Distinguish standards, empirical studies, practitioner heuristics, research proposals, and project preferences.
Established guidance remains useful without being a universal law; recent publication does not establish superiority.
The interview established responsiveness-first preference, context-derived resource exceptions, whole-task effort, accessible task completion, evidence-backed challenges to precedent, and proportional agent judgment.
Those preferences govern this skill's recommendations; they are not findings proven by the cited literature.
Project goals versus informed choice remained a project-priority decision in the interview, including possible conflicts with truthfulness.
This records a preference, not research consensus, legal advice, or blanket permission for future deceptive actions.
Unknown retention stakes require evidence or clarification before weakening protection; the interview's disposable-data exception was narrower than general permission to discard work.
Established interaction principles
ISO 9241-110:2020, Interaction principles describes a technology-independent framework for interactive systems.
The public abstract informed scope; the full standard was not inspected.
Nielsen, 10 Usability Heuristics provides practitioner guidance on state visibility, mental-model fit, control, consistency, error prevention, recognition, efficiency, relevance, and recovery.
Heuristics identify plausible problems; they do not prove that a particular redesign improves actual tasks.
Shneiderman, Eight Golden Rules emphasizes consistency, universal usability, feedback, closure, prevention, reversal, control, and reduced memory burden.
The author explicitly requires interpretation and validation in context.
Do not turn its historical memory-capacity shorthand into a fixed limit on interface options.
Nielsen, Progressive Disclosure addresses learnability versus power by deferring secondary complexity while preserving obvious access.
This is not evidence that every task benefits from hiding information or adding stages.
Medium fit
Command Line Interface Guidelines is a practitioner guide connecting human use with composability, discovery, conventions, robustness, and sufficient feedback.
Its philosophy transfers; terminal-specific mechanics do not become requirements for every medium.
Apple, Designing for iPadOS relates interaction to ergonomics, viewing distance, multiple input modes, sustained work, and inter-app capabilities.
Use it as evidence that actual capabilities matter, not as a cross-platform style authority.
Chalmers and Galani, Seamful Interweaving, DIS 2004 argues that selectively exposing technological differences and limitations can support understanding and appropriation.
Its mixed-reality context does not establish that every implementation detail should become visible.
Portable application:
CLI: exploit composition, streams, repeatable invocation, searchable output, and surrounding tools where useful.
TUI: exploit keyboard-driven spatial work and incremental updates while accounting for terminal capabilities, remote transport, resize, and assistive access.
Mobile: account for reach, touch precision, interruptions, battery, and variable connectivity without assuming touch is the only input.
Desktop: exploit sustained work, precise input, simultaneous views, and inter-application workflows where they help.
Web: preserve meaningful navigation, addressing, browser controls, and operation across variable devices and networks.
These categories overlap; a remote terminal, mobile browser, and keyboard-equipped tablet invalidate simplistic platform stereotypes.
Machine semantics and resource constraints
CLI Guidelines: Output distinguishes human formatting from stable machine-oriented output and discusses explicit structured formats.
This supports separate presentation and semantic contracts, not a universal requirement for JSON or an external API.
Kleppmann et al., Local-first software, 2019 connects architecture to responsiveness, offline continuity, ownership, and collaboration.
Local operation can remove network dependencies from immediate work while introducing synchronization, conflict, storage, and lifecycle costs.
Its historical implementation assessments are not current product recommendations.
Internal data should preserve meaning independently of presentation; formatting, localization, rounding, and summarization must not become accidental storage semantics.
Mechanical sympathy requires reasoning about the actual substrate and workload, not assuming fewer allocations, more caching, or more parallelism always improves interaction.
Latency, throughput, memory, bandwidth, energy, and correctness can conflict.
Batching may increase throughput while delaying feedback; speculation may reduce waiting while wasting transfers or showing stale state.
Responsiveness-first is the user's preference, not permission for unlimited speculation.
Infer relevant limits from domain and project evidence; ask when the trade-off remains consequential and unresolved.
Time and unavoidable delay
Nielsen, Response Times: The 3 Important Limits offers approximate 0.1 s, 1 s, and 10 s guidance for perceived immediacy, flow, and sustained attention.
These are historical rules of thumb, not universal acceptance thresholds.
Ng et al., Designing for Low-Latency Direct-Touch Input, UIST 2012 found perceptible benefits well below 10 ms in studied direct-touch interactions.
The study also explored immediate low-fidelity feedback followed by more complete rendering.
This does not prescribe that every application needs specialized hardware or a 1 ms budget.
Interaction to Next Paint is a current web responsiveness metric covering input delay, event processing, and presentation through the next paint.
Its good-responsiveness threshold of 200 ms is a web measurement convention, not a universal human limit or proof of eventual task completion.
RAIL remains useful for distinguishing response, animation, idle work, and loading.
Its own documentation points to newer Core Web Vitals guidance; do not describe it as the latest web acceptance model.
Evaluate acknowledgment, useful results, and durable completion separately.
An immediate pending state is honest feedback; an unsupported success claim is not a latency optimization.
Remove unnecessary work and critical-path dependencies before adding waiting treatments.
When waiting remains, preserve other useful work, orientation, truthful progress, and meaningful control.
Progress estimates need evidence; unknown total work does not justify invented percentages.
Expose actionable boundaries rather than every internal detail.
Accessibility and agency
WCAG 2.2 provides testable accessibility criteria for web content.
Compliance alone does not establish overall usability or address every cognitive need.
WCAG2ICT explains applying WCAG concepts to non-web software, including text, command-line, and terminal applications.
It is informative guidance, not an independent conformance standard; the linked editor's version can change.
Text-only output and keyboard operation do not automatically make an interface accessible.
W3C cognitive accessibility supplemental guidance adds objectives concerning understanding, finding information, avoiding mistakes, focus, memory, and support.
Supplemental guidance is not identical to normative WCAG success criteria.
W3C TR and WAI fetches were blocked during research; indexed excerpts were available, and WCAG2ICT was inspected through its editor's publication.
FTC, Bringing Dark Patterns to Light, 2022 documents how interface design can impair autonomy, decision-making, and choice.
Improved business metrics do not by themselves demonstrate improved user outcomes.
Report project preferences and affected-user consequences distinctly from research conclusions.
Adaptive and automated interaction
Amershi et al., Guidelines for Human-AI Interaction, CHI 2019 synthesizes 18 guidelines, evaluated with 49 practitioners across 20 AI-infused products.
Relevant concerns include capability boundaries, correction, adaptation, feedback, and user control.
This is not proof that every interface should use conversation or automation.
Balance adaptation against predictable behavior, acquired skill, accessible operation, and ability to inspect or override consequential automation.
# Research and interpretation
## Contents
- [Evidence and authority](#evidence-and-authority)
- [Established interaction principles](#established-interaction-principles)
- [Medium fit](#medium-fit)
- [Machine semantics and resource constraints](#machine-semantics-and-resource-constraints)
- [Time and unavoidable delay](#time-and-unavoidable-delay)
- [Accessibility and agency](#accessibility-and-agency)
- [Adaptive and automated interaction](#adaptive-and-automated-interaction)
## Evidence and authority
- Distinguish standards, empirical studies, practitioner heuristics, research proposals, and project preferences.
- Established guidance remains useful without being a universal law; recent publication does not establish superiority.
- The interview established responsiveness-first preference, context-derived resource exceptions, whole-task effort, accessible task completion, evidence-backed challenges to precedent, and proportional agent judgment.
- Those preferences govern this skill's recommendations; they are not findings proven by the cited literature.
- Project goals versus informed choice remained a project-priority decision in the interview, including possible conflicts with truthfulness.
- This records a preference, not research consensus, legal advice, or blanket permission for future deceptive actions.
- Unknown retention stakes require evidence or clarification before weakening protection; the interview's disposable-data exception was narrower than general permission to discard work.
## Established interaction principles
- [ISO 9241-110:2020, Interaction principles](https://www.iso.org/standard/75258.html) describes a technology-independent framework for interactive systems.
The public abstract informed scope; the full standard was not inspected.
- [Nielsen, 10 Usability Heuristics](https://www.nngroup.com/articles/ten-usability-heuristics/) provides practitioner guidance on state visibility, mental-model fit, control, consistency, error prevention, recognition, efficiency, relevance, and recovery.
Heuristics identify plausible problems; they do not prove that a particular redesign improves actual tasks.
- [Shneiderman, Eight Golden Rules](https://www.cs.umd.edu/users/ben/goldenrules.html) emphasizes consistency, universal usability, feedback, closure, prevention, reversal, control, and reduced memory burden.
The author explicitly requires interpretation and validation in context.
Do not turn its historical memory-capacity shorthand into a fixed limit on interface options.
- [Nielsen, Progressive Disclosure](https://www.nngroup.com/articles/progressive-disclosure/) addresses learnability versus power by deferring secondary complexity while preserving obvious access.
This is not evidence that every task benefits from hiding information or adding stages.
## Medium fit
- [Command Line Interface Guidelines](https://clig.dev/) is a practitioner guide connecting human use with composability, discovery, conventions, robustness, and sufficient feedback.
Its philosophy transfers; terminal-specific mechanics do not become requirements for every medium.
- [Apple, Designing for iPadOS](https://developer.apple.com/design/human-interface-guidelines/designing-for-ipados.md) relates interaction to ergonomics, viewing distance, multiple input modes, sustained work, and inter-app capabilities.
Use it as evidence that actual capabilities matter, not as a cross-platform style authority.
- [Chalmers and Galani, Seamful Interweaving, DIS 2004](https://www.dcs.gla.ac.uk/~matthew/papers/SeamfulInterweaving.pdf) argues that selectively exposing technological differences and limitations can support understanding and appropriation.
Its mixed-reality context does not establish that every implementation detail should become visible.
Portable application:
- CLI: exploit composition, streams, repeatable invocation, searchable output, and surrounding tools where useful.
- TUI: exploit keyboard-driven spatial work and incremental updates while accounting for terminal capabilities, remote transport, resize, and assistive access.
- Mobile: account for reach, touch precision, interruptions, battery, and variable connectivity without assuming touch is the only input.
- Desktop: exploit sustained work, precise input, simultaneous views, and inter-application workflows where they help.
- Web: preserve meaningful navigation, addressing, browser controls, and operation across variable devices and networks.
- These categories overlap; a remote terminal, mobile browser, and keyboard-equipped tablet invalidate simplistic platform stereotypes.
## Machine semantics and resource constraints
- [CLI Guidelines: Output](https://clig.dev/#output) distinguishes human formatting from stable machine-oriented output and discusses explicit structured formats.
This supports separate presentation and semantic contracts, not a universal requirement for JSON or an external API.
- [Kleppmann et al., Local-first software, 2019](https://www.inkandswitch.com/essay/local-first/) connects architecture to responsiveness, offline continuity, ownership, and collaboration.
Local operation can remove network dependencies from immediate work while introducing synchronization, conflict, storage, and lifecycle costs.
Its historical implementation assessments are not current product recommendations.
- Internal data should preserve meaning independently of presentation; formatting, localization, rounding, and summarization must not become accidental storage semantics.
- Mechanical sympathy requires reasoning about the actual substrate and workload, not assuming fewer allocations, more caching, or more parallelism always improves interaction.
- Latency, throughput, memory, bandwidth, energy, and correctness can conflict.
Batching may increase throughput while delaying feedback; speculation may reduce waiting while wasting transfers or showing stale state.
- Responsiveness-first is the user's preference, not permission for unlimited speculation.
Infer relevant limits from domain and project evidence; ask when the trade-off remains consequential and unresolved.
## Time and unavoidable delay
- [Nielsen, Response Times: The 3 Important Limits](https://www.nngroup.com/articles/response-times-3-important-limits/) offers approximate 0.1 s, 1 s, and 10 s guidance for perceived immediacy, flow, and sustained attention.
These are historical rules of thumb, not universal acceptance thresholds.
- [Ng et al., Designing for Low-Latency Direct-Touch Input, UIST 2012](https://www.tactuallabs.com/papers/designingLowLatencyDirectTouchInputUIST12.pdf) found perceptible benefits well below 10 ms in studied direct-touch interactions.
The study also explored immediate low-fidelity feedback followed by more complete rendering.
This does not prescribe that every application needs specialized hardware or a 1 ms budget.
- [Interaction to Next Paint](https://web.dev/articles/inp) is a current web responsiveness metric covering input delay, event processing, and presentation through the next paint.
Its good-responsiveness threshold of 200 ms is a web measurement convention, not a universal human limit or proof of eventual task completion.
- [RAIL](https://web.dev/articles/rail) remains useful for distinguishing response, animation, idle work, and loading.
Its own documentation points to newer Core Web Vitals guidance; do not describe it as the latest web acceptance model.
- Evaluate acknowledgment, useful results, and durable completion separately.
An immediate pending state is honest feedback; an unsupported success claim is not a latency optimization.
- Remove unnecessary work and critical-path dependencies before adding waiting treatments.
When waiting remains, preserve other useful work, orientation, truthful progress, and meaningful control.
- Progress estimates need evidence; unknown total work does not justify invented percentages.
Expose actionable boundaries rather than every internal detail.
## Accessibility and agency
- [WCAG 2.2](https://www.w3.org/TR/WCAG22/) provides testable accessibility criteria for web content.
Compliance alone does not establish overall usability or address every cognitive need.
- [WCAG2ICT](https://w3c.github.io/wcag2ict/) explains applying WCAG concepts to non-web software, including text, command-line, and terminal applications.
It is informative guidance, not an independent conformance standard; the linked editor's version can change.
Text-only output and keyboard operation do not automatically make an interface accessible.
- [W3C cognitive accessibility supplemental guidance](https://www.w3.org/WAI/WCAG2/supplemental/) adds objectives concerning understanding, finding information, avoiding mistakes, focus, memory, and support.
Supplemental guidance is not identical to normative WCAG success criteria.
W3C TR and WAI fetches were blocked during research; indexed excerpts were available, and WCAG2ICT was inspected through its editor's publication.
- [FTC, Bringing Dark Patterns to Light, 2022](https://www.ftc.gov/system/files/ftc_gov/pdf/P214800%20Dark%20Patterns%20Report%209.14.2022%20-%20FINAL.pdf) documents how interface design can impair autonomy, decision-making, and choice.
Improved business metrics do not by themselves demonstrate improved user outcomes.
Report project preferences and affected-user consequences distinctly from research conclusions.
## Adaptive and automated interaction
- [Amershi et al., Guidelines for Human-AI Interaction, CHI 2019](https://www.microsoft.com/en-us/research/publication/guidelines-for-human-ai-interaction/) synthesizes 18 guidelines, evaluated with 49 practitioners across 20 AI-infused products.
Relevant concerns include capability boundaries, correction, adaptation, feedback, and user control.
This is not proof that every interface should use conversation or automation.
- [What does Generative UI mean for HCI Practice?, CHI 2026 workshop](https://www.microsoft.com/en-us/research/publication/what-does-generative-ui-mean-for-hci-practice/) explores opportunities and challenges of generated interfaces.
A workshop agenda establishes an active research question, not validated superiority over stable interaction.
- Balance adaptation against predictable behavior, acquired skill, accessible operation, and ability to inspect or override consequential automation.