{"schema_version":1,"experiment_id":"eoa_inverse_innovation_exp09_archetype_breadth150_20260804","research_id":"eoa_inverse_innovation_exp09_light_prior_art_20260804","cell_id":"synchronized_release_dampening__computer_science","search_lanes":{"direct_problem_and_intervention":{"queries":["CI dependency completion thundering herd blocked workflows environment provisioning","CI build stampede duplicate environment preparation request coalescing","\"thundering herd\" \"continuous integration\" scheduler","\"thundering herd\" CI pipeline jobs resource acquisition"],"source_ids":["SRC1"],"no_result_note":"No direct public source was found describing pull-request workflows simultaneously unblocked by one dependency-ready event and then dampened with the proposal's full controller."},"synonyms_and_historical_terms":{"queries":["CI thundering herd stampede cache singleflight admission control","build system request coalescing duplicate work","singleflight token bucket admission control cohort release CI workflows","site:aws.amazon.com/builders-library token bucket jitter overload admission control"],"source_ids":["SRC1","SRC4"],"no_result_note":null},"products_practices_and_standards":{"queries":["site:docs.github.com actions concurrency control workflows official","site:buildkite.com/docs concurrency groups queue CI official","site:bazel.build remote caching avoid duplicate builds official","site:docs.gitlab.com resource_group process_mode CI concurrency official"],"source_ids":["SRC2","SRC3"],"no_result_note":null},"component_combination":{"queries":["CI workflow singleflight token bucket cohort release shared environment snapshot","dependency build completion fanout rate limit duplicate environment preparation","CI duplicate builds coalescing identical work build system","site:cloud.google.com token bucket rate limiting jitter thundering herd official"],"source_ids":["SRC2","SRC3","SRC4"],"no_result_note":"The components were individually established, but the bounded search did not find their complete combination keyed to a CI dependency-ready boundary."}},"sources":[{"source_id":"SRC1","title":"Continuous Pipelines at Google","publisher":"Google","url":"https://static.googleusercontent.com/media/research.google.com/en//pubs/archive/43790.pdf","source_type":"PRIMARY_RESEARCH","claims_supported":["Google operational experience identifies a thundering-herd problem in which thousands of pipeline workers can start immediately and overwhelm dependent servers, shared cluster services, and networking infrastructure.","Simultaneous pipelines can create peak load on a common shared resource even when the problem is not apparent from an isolated or aggregated view.","Pipeline fragility can manifest as resource exhaustion, missed deadlines, and scheduling delays."]},{"source_id":"SRC2","title":"Controlling concurrency","publisher":"Buildkite","url":"https://buildkite.com/docs/pipelines/configure/workflows/controlling-concurrency","source_type":"FIRST_PARTY_PRODUCT","claims_supported":["A production CI product provides keyed concurrency groups and limits to govern access to shared resources.","Concurrency groups queue excess jobs and admit new jobs as active jobs finish.","Buildkite supports concurrency gates around parallel work and exposes concurrency wait time, making admission delay measurable.","The documented mechanism is a configured shared-resource gate, not dependency-event-specific release dampening or duplicate preparation coalescing."]},{"source_id":"SRC3","title":"Remote Caching","publisher":"Bazel","url":"https://bazel.build/remote/caching","source_type":"OFFICIAL_GUIDANCE","claims_supported":["Bazel allows developers and CI systems to share reproducible build outputs through an action-keyed remote cache instead of rebuilding them on every machine.","The cache maps action hashes to results and stores content-addressed outputs, establishing keyed reuse of immutable or reproducible work.","The guidance warns about cache security, write authority, poisoned results, concurrent input changes, and environment differences, supporting the proposal's isolation and immutability safeguards."]},{"source_id":"SRC4","title":"Timeouts, retries and backoff with jitter","publisher":"Amazon Web Services","url":"https://aws.amazon.com/builders-library/timeouts-retries-and-backoff-with-jitter/","source_type":"OFFICIAL_GUIDANCE","claims_supported":["Correlated clients retrying at the same time can recreate contention or overload.","Jitter spreads correlated work over time, including delayed and scheduled work whose brief peaks may be hidden by aggregate metrics.","AWS describes token-bucket limiting as a way to constrain retry load to a controlled rate."]}],"problem_evidence":{"status":"PARTLY_SUPPORTED","finding":"The general diagnosis is visible: primary operational evidence documents pipeline thundering herds, short peak loads on shared resources, resource exhaustion, and delay, while CI products explicitly queue jobs to protect shared resources. Keyed reuse of reproducible build outputs and temporal dispersion are also established practices. The retained sources do not directly demonstrate the proposal's narrower empirical premise that one upstream dependency-ready event simultaneously releases many pull-request workflows, creates duplicate base-environment preparations, and causes broker timeouts despite spare daily capacity.","source_ids":["SRC1","SRC2","SRC3","SRC4"]},"closest_prior_art":[{"name":"Buildkite concurrency groups and concurrency gates","source_ids":["SRC2"],"overlap":"Keys CI jobs into shared-resource groups, queues excess jobs, admits bounded parallel work, preserves an ordering policy, and measures concurrency wait time.","remaining_difference":"It does not detect a dependency-ready release population, coalesce one shared environment preparation per dependency and base-revision key, price admission from broker recovery signals, or release bounded cohorts with deadlines and a capped escape lane."},{"name":"Bazel action-keyed remote caching","source_ids":["SRC3"],"overlap":"Reuses reproducible outputs across developers and CI machines through action hashes and a content-addressed store, avoiding repeated completed work.","remaining_difference":"A cache hit reuses already-produced output; the guidance does not specify single-flight suppression of simultaneous cache misses combined with event-aware workflow admission and cohort dispersion."},{"name":"AWS token-bucket limiting and jitter","source_ids":["SRC4"],"overlap":"Limits offered load and spreads correlated arrivals that otherwise reproduce contention or overload; it also recognizes that short spikes can disappear in aggregate metrics.","remaining_difference":"The guidance is general distributed-systems load control rather than a CI controller keyed to dependency version and repository base revision, and it does not add shared environment preparation or CI-specific fairness deadlines."},{"name":"Google pipeline thundering-herd and shared-resource load analysis","source_ids":["SRC1"],"overlap":"Identifies correlated pipeline worker starts, overload of dependencies and shared infrastructure, and simultaneous-pipeline peak patterns.","remaining_difference":"It describes the problem class and continuous-pipeline alternatives, not the proposed dependency-ready controller combining single-flight preparation, capacity feedback, bounded cohorts, repository fairness, and an emergency lane."}],"prior_art_disposition":"ADJACENT_PRIOR_ART","contrastive_claim_remaining":"For CI workloads that are already admitted but become runnable together at one dependency-ready boundary, a controller combining dependency-version/base-revision single-flight preparation, broker-feedback-priced tokens, bounded cohort reactivation, per-repository fairness deadlines, and a capped security lane will reduce event-window lease-arrival peaks, duplicate preparations, broker queue depth, and infrastructure failures more than ordinary concurrency gating or cached-output reuse alone, without violating maximum-wait and isolation constraints.","contrastive_claim_falsifier":"The claim is falsified if event-aligned traces show no material dependency-ready arrival spike or shareable duplicate preparation, or if a controlled replay shows that the combined controller fails to outperform a matched conventional keyed-concurrency baseline on collision and duplicate-work measures while satisfying preset completion-latency, maximum-wait, fairness, priority-cap, and isolation limits.","gates":{"adequate_source_search":{"status":"PASS","rationale":"The bounded search covered direct phrasing, thundering-herd and stampede terminology, CI concurrency products, build-output reuse, token buckets, jitter, and component combinations. Four opened sources span four publishers and include primary operational research plus official and first-party documentation.","source_ids":["SRC1","SRC2","SRC3","SRC4"]},"supported_problem":{"status":"PASS","rationale":"The broader correlated pipeline-load and shared-resource collision problem is directly supported, while the exact dependency-ready and duplicate-environment scenario remains unverified; therefore the problem evidence is partly supported rather than fully supported.","source_ids":["SRC1","SRC2","SRC3","SRC4"]},"distinct_testable_claim":{"status":"PASS","rationale":"Although every major mechanism has adjacent prior art, the remaining claim distinguishes the event boundary, composite control policy, comparator, measurable outcomes, and fairness and isolation constraints.","source_ids":["SRC1","SRC2","SRC3","SRC4"]},"bounded_next_test":{"status":"PASS","rationale":"A staging replay of at most 500 descriptors, comparing baseline, conventional keyed concurrency, and the combined controller, is bounded and can measure arrival peaks, queue depth, duplicate preparations, latency, maximum wait, service order, and priority-token use without changing production.","source_ids":["SRC1","SRC2"]},"no_obvious_safety_or_authority_stop":{"status":"PASS","rationale":"The proposed first step is staging-only and preserves required tests and merge authority. Immutable, repository-scoped reuse, restricted cache-write authority, maximum-wait rollback, and priority caps address the most obvious cache-poisoning, trust-boundary, starvation, and priority-capture hazards.","source_ids":["SRC3"]}},"screen_survival":true,"world_novelty_boundary":"This bounded public-web screen found adjacent components and practices but no retained source describing the complete dependency-ready CI controller. It cannot establish world novelty, patentability, market size, expert acceptance, or realized value; unindexed internal systems, patents, products, and differently described implementations may still collide."}