Behavioral measurement framework
I designed the temporal coding framework used to represent visible behavior across extended decisions, including posture, gaze, self-touch, chess interaction, and visibility.
HOT FILE // 01
research index ↙ACM CHI 2027 · UNDER REVIEW
No Single Tell
What can visible behavior actually tell us about thinking?
I study how gaze, posture, self-touch, and other visible behaviors unfold during extended expert decision-making in elite chess. The project aligns long-form World Championship video with human behavioral annotations, move timing, board state, clocks, and engine-derived decision context.
Rather than searching for a universal behavioral “tell,” we ask when visible behavior becomes meaningful once it is situated in time, individual baseline, and decision context.
peer review currently has custody of the interesting plots
research brief // 01
At a Glance
- Question
- How does visible behavior change across extended, difficult decisions?
- Setting
- 14 World Chess Championship games · 1,270 deliberation episodes
- Measurement
- Temporally localized human coding of gaze, posture, self-touch, chess interaction, and visibility.
- Context
- Move timing · clocks · board state · engine-derived decision difficulty and quality
- My role
- Behavioral measurement design · annotation/reviewer interfaces · multimodal alignment · temporal analysis infrastructure
- Status
- ACM CHI 2027 · under review
from footage to measurement
How the Study Works
The workflow moves from raw broadcast footage to temporally aligned behavioral measurements that can be interpreted alongside the decision being made.
-
01
Observe
Long-form broadcast video
World Championship footage provides continuous, naturally occurring expert decision-making rather than isolated laboratory trials.
-
02
Code
Visible behavior over time
Human reviewers temporally code observable behavior including gaze, posture, self-touch, chess interaction, and visibility.
custom annotation + reviewer workflow -
03
Align
Synchronize behavior with the game
Behavioral events are aligned with move timing, clocks, board state, and the surrounding deliberation episode.
-
04
Contextualize
Describe the decision itself
Engine-derived measures characterize the surrounding choice structure, pressure, and decision quality.
-
05
Analyze
Look for temporal and individual patterns
We examine behavior relative to context and player baseline rather than treating any single action as a universal signal.
contribution ledger
What I Built
Human annotation + review tools
I built custom interfaces for temporally localized annotation and multi-reviewer validation, allowing coders to scrub tightly around decision episodes and record behavior consistently over time.
Multimodal alignment infrastructure
I developed the pipeline linking behavioral annotations with synchronized video, move timing, board state, clocks, and engine-derived decision context.
Temporal analysis workflow
The resulting representation supports within-player and across-game analyses of how behavior evolves alongside decision context rather than reducing behavior to isolated snapshots.
measurement validity
Why This Is an HCI Problem
Systems increasingly treat visible behavior as evidence of internal state: attention, uncertainty, stress, engagement, or cognitive load.
But behavior is not self-interpreting.
This project asks a more basic measurement question: what can an observable action actually support when it is separated from the person, the surrounding task, and the moment in which it occurred?
By treating visible behavior as situated, temporal, and person-dependent, the work focuses not only on prediction, but on the validity and interpretation of behavioral measurement itself.
public takeaway
What We Can Say for Now
The project does not support the idea of a single universal behavioral marker of decision difficulty.
What becomes interesting is the context around behavior: who is doing it, when it occurs, what the decision looks like, and how the behavior changes relative to that person’s own baseline.
The detailed analyses, effect sizes, and plots are staying off the public page until the review process is complete.
the receipts are temporarily with anonymous review
methods
Methods
temporal behavioral coding · human annotation · multi-reviewer validation · multimodal alignment · computer vision · chess-engine analysis · within-player temporal analysis · interactive visualization
research, compressed
Three Minutes, Apparently
HKUST 3MT · 2026 · First Runner-Up
This project also became my HKUST Three Minute Thesis talk, where I received First Runner-Up.
five years of TA-ing may have accidentally prepared me for one microphone and three minutes
paper // under review
No Single Tell: Studying Context-Dependent Visible Behavior During Extended Decision Making in Elite Chess
Eman Ansar · Arpit Narechania · Atanu Sinha
ACM CHI 2027 · Under Review