Where the numbers come from.

Real vendor pricing, verified for independence. Benchmarks are built from published price lists, vendor websites, trade show prospectuses, rate cards, and customer-contributed audit data. Every price is verified for independence and accuracy before inclusion.

Normalized at the door. Every observation is normalized to a per-unit basis at ingest, so quotes written against different quantities and formats compare on the same footing.

Three independent vendors, or nothing. A band publishes only when three or more independent vendors back it. That is not an editorial guideline — it is a constant in the engine (MIN_VENDORS_FOR_BAND = 3), so a thin band cannot slip through.

Below the gate, we say so. Items that have not cleared the three-vendor gate show as BUILDING, with no published numbers. An honest gap beats a shaky band.

Floor, typical, ceiling. Every published band carries a floor, a typical, and a ceiling, with per-tier vendor counts — so you can see how deep the data runs at the quantity you actually care about.

Confidence tiers. Every band carries a confidence rating driven by vendor depth: HIGH (10+ independent vendors with consistent pricing across quantity tiers), MEDIUM (5–9 vendors with moderate spread), or LOW (3–4 vendors, limited coverage). The rating ships with the number so you can judge how hard to lean on it.

Independence verification. Vendors must be operationally independent — shared ownership groups are identified and deduplicated so that a parent company with three subsidiaries counts as one vendor, not three. A band built on three prices from the same ownership group does not clear the gate.

Outlier handling. Extreme prices are excluded using statistical clustering before bands are computed. This means a single anomalous quote — a rush surcharge, a liquidation price, a data-entry error — does not warp the band that everyone else sees.

Quantity normalization. Vendors quote different quantities. The engine normalizes to per-unit pricing at each quantity tier so that a quote for 500 units and a quote for 50 units are compared at the right scale, not averaged together.

Market structure tags. Every item carries a market-structure classification: COMMODITY (many interchangeable vendors, tight pricing), OLIGOPOLY (a few dominant suppliers), SEMI-CONCENTRATED (moderate concentration), or MONOPOLY (single-source). The tag sets expectations — a wide band on a commodity item is a different signal than a wide band on an oligopoly item.

Models are labeled, loudly. When you ask about a spec we do not track, the spec sheet interpolates deterministically between benchmarked neighbors. The result is always labeled modeled, and modeled numbers are never blended into benchmarked bands — the two kinds of numbers never mix.

Quarterly snapshots. The data refreshes as quarterly snapshots. The current snapshot is 2026-05.

Counts are floored, never inflated. Every count on this site is live and floored — “over N” means at least N. Right now the benchmark holds over 76,300 observations across over 30,850 items.

A buyer, not a directory. The catalog is curated by a procurement buyer with twelve years in trade marketing — someone who has sat on your side of the quote.

Does not set prices or recommend vendors. The benchmark tells you where the market sits. What you negotiate and who you work with is your call.

Does not publish vendor identities. No band, audit result, or catalog page reveals which vendors contributed specific prices. Sources are aggregated, never individually attributed.

Does not use AI-generated price estimates. Every benchmark derives from real vendor pricing data. When the engine interpolates for an untracked spec, the result is deterministic math labeled MODELED — never a language-model guess.

Shows ranges, not point estimates. Target buy prices are always presented as floor-to-median ranges. Single-number targets invite antitrust risk and false precision; ranges reflect how markets actually work.

These rules are coded into the platform, not just written down. They govern how data enters, how it is stored, and what the system will and will not produce.

1. Buyer activity is permanently confidential. Who audited what, when, and at what price is never exposed — not to vendors, not to other buyers, not in aggregate.
2. No vendor-side revenue. ClearQuoter is funded by buyer subscriptions. Vendors cannot pay to appear, rank higher, or influence a band.
3. No AI free-form estimates. The engine uses deterministic math on real data. Language models handle document parsing and spec matching, never price generation.
4. ABSENT is a valid answer. If the data does not support a benchmark, the system says so. Bands are never fabricated to fill a gap.
5. Quarantine over delete. Suspect observations are quarantined, not destroyed. Corrections are reversible; data is never silently dropped.
6. Antitrust compliance by design. All target prices are ranges (floor to median), never single figures. The system cannot produce a point-estimate recommendation.
7. Three-vendor minimum is a constant. The minimum vendor count for a published band is a code constant, not a configurable threshold. It cannot be overridden per item.
8. Anonymous audits are ephemeral. Audits run without an account are processed in memory, shown back to the user, and discarded. Nothing is stored.
9. Contributions are anonymized at ingest. If a buyer opts in to contribute pricing data, the observation enters as item, price, quantity, date. The buyer's identity is stripped before it reaches the benchmark.
10. Counts are floored, never inflated. Every count shown on this site is a live floor — “over N” means at least N, never a rounded-up marketing number.