Web-based temporal applications face architectural challenges separating computational time accuracy from visual rendering latency, particularly under resource contention and browser background throttling. The central premise of this preregistration-ready protocol is that a paused rendering loop does not necessarily imply temporal drift when the application state is reconstructed from an independently anchored monotonic clock. This cross-platform controlled experimental protocol presents a hierarchical measurement framework to isolate and quantify wall-clock synchronization error, monotonic scheduling delay, rendering execution, and physical optical presentation latency.
The conceptual treatment space represents a six-factor factorial design (1,080 combinations). To ensure feasibility, the executed experiment utilizes a preregistered D-optimal fractional-factorial allocation constrained to permit unbiased estimation of all pre-specified primary main effects and causal interactions. The protocol introduces a mathematically consistent additive endpoint decomposition, culminating in hardware-calibrated photodiode measurement of browser-event-to-optical presentation latency. Direct UTC traceability is established separately for the synchronization layer, avoiding the assumption that optical detection intrinsically equates to absolute UTC accuracy.
Web-based temporal applications operate within the restrictive confines of the browser sandbox, relying on ECMAScript engines to process high-resolution time, schedule callbacks, and execute rendering pipelines. A fundamental architectural challenge arises under resource contention and lifecycle transitions (e.g., backgrounding a tab). While pipelines like requestAnimationFrame (rAF) and OffscreenCanvas optimize active frame pacing, they are fundamentally subject to browser scheduling policies that aggressively throttle background execution to preserve battery life.
However, the web platform explicitly distinguishes monotonic high-resolution timing (performance.now()) from system-clock epoch time (Date.now()). Consequently, a presentation-continuity failure (e.g., dropped frames in a background tab) is structurally distinct from a temporal-state failure. If a web clock relies on an incremental ticker (tn+1=tn+Δ), background throttling directly induces temporal drift. Conversely, reconstructing the logical time from an anchored monotonic clock may preserve temporal correctness even when rendering is entirely suspended.
This protocol establishes a formal causal measurement framework to separate wall-clock correctness, browser scheduling delay, and physical presentation latency. By applying rigorous physical metrology and a fractional-factorial experimental design, this study will quantify the degree to which browser engines decouple temporal correctness from visual continuity.
The problem of distributed time synchronization is foundational to networked systems, classically addressed by the Network Time Protocol (NTP) and the Precision Time Protocol (PTP) [1, 2]. However, browsers abstract away direct socket and hardware clock access. While W3C specifications establish performance.now() as a monotonically increasing clock resistant to system-clock skew [3], client-side JavaScript execution remains vulnerable to event-loop saturation and OS-level thread starvation [4, 5].
Display metrology has demonstrated that rendering completion in software does not equal optical presentation. End-to-end latency includes pipeline dispatch, compositor buffering, display scanout, and pixel response time [6, 7]. Variable Refresh Rate (VRR) displays further complicate this by decoupling the physical refresh from rigid 16.67 ms intervals [8]. Previous studies benchmarking web performance frequently conflate computational callback latency with physical optical latency [9, 10]. Furthermore, experimental designs evaluating web performance rarely utilize robust fractional-factorial methods, often relying on one-at-a-time factor manipulation that fails to capture critical structural interactions [11, 12].
This protocol tests the causal mechanisms separating computational time state from rendering continuity via the following primary hypotheses:
To ensure mathematically valid endpoint decomposition, we distinguish wall-clock time, monotonic performance time, and physical execution events on a common calibrated time base:
The end-to-end latency (Le2e) is modeled as a strictly additive chain:
Where the components are:
Concurrently, the underlying logical state error is defined as:
Temporal-State Error: Estate(t)=T̂app(t)-TUTC(t)
The conceptual treatment space represents a six-factor factorial design (1,080 combinations). Causal interpretation is restricted explicitly to experimentally manipulated factors and their randomized within-device contrasts. Operating system, CPU architecture, GPU, browser engine, and device performance tier are treated as device-level stratification blocks rather than randomized causal treatments.
The executed allocation will utilize a preregistered D-optimal fractional-factorial design constrained to permit unbiased estimation of all primary main effects and the H1--H3 interactions. Treatment order within each device will be randomized and Latin-square balanced across sessions.
Run durations are tailored to specific endpoints: short intervals (minutes) for scheduling/rendering analysis, and long intervals (hours) for estimating clock drift (ppm).
The combined standard uncertainty (uc) uses a general covariance formulation to account for correlated components (e.g., GPU scheduling and rendering latency):
Sources include uref (reference source), unet (transfer asymmetry bounds), uclk (device offset instability), ures (API resolution granularity), udisp (Variable Refresh Rate / scanout phase), and usens (photodiode calibration).
To strengthen metrology without assuming normality, final measurement uncertainty intervals ([P2.5,P97.5]) are derived via Monte Carlo uncertainty propagation (105 iterations) applied to the empirically calibrated component distributions [15].
The physical device is the highest-level independent experimental unit. Data structure follows: Device→Session→Run→Frame.
To handle right-skewed frame intervals and structural dependencies, run-level P99 will be calculated within each run and analyzed via a linear mixed-effects model (LMM) [17]:
(Note: The structural interaction B×D explicitly accounts for the known dependency of rAF on background states).
Statistical power is formally simulated [18]. The protocol generates 10,000 Monte Carlo datasets simulating the full hierarchical covariance structure, explicitly modeling predefined variance components (σdevice2,σsession2,σrun2,σframe2). Power is calculated as the proportion of simulations successfully rejecting H0. Final physical device counts (Ndevices), sessions, and runs will be pinned once pilot Intraclass Correlation Coefficient (ICC) distributions guarantee 80% power at α=0.05 for the TOST equivalence margins.
An independently implemented reference evaluator, generated directly from the version-pinned IANA tzdb source distribution, evaluates the browser's Intl.DateTimeFormat ECMA-402 implementation [19]. The suite tests:
This protocol adheres strictly to Registered Report and SPIRIT principles to prevent undisclosed flexibility [20].
OS build, browser binary checksum, ICU version, exact IANA tzdb release (e.g., 2026c), CPU architecture, GPU driver, and display mode (including VRR state) are frozen per device prior to data collection.
Invalid sessions caused by predefined instrumentation failures (e.g., wall-clock step/adjustment exceeding tolerance, OS sleep) will be reported in a CONSORT-style run-flow diagram. No outcome-based exclusions are permitted. Upon study completion, the preregistered Git commit, Docker container digests, exact dependency lockfiles, raw telemetry, and Python/R analysis scripts will be openly published.
[1] Mills, D. L., et al. (2010). Network Time Protocol Version 4: Protocol and Algorithms Specification. IETF RFC 5905.
[2] IEEE. (2019). IEEE Standard for a Precision Clock Synchronization Protocol for Networked Measurement and Control Systems. IEEE Std 1588-2019.
[3] W3C Web Performance Working Group. (Current Working Draft). High Resolution Time Level 3. W3C.
[4] Mozilla Developer Network (MDN). OffscreenCanvas and Background Execution Policies for requestAnimationFrame.
[5] Grigorik, I. (2013). High Performance Browser Networking. O'Reilly Media.
[6] Ito, K., et al. (2019). Measurement and evaluation of end-to-end latency in interactive displays. Journal of the Society for Information Display, 27(1), 15-25.
[7] Stauffert, J. P., et al. (2020). Latency and Cybersickness: Impact, Causes, and Measures. A Review. Frontiers in Virtual Reality, 1, 31.
[8] Vesa (Video Electronics Standards Association). Adaptive-Sync Display Standard.
[9] Sundararajan, K., et al. (2020). Analyzing Client-Side Temporal Drift in Distributed Web Applications. Journal of Web Engineering, 19(3), 311-334.
[10] Netravali, R., et al. (2018). Polaris: Faster Page Loads Using Fine-grained Dependency Tracking. USENIX Symposium on Networked Systems Design and Implementation (NSDI).
[11] Montgomery, D. C. (2017). Design and Analysis of Experiments (9th ed.). Wiley.
[12] Goos, P., & Jones, B. (2011). Optimal Design of Experiments: A Case Study Approach. Wiley.
[13] Internet Assigned Numbers Authority (IANA). Time Zone Database (tzdb).
[14] W3C Web Performance Working Group. Long Tasks API. W3C.
[15] Joint Committee for Guides in Metrology (JCGM). (2008). Evaluation of measurement data — Supplement 1 to the "Guide to the expression of uncertainty in measurement" — Propagation of distributions using a Monte Carlo method. JCGM 101:2008.
[16] Lakens, D., et al. (2018). Equivalence Testing for Psychological Research: A Tutorial. Advances in Methods and Practices in Psychological Science, 1(2), 259-269.
[17] Galwey, N. W. (2006). Introduction to Mixed Modelling: Beyond Regression and Analysis of Variance. Wiley.
[18] Arnold, B. F., et al. (2011). Simulation Methods to Estimate Design Power: An Overview for Applied Research. BMC Medical Research Methodology, 11(1), 94.
[19] Ecma International. ECMAScript Internationalization API Specification (ECMA-402). (Edition exact-pinned at preregistration).
[20] Nosek, B. A., & Lakens, D. (2014). Registered Reports: A Method to Increase the Credibility of Published Results. Social Psychology, 45(3), 137–141.