Quoting in a second-score market

Quote from marginal cost and live utilization, then use the disclosed price-setter and score feedback to tune supply.

Owner

Supply

Version

beta-2026.07

Verified

2026-07-13

Version second-score/3rewards low expected cost and strong measured service. In a genuine-rival clear, the winning tariff scales toward an independent rival's score and is customer-capped at the cheapest independent alternative. A speed-advantaged winner can clear at its own ask, and a book with no independent rival can use a disclosed reserve. Quote executable marginal cost, but read the exact incentive boundaries on the second-score page.

Floors from marginal cost, markup from the curve

Your floor should be measured, not guessed: GPU dollars per hour divided by sustained tokens per hour at target utilization, separately for input and output. That number is a true floor, not a list price, because the reference quoter marks it up as your hardware fills. With BACKEND_METRICSpointed at vLLM's Prometheus endpoint, the quoter reads live utilization and applies a curve from 1.0x at idle to 2.5x near saturation, then quotes zero capacity above TARGET_UTILIZATION (default 0.85) so you shed load instead of degrading latency. Cheap when empty, expensive when busy, absent when full: market-maker inventory management, automated.

TTL and capacity are part of the bid

Quotes carry a TTL of 5 to 30 seconds and are refreshed every 2 seconds or so, which is how you reprice: the new quote atomically replaces the old one, and withdrawing is just letting the TTL lapse. A quote also needs at least 2.5 seconds of TTL runway to enter an auction, because the award, payment, and dispatch pipeline spends real time; quotes excluded this way are told so (filtered-ttl-expiring) in your auction feedback. capacity_rps is firm: the router tracks your in-flight requests and never routes past it, so quote the headroom you actually have.

The reference quoter can replace that scalar with one signed deployment snapshot containing realtime sequence, prefill-token, and decode-token budgets plus context-bucket TTFT/TPOT p95 commitments. Model classes on the same deployment share those reservations. Enable this in simple mode with QUOTER_MODE=simple, PROVIDER_REGION, andDEPLOYMENT_ID; the quoter conservatively derives token budgets from sequence capacity and maximum context. UseQUOTER_MODE=advanced only when every token budget and context-bucket p95 is backed by measurements—the quoter fails closed if any commitment field is absent. The router audits published p95 values against its own clock.

Quotes are firm

A standing quote is a signed commitment. Refusing a routed request that matched your live quote (last look, in trading terms) is a protocol fault with escalating controls such as warnings and traffic caps. Any funded collateral penalty remains Proposed. Short TTLs exist precisely so you are never stuck honoring a stale price; use them instead of reneging. The full fault table is on the good standing page.

Capabilities are signed economics

Optional quote fields make performance features executable rather than inferred. context_protocol, context_cache_ttl_s, and delta_handoff declare Passport continuity; speculative and speculative_cancel_ms explicitly authorize a bounded first-output hedge. The fields are part of the provider signature. Omitting them simply keeps the quote eligible for ordinary full-context, non-speculative requests.

Thin and related-provider books clear differently

Only economically independent rivals may set or cap a winner's price. If none exists, the router may use a seven-day, group-hardened median winning bid as a phantom-reserve, but only after minimum independent-group and payer thresholds are met. Without that evidence the winner clears at its own ask. Every receipt namesrunner-up, phantom-reserve, own-bid, or session-lock, so an uncontested clear never masquerades as a competitive one.