{"schema_version":1,"experiment_id":"eoa_inverse_innovation_exp11_mechanism_context_external20_20260804","research_id":"eoa_inverse_innovation_exp11_external_scrutiny_20260804","cell_id":"negative_space_design__computer_science","opaque_id":"negative_space_design__computer_science__C","search_lanes":{"direct_problem":{"queries":["AI code completion interruptions developer flow cognitive load study inline suggestions","GitHub Copilot interruptions flow study developers suggestion rejection","programmers AI code suggestions automation bias premature acceptance study","software developer interruptions inline code completion study cognitive load"],"source_ids":["SRC1","SRC2","SRC5","SRC8"],"no_result_note":null},"closest_prior_art":{"queries":["GitHub Copilot disable inline suggestions temporarily snooze completion","VS Code inline suggestions trigger automatic explicit suppress suggestions","AI coding assistant snooze autocomplete suggestions","When to show a suggestion selectively hide code suggestions programmer state"],"source_ids":["SRC2","SRC3","SRC4","SRC5","SRC7"],"no_result_note":"No retained source implemented the complete package of automatic entry after rejection, test failure, or rapid editing; a bounded quiet interval; a visible state cue; immediate manual recall; and preservation of non-AI diagnostics."},"historical_terminology":{"queries":["mixed-initiative user interface interruption attention notification suppression Horvitz","context-aware notification interruption cost defer alerts breakpoint older terminology","code completion negotiated interruption programmer workflow","autocomplete content assist code completion disruption professional developers"],"source_ids":["SRC2","SRC6","SRC8"],"no_result_note":null},"products_practices_standards":{"queries":["site:code.visualstudio.com Snooze Inline Suggestions","site:devblogs.microsoft.com/visualstudio Better Control Copilot Code Suggestions","site:w3.org WCAG interruptions postponed suppressed","CodeArts do not disturb mode code completion"],"source_ids":["SRC3","SRC4","SRC5","SRC6","SRC7"],"no_result_note":null},"non_english_regional":{"queries":["KI Codevervollständigung Vorschläge pausieren Ablenkung Programmierer","補完 提案 一時停止 AI コーディング 集中","暂停 AI 代码补全 建议 专注 编辑器","autocomplétion IA mettre en pause suggestions code concentration"],"source_ids":["SRC5","SRC7"],"no_result_note":"The regional search found Chinese first-party documentation for a code-completion do-not-disturb mode and localized Microsoft documentation describing pause-sensitive or manual triggering; no non-English source located the full event-triggered cooldown composition."},"composition_subproblems":{"queries":["AI code completion cooldown after rejection rapid typing manual override","inline suggestion silence unresolved diagnostics preserve warnings","bounded snooze code completion visible status cue manual invoke","conditional suggestion display rejection telemetry programmer latent state"],"source_ids":["SRC2","SRC3","SRC4","SRC5","SRC6","SRC7"],"no_result_note":"The component mechanisms are documented separately, but the bounded search found no direct evaluation of their proposed rule-based composition against continuous eligibility, manual snooze, and relevance filtering."}},"sources":[{"source_id":"SRC1","title":"An Eye for AI: Eye-Tracking the Micro-Interruptions of GenAI Code Suggestions","url":"https://hasel.dev/publication/an-eye-for-ai-eye-tracking-the-micro-interruptions-of-genai-code-suggestions/","publisher":"Human Aspects of Software Engineering Lab, University of Zurich","date_or_year":"2026","source_type":"PRIMARY_RESEARCH","language":"English","claims_supported":["An eye-tracking study recorded gaze, suggestion interactions, and editing activity during AI-assisted programming.","Approximately half of generated suggestions were not viewed, and more than 75% of viewed suggestions were not used.","Reviewed suggestions averaged about 0.9 seconds of attention and were characterized as flow-disrupting micro-interruptions."]},{"source_id":"SRC2","title":"When to Show a Suggestion? Integrating Human Feedback in AI-Assisted Programming","url":"https://ojs.aaai.org/index.php/AAAI/article/view/28878","publisher":"AAAI Press","date_or_year":"2024","source_type":"PRIMARY_RESEARCH","language":"English","claims_supported":["Conditional Suggestion Display from Human Feedback uses acceptance, rejection, prompt, suggestion, and session signals to decide whether to display or withhold a suggestion.","A retrospective evaluation using data from 535 programmers found that a substantial fraction of suggestions predicted to be rejected could be hidden.","Programmer latent state materially affected acceptance; prior interaction data reported substantial time spent verifying suggestions and lower acceptance while editing or thinking about new code.","The authors warn that acceptance alone is an imperfect reward because developers sometimes accept without complete verification or reject without viewing."]},{"source_id":"SRC3","title":"Towards More Effective AI-Assisted Programming: A Systematic Design Exploration to Improve Visual Studio IntelliCode's User Experience","url":"https://par.nsf.gov/servlets/purl/10465130","publisher":"IEEE/ACM International Conference on Software Engineering; NSF Public Access Repository","date_or_year":"2023","source_type":"PRIMARY_RESEARCH","language":"English","claims_supported":["Seven laboratory studies involving 61 programmers examined 19 interfaces for AI-assisted code suggestions.","The resulting design principles include snoozability to prevent interruptions when the user intends, sufficient visibility to reduce premature commitment, and reduced visual clutter.","Participants sometimes accepted and undid suggestions to compare them with original code, demonstrating measurable rework around suggestion inspection.","The paper also supplies counterevidence: proactive inline display improves discoverability, so indiscriminate silence can reduce useful tool access."]},{"source_id":"SRC4","title":"Inline suggestions from GitHub Copilot in VS Code","url":"https://code.visualstudio.com/docs/editing/ai-powered-suggestions","publisher":"Microsoft Visual Studio Code","date_or_year":"2026-07-29","source_type":"FIRST_PARTY_PRODUCT","language":"English","claims_supported":["VS Code provides a Snooze control that temporarily disables inline suggestions in five-minute increments and a Cancel Snooze control that resumes them.","The snooze state is exposed through the Copilot status-bar menu and Command Palette commands.","Users can disable suggestions globally or by language, while organizations can manage some suggestion settings.","VS Code exposes a minimum display-delay setting and a collapsed presentation option for reducing distraction."]},{"source_id":"SRC5","title":"Better Control over Your Copilot Code Suggestions","url":"https://devblogs.microsoft.com/visualstudio/better-control-over-your-copilot-code-suggestions/","publisher":"Microsoft Visual Studio Blog","date_or_year":"2025-08-21","source_type":"FIRST_PARTY_PRODUCT","language":"English","claims_supported":["Microsoft reports user feedback that completions can appear too quickly, interfere with typing, and distract during deep work.","Visual Studio can debounce completions during rapid typing and show them only after a pause.","Automatic completions can be disabled while retaining explicit keyboard invocation.","Next-edit suggestions can be collapsed behind a visible margin indicator until the developer requests review."]},{"source_id":"SRC6","title":"Web Content Accessibility Guidelines (WCAG) 2.2","url":"https://www.w3.org/TR/WCAG22/","publisher":"World Wide Web Consortium","date_or_year":"2024","source_type":"OFFICIAL_STANDARD","language":"English","claims_supported":["Success Criterion 2.2.4 states that users can postpone or suppress interruptions except those involving an emergency.","The exception supports preserving safety-, security-, and data-loss-related warnings while allowing nonessential AI output to be suppressed.","The guidance supports user control and recoverability as guardrails, although WCAG conformance is scoped to web content rather than IDE research generally."]},{"source_id":"SRC7","title":"Huawei Cloud CodeArts Code Agent User Guide (IDE)","url":"https://support.huaweicloud.com/usermanual-codeartssnap/%E5%8D%8E%E4%B8%BA%E4%BA%91%E7%A0%81%E9%81%93%EF%BC%88CodeArts%EF%BC%89%E4%BB%A3%E7%A0%81%E6%99%BA%E8%83%BD%E4%BD%93%20%E7%94%A8%E6%88%B7%E6%8C%87%E5%8D%97%EF%BC%88IDE%EF%BC%89-pdf.pdf","publisher":"Huawei Cloud","date_or_year":"2026-05-30","source_type":"FIRST_PARTY_PRODUCT","language":"Chinese","claims_supported":["Huawei CodeArts documents a do-not-disturb mode that hides code-completion and conversation-related prompts to help users focus on coding.","The same product supports real-time completion and shortcut-triggered code generation, establishing regional first-party use of silence and explicit invocation controls.","The documented mode is user-configured rather than an automatically entered bounded cooldown."]},{"source_id":"SRC8","title":"An Empirical Investigation of Code Completion Usage by Professional Software Developers","url":"https://ppig.org/files/2015-PPIG-26th-Marasoiu.pdf","publisher":"Psychology of Programming Interest Group","date_or_year":"2015","source_type":"PRIMARY_RESEARCH","language":"English","claims_supported":["The study frames code completion as a negotiated interruption that users may ignore, engage with, or retrigger manually.","Only 40.1% of observed completion windows led to acceptance; an industrial sample of 10,000 completions had a 44.2% acceptance fraction.","Unexpected, missing, or disappearing completions caused disruptions and recovery actions, but completion was also heavily used for speed, correctness, exploration, and debugging.","The findings provide older terminology and counterevidence against treating every completion as net harmful."]}],"problem_evidence":{"status":"PARTLY_SUPPORTED","finding":"The interruption component is externally supported: eye tracking identifies suggestion review as a micro-interruption, telemetry research shows substantial verification cost and strong dependence on developer state, and first-party product documentation acknowledges distraction during rapid typing and deep work. Older completion research also documents dismissals and recovery behavior. However, the stronger claims that individually relevant suggestions cumulatively impair problem-state formation, cause premature acceptance, or increase downstream defects and completion time have not been directly established by the retained sources.","source_ids":["SRC1","SRC2","SRC3","SRC5","SRC8"],"uncertainty":"Effects are heterogeneous: short or timely completions can preserve flow, improve discoverability, and support correctness. Existing studies do not isolate timing from suggestion relevance, task difficulty, length, developer experience, or accessibility needs."},"adopter_evidence":{"status":"SUPPORTED","finding":"An identifiable adoption path exists. Individual developers already control local snooze, delay, manual invocation, and do-not-disturb settings; VS Code documentation also identifies organization administrators as authorizers for managed suggestion features. This supports a developer-authorized local pilot with engineering or repository approval for project work.","source_ids":["SRC4","SRC5","SRC7"],"uncertainty":"The sources establish product-level user and organizational control, not a universal rule for who owns telemetry or experimentation authority in every repository."},"implementation_evidence":{"status":"PARTLY_SUPPORTED","finding":"Most components are technically demonstrated in adjacent systems: timed snooze with a visible cancel control, rapid-typing debounce, explicit invocation, collapsed suggestions, conditional withholding based on rejection and session context, and a regional do-not-disturb mode. The retained literature does not demonstrate the full state machine, especially automatic fixed cooldowns after rejection or unresolved test failure combined with immediate manual recall and preservation of diagnostic channels.","source_ids":["SRC2","SRC3","SRC4","SRC5","SRC6","SRC7"],"uncertainty":"It remains unknown whether simple rules can classify receptive states accurately enough, whether manual invocation remains available during every suppression mode, and whether diagnostic-triggered silence would withhold useful fixes at precisely the wrong time."},"prior_art":{"disposition":"ADJACENT_PRIOR_ART","closest_analogues":[{"name":"Conditional Suggestion Display from Human Feedback (CDHF)","source_ids":["SRC2"],"same_problem":true,"same_causal_lever":true,"overlap":"Uses acceptance, rejection, session telemetry, and inferred developer state to decide when suggestions should be displayed or withheld, explicitly targeting verification time and mistimed suggestions.","remaining_difference":"It uses learned utility and acceptance thresholds rather than a transparent, bounded cooldown after specified events; it does not supply the proposed quiet-state cue, guaranteed manual override, preserved-diagnostics rule, or prospective crossover evidence."},{"name":"VS Code Snooze Inline Suggestions","source_ids":["SRC4"],"same_problem":true,"same_causal_lever":true,"overlap":"Provides bounded suggestion silence, a visible status-bar control, five-minute increments, and immediate cancellation without deleting the underlying capability.","remaining_difference":"Entry is manual rather than automatically triggered by rejection, rapid edits, or unresolved failure, and the documentation does not establish causal effects on interruption, rework, quality, or control."},{"name":"Visual Studio completion timing and on-demand controls","source_ids":["SRC5"],"same_problem":true,"same_causal_lever":true,"overlap":"Suppresses completion during rapid typing through debounce, permits manual-only invocation, and collapses unsolicited next-edit content behind a visible indicator.","remaining_difference":"The controls are persistent settings or presentation policies, not a short event-triggered cooldown with automatic recovery after rejection or test failure."},{"name":"Snoozability design principle for inline code suggestions","source_ids":["SRC3"],"same_problem":true,"same_causal_lever":true,"overlap":"Explicitly recommends that users be able to snooze inline suggestions to prevent interruptions and pairs this with visibility and premature-commitment safeguards.","remaining_difference":"It is a design principle within a broader interface study and does not evaluate automatic context-sensitive silence or the proposed outcome measures."},{"name":"Huawei CodeArts do-not-disturb mode","source_ids":["SRC7"],"same_problem":true,"same_causal_lever":true,"overlap":"Hides completion and conversation prompts to protect coding focus while retaining the assistant product.","remaining_difference":"It is a user-enabled broad mode, with no documented bounded duration, event detector, selective preservation policy, or comparative evaluation."}],"contrastive_claim_remaining":"Compared with continuous eligibility, relevance filtering, rapid-typing debounce, and manual snooze, a transparent rule-based cooldown entered automatically after recent rejection or a rapid-edit burst will reduce involuntary suggestion inspection, dismissals, and accept-then-undo behavior and improve reported control, without worsening task time, correctness, or access to critical diagnostics; unresolved-test-failure entry should remain a separately randomized exploratory condition because useful fixes may be especially valuable then.","contrastive_claim_falsifier":"In an adequately instrumented within-person crossover study among developers who exhibit timing-related interruption, the automatic bounded-cooldown condition produces no preregistered improvement in attention shifts, dismissals, rework, quality, or experienced control, or any improvement is offset by slower completion, missed critical cues, reduced task success, or frequent manual overrides.","confidence":"HIGH","search_limitations":"This was a bounded public-web search retaining exactly eight direct sources across six adversarial lanes. It was not an exhaustive search of patents, source repositories, proprietary product telemetry, unpublished experiments, or every regional product. Product behavior may change after 2026-08-04."},"researchability_gates":{"externally_supported_problem":{"status":"PASS","rationale":"Multiple independent empirical and first-party sources support a real timing, verification, and distraction problem, even though downstream defect and premature-acceptance effects remain hypotheses.","source_ids":["SRC1","SRC2","SRC5","SRC8"]},"identifiable_adopter_or_authorizer":{"status":"PASS","rationale":"The participating developer can authorize local use, while documented organization-managed settings support an engineering or repository authority role for project deployment.","source_ids":["SRC4","SRC5","SRC7"]},"distinct_testable_incremental_claim":{"status":"PASS","rationale":"Existing work supplies manual snooze, debounce, conditional filtering, and do-not-disturb controls, but leaves a falsifiable comparison of automatic event-triggered bounded cooldowns against those rivals.","source_ids":["SRC2","SRC3","SRC4","SRC5","SRC7"]},"bounded_next_evidence_step":{"status":"PASS","rationale":"A local, opt-in, within-person crossover pilot can compare baseline, manual snooze, and fixed post-rejection or rapid-edit cooldowns on non-production tasks using preregistered behavioral, quality, time, and control outcomes.","source_ids":["SRC1","SRC2","SRC3","SRC4"]},"no_unresolved_safety_or_authority_stop":{"status":"PASS","rationale":"The proposed local flag, immediate override, exclusion of compiler/security/data-loss warnings, and prohibition on personnel evaluation bound the principal risks. WCAG's interruption exception reinforces preserving urgent warnings. Consent and minimization remain required for telemetry but do not create an unresolved stop for a local pilot.","source_ids":["SRC2","SRC4","SRC6"]},"adequate_search_evidence":{"status":"PASS","rationale":"All six required lanes were searched; exactly eight retained sources were opened. They span six independent publisher groups and include primary research, an official standard, and multiple first-party product sources, including a Chinese regional source and pre-GenAI terminology.","source_ids":["SRC1","SRC2","SRC3","SRC4","SRC5","SRC6","SRC7","SRC8"]}},"strict_success":true,"screen_survival":true,"remaining_research_value":"MODERATE","recommended_next_step":"Preregister and run an opt-in local crossover pilot on non-production tasks. Compare continuous eligibility, the editor's existing manual snooze or debounce control, and automatic fixed cooldowns after rejection and rapid-edit bursts; randomize the unresolved-test-failure trigger separately. Log presentation, gaze-proxy or focus shifts, dismissal, explicit invocation, acceptance, undo, task time, tests, defects, and control ratings; preserve all non-AI diagnostics, minimize telemetry, and halt on missed critical cues, quality degradation, or unavailable override.","world_novelty_boundary":"The bounded search supports only a contrastive research gap around prospective evaluation of transparent, event-triggered bounded silence. It cannot establish world novelty, patentability, freedom to operate, market size, routine production availability across all editors, or realized impact."}