Bound parallel carrier lookups without serializing every quote

Project context

A fictional freight broker sees slow quote responses at dispatch handover. The service looks healthy in average-latency charts, yet a few long requests occupy every worker. Build a small synthetic quote service and controlled dependency stub before taking implementation tickets.

Setup prerequisites

• Create a local quote endpoint and a deterministic carrier-price stub; no repository or dataset is supplied.
• Use synthetic routes and a fixed workload manifest. Record runtime, machine resources and instrumentation settings.

Preceding work

Complete these dependencies, or supply their agreed outputs before taking this ticket.

• PLATENCY-101 · Write down the traffic mix before comparing quote timings
• PLATENCY-102 · Show quote latency percentiles alongside rejected and timed-out requests
• PLATENCY-103 · Separate quote queue time from carrier lookup time
• PLATENCY-105 · Stop expired quotes from holding carrier connections

Acceptance criteria

• Apply a configurable global in-flight limit and a bounded admission queue.
• Remove cancelled entries promptly and define fairness so a large quote cannot monopolize all new slots.
• Preserve the required carrier results or return an explicit incomplete-quote outcome; never silently price from a partial result.

Implementation constraints

• Compare limits using the fixed traffic mix and a stub with a declared service-time distribution.

Verification to include

• Run single-stop and multi-stop requests together and verify the configured concurrency ceiling.
• Overfill the queue, cancel its head and inject a carrier failure; verify forward progress and no leaked permit.

Deliverables

• Admission controller, mixed-traffic measurements and permit invariants

Rollout and recovery

Start with the current safe connection limit; queue rejection is observable and the feature switch restores the previous dispatch path.

Value of the work

For the engineer: Learn to distinguish queueing, dependency delay, CPU work and measurement mistakes using reproducible observations.

For the team: Produce a reviewable diagnosis and guarded changes that could guide a team investigating customer-visible latency.

Evidence boundaries

Outcome Evidence: Tests, patches, and runbooks are requested deliverables. They become Outcome Evidence only through a qualified Mission and immutable Evidence IDs.

Ownership Evidence: Independent adaptation must be observed under a declared verification policy and cite immutable Evidence IDs. Completing a planning ticket establishes no Ownership Evidence.