The Purchase Mechanics™ Standard for Consumer Market Datasets

Abstract

The Purchase Mechanics Standard defines how Measured Choice separates product identity from retailer commerce, preserves missing and conflicting information, distinguishes source classes, and applies decision methods only when the evidence can support them. Its purpose is not to make consumer research look mathematical. Its purpose is to make unsupported precision difficult to introduce.

The standard also separates the consumer layer from the technical layer. A shopper should be able to understand the decision quickly. An auditor should be able to trace a material claim to a dated observation, source class, field definition, calculation and release.

1. Core operating rules

• Separate identity from commerce. A product configuration and a retailer offer are different records.
• Preserve provenance. Decision-critical facts retain source class, timestamp and material qualifications.
• Preserve missingness. Unknown price, lifespan, failure rate or repair cost stays unknown unless a modeled value is explicitly authorized and labeled.
• Preserve conflicts. Credible disagreements are carried into the decision rather than silently averaged away.
• Separate evidence types. Manufacturer claims, independent tests, owner signals, direct observations and modeled assumptions are not interchangeable.
• Use hard gates before preference trade-offs. A favorable score cannot rescue a product that fails a basic evidence, identity, safety or ownership-quality requirement.
• Allow zero winners. A category may finish with no overall Choice.
• Stop advanced modeling when inputs are not ready. More mathematics does not repair weak evidence.

2. Data architecture

Configuration Master

The canonical configuration record holds the relatively stable identity and specification layer. Internal configuration IDs remain stable even when retailer naming changes. GTIN, UPC, MPN and model aliases are identity evidence where available; they are not required to exist for every record.

Offers

Retailer offers are time-stamped commerce observations linked to a configuration. They may contain retailer, seller, price, regular price, promotion, shipping, availability, condition, source class and qualification status.

Raw Facts

Raw Facts preserve source-level observations before they are compressed into decision fields. They support traceability for warranty terms, cleaning instructions, hands-on tests, owner-duration signals, parts findings and similar evidence.

Decision Dataset

The Decision Dataset is deliberately narrower. It contains only fields that materially affect buyer value or recommendation confidence.

Release Manifest

The Release Manifest records population counts, observation counts, collection dates, method version, QA status, limitations and supersession history. Source: release summary It is the current reproducibility layer. Measured Choice does not currently claim a fully event-sourced or bitemporal production database.

3. Market population and freeze discipline

A category starts with written inclusion and exclusion rules covering geography, product function, form factor, configuration treatment and release timing. A release may then freeze the configuration set for analysis.

A frozen set is a versioned research frame, not a claim that no other product exists. For the nugget-ice pilot, the frozen U.S. frame contains 84 configurations. MCDD-ICE-001 release Post-freeze discoveries are recorded separately so the denominator is not silently changed after analysis starts.

4. Entity resolution and shared-platform evidence

Entity resolution decides whether multiple listings represent the same configuration, related variants, or different products. Strong evidence includes exact model numbers, GTIN/UPC/MPN, manufacturer documentation and exact retailer/manufacturer crosswalks. Supporting evidence can include dimensions, wattage, capacities, manuals and distinctive chassis details.

Similarity by appearance or dimensions alone is not enough to declare two brands identical.

5. Offer qualification and price statistics

A qualified current price is an observed, actionable offer—not a transaction price and not a permanent market norm. Displayed base price and promotional codes are stored separately when possible.

Snapshot statistics may be calculated over the qualified-price population. Worked release table In MCDD-ICE-001 Screening V0.1:

These values describe the dated snapshot. They do not prove that $239.99 is the long-run “normal” transaction price.

Longitudinal pricing

MODAL_14D_PRICE, PRICE_30D_RANGE and PRICE_VOLATILITY_30D remain in development. A single sale is not labeled “normal” until sufficient repeated observations exist and collection coverage is documented.

6. Missing data, conflicts and imputation

Missing decision facts remain missing. Conflicting credible values remain conflicts until resolved. The current standard does not use carry-forward, peer-class mean or hedonic imputation to manufacture current product facts.

If modeled values are introduced later, they must remain visibly different from observed values and must carry their own method version and assumptions.

7. Evidence provenance

Evidence strength is claim-specific. A manual can be stronger than a review for cleaning instructions; an independent test can be stronger than a specification sheet for observed first-hour output.

8. Public claim labels

Measured data is directly observed or computed from a defined release. Evidence signal is real and decision-relevant but not mature enough for a population statistic. Our read is a transparent Measured Choice interpretation. In development is not allowed to influence the published verdict.

Examples currently held in development include the proposed Biofilm Vulnerability Tier and normalized CPSC Incident Ratio. Current status

9. Decision derivations and Choice logic

Diagnostic ratios

Ratios such as price per claimed daily output can illuminate one dimension of value, but they cannot stand in for reliability, cleaning burden, support, footprint or buyer fit.

Pareto screening

Pareto analysis can identify non-dominated products under explicitly chosen dimensions. Non-dominated does not mean “indisputably best.” The result depends on the dimensions and evidence quality.

Hard gates and role frontiers

Products with materially different functions may remain on different buyer-specific frontiers. Identity, current-price qualification, utility floor, evidence quality, safety handling and lifecycle evidence can all operate as gates before a product is eligible for an overall Choice designation.

Stop rule

For MCDD-ICE-001 V0.1, 0 of 9 finalists cleared the full upstream evidence plus lifecycle/cost-to-own requirements. Decision release Downstream Monte Carlo uncertainty propagation, Kneedle and DEA were therefore stopped rather than used to force a winner.

Category result: ZERO CATEGORY-LEVEL CHOICE PRODUCTS — V0.1.

10. Reliability, repairability and cost to own

A long warranty can reduce buyer downside; it is not proof of population lifespan. Manufacturer durability testing can be useful evidence about the test performed; it is not automatically a field failure rate.

Repairability should state the path actually verified: open parts, support-only parts, authorized service, no open catalog found, or unknown. Evidence-status example “No catalog found” must not become “no parts exist.”

Cost-to-own models can include acquisition, energy, consumables, repair and replacement only when the inputs are supportable. If lifespan or failure probability is unknown, the output should remain partial or scenario-based.

11. Versioning and change control

Every decision release should identify the dataset ID, dataset version, schema version, method version, frozen population, observation counts, collection dates, known limitations, QA status and superseded release. Example release

Published decisions should also state their reopen triggers—for example material price movement, new recall information, stronger duration-qualified reliability evidence, repair-path changes, warranty changes or a new product that materially changes the frontier.

12. Publication and citation design

The technical layer should publish as stable canonical HTML first, with a PDF mirror and downloadable CSV/JSON when the data rights and quality support it. Finished papers should include title, paper ID, version, date, abstract, definitions, methods, results or worked examples, limitations, source mapping, change log and a short citation block.

Structured data should match visible content. No markup format can guarantee citation by search systems or AI tools; the strategy is to become the original source for facts Measured Choice actually measured or derived.

Download PDF mirror →

13. Source map

Method rules are normative statements defined by this paper. Worked examples and category facts must point to a versioned release or evidence class. The table below shows the current mapping used in this publication candidate.

The public source map intentionally exposes release facts and locators without publishing private notes, personal data, or copyrighted source material.

14. Current limitations

• The market frame is broad and versioned but is not claimed to be an exhaustive global census.
• Current prices are offer observations, not transaction data.
• 14/30-day longitudinal price metrics are not mature for the nugget-ice release.
• Shared-OEM resolution remains partial.
• Long-horizon reliability and population failure denominators remain weak across the finalist set.
• Cost-to-own modeling is incomplete for most finalists.
• Biofilm vulnerability scoring and normalized incident ratios are not published.
• The current system uses release snapshots, not a fully implemented bitemporal datastore.
• No measured operational error-rate thresholds are yet published for the pipeline.