable. |
| Duplicate-intent stress test | A second near-identical implant-longevity phrase | No new stratum, geography or buyer question is added. | The decision remained unchanged; the duplicate is excluded and recorded. |
Source limitations must shape the sample
Google documents that anonymised queries are omitted from Search Console query tables and that row limits can also affect visible data. Google also describes Trends as a random sample of aggregated, anonymised and categorised searches. A defensible sample therefore combines sources, records exclusions and treats every source as incomplete rather than mistaking one export for the market.
- Google Search Central: Search Console performance-data filtering and limits
- Google Search Central: how Google Trends data is sampled
Decision hand-off: send the frozen sample to the execution-readiness gate; then compare candidate territories in opportunity size versus execution readiness, translate the winner into a constraint-led budget, and re-test the sample during the post-investment review when new first-party language appears.
Evidence boundary: this is a reproducible worked sampling record. The listed queries are examples; no search volume, ranking outcome or commercial value is claimed.
Methodological context and references
- Google Search Central — Creating helpful, reliable, people-first content
- Google Search Central — Link best practices
- Google Search Central — Article structured data
- UK Government Green Book guidance — Multi-Criteria Decision Analysis
- Center for Evidence-Based Management — Evidence-based decision resources
The references provide external context for this page. They do not certify the proprietary model or validate an unstated client outcome.
The governing rule
A query sample becomes defensible when another reviewer can reconstruct not only what was included, but why everything else was excluded.
Reconstructability strengthens auditability. It does not by itself prove that the sample represents all search demand or predicts rankings, traffic, leads or commercial outcomes.
Defensible Query Sample Field
How to read it: select a checkpoint to isolate that part of the method. The canvas shows the relationship; the inspector explains what the checkpoint changes in the decision.