Beyond the BeatSIYAA’S PULSELABSIYAA’S SCIENCE & ENGINEERING JOURNAL
Decoding Individual Heart-Rate Response and Recovery Using Wearable Technology
A scientific question about individual heart-rate response. An Apple Watch and iPhone platform built to investigate it.
01 THE SPARK
“What if I could make my heart’s response to exercise visible?”Siyaa · project motivation
Basketball, rugby and karate made me notice that my heart seems to respond differently to different activities. News stories about deaths during or after exercise also sparked my curiosity.
What if I could make my heart’s response to exercise visible?A personal observation became a question: do different people respond differently to the same exercise? I began with a question, not an answer.
What can age and exercise type help explain?To investigate, I needed a consistent way to capture heart-rate samples, guide an exercise session, calculate changes and make recovery patterns visible.
Make the invisible measurable.That need became Siyaa’s PulseLab: an Apple Watch and iPhone research platform, with timestamped measurements, calculations and graphs.
One study. A connected platform.News stories sparked curiosity, not a medical hypothesis about deaths. PulseLab does not predict sudden death, diagnose disease or determine whether exercise is safe.
Personal account supplied by Siyaa in the website brief; research identity and progression documented in PROJECT.md.
02 THE RESEARCH
SCIENTIFIC QUESTION
TECHNOLOGY QUESTION
Implemented; physical transfer validation remains.
Versioned derivations and automated tests are implemented.
Disabled. Thresholds and rules are not approved.
Not established: no reference device and one session per person.
Exact age and exercise type; exercise position is a control term. Age is observed, not experimentally assigned.
HR samples and derived response/recovery: BPM increase, percentage increase, one/two-minute recovery and confirmed threshold time.
Five-minute seated baseline, 30-second duration, same study Watch and fit check, indoor testing, consistent instructions and counterbalanced orders.
Comfortable self-selected pace, caffeine and differences between similar staircases remain limitations. Sex/activity/anthropometrics are exploratory.
No finalized hypothesis for the current Beyond the Beat question is recorded in the reviewed authoritative documents. Siyaa should write and confirm it before presenting it; the earlier project’s prediction is not silently reused.
03 EXPERIMENTAL DESIGN
Jog in place with alternating knees and upright posture. Raise each knee toward hip height only as comfortably possible. No forced height or unnecessary jumping.
30 secondsClimb a normal flight and walk back down, repeating for 30 seconds at a controlled pace. Enter the number of stairs in the flight beforehand; this is not a count of steps performed.
30 secondsSit on the floor with legs extended. Reach comfortably toward the toes and return upright. A slight knee bend is allowed; do not force the stretch or bounce.
30 secondsFor stairs, the documented protocol requires similar normal flights, clear dry surroundings, no handrail use and a nearby adult. Anyone unable to perform without support does not perform this exercise. Stop safely when the timer ends; do not rush on the stairs.
Stop for pain, dizziness, loss of balance or discomfort. No app HR value determines medical fitness or clearance.
Use the same physical Watch with a snug, comfortable fit. Testing is indoors; avoid vigorous exercise for 30 minutes beforehand. Comfortable pace is not controlled intensity, and caffeine is an uncontrolled limitation.
04 RECRUITMENT DESIGN
PLANNED PARTICIPANT DISTRIBUTION
Exact age remains the main variable; groups guide balance and simple charts.
RECRUITMENT GOALS, NOT ENROLMENT
The sex-balance goal is supplied in Siyaa’s current website brief. The app also permits “another sex variation” and “prefer not to answer”; this goal is not a new eligibility exclusion.
One session per participant. The 100-person goal is not a minimum for descriptive analysis or proof of adequate statistical power.
The documented recruitment targets are overridable. Actual age is never altered to meet a target.
| Exact age | Target count |
|---|---|
| 12 | 5 |
| 13 | 4 |
| 14 | 4 |
| 15 | 4 |
| 16 | 4 |
| 17 | 4 |
| 18 | 3 |
| 19 | 2 |
| 20 | 2 |
| 21 | 2 |
| 22 | 2 |
| 23 | 2 |
| 24 | 2 |
| 25 | 2 |
| 26 | 2 |
| 27 | 2 |
| 28 | 2 |
| 29 | 2 |
| 30 | 3 |
| 31 | 3 |
| 32 | 3 |
| 33 | 3 |
| 34 | 3 |
| 35 | 2 |
| 36 | 2 |
| 37 | 2 |
| 38 | 2 |
| 39 | 2 |
| 40 | 3 |
| 41 | 3 |
| 42 | 3 |
| 43 | 2 |
| 44 | 2 |
| 45 | 2 |
| 46 | 2 |
| 47 | 2 |
| 48 | 2 |
| 49 | 2 |
| 50 | 2 |
Anonymous sequential ID; exact age; sex; activity category; height and weight; assigned exercise order; automatically recorded session time; protocol deviations. No participant names.
Activity: low <1 hour/week; moderate about 1–4; high >4 hours/week of moderate/vigorous activity. Height is saved in cm and weight in kg regardless of entry units. Sex, activity and anthropometrics remain exploratory.
05 THE SESSION
Anonymous ID, exact age, unit-normalized fields, assigned order and participant confirmations.
Sit quietly. Derive resting HR from the median of actual samples in the final 60 seconds.
The visible countdown and the same music start and stop together. Music does not set pace.
HR at or below 110% of resting for 30 confirmed seconds. Repeat exercise/recovery for the assigned sequence.
Save the session and verify transfer completeness. Missing Watch samples leave research metrics provisional.
The 110% recovery threshold is an operational study rule, not a medical cutoff or clearance to exercise.
| Order | Exercise sequence | 12–17 | 18–29 | 30–39 | 40–50 |
|---|---|---|---|---|---|
| 1 | Knees → Stairs → Toes | 5 | 4 | 4 | 4 |
| 2 | Knees → Toes → Stairs | 4 | 5 | 4 | 4 |
| 3 | Stairs → Knees → Toes | 4 | 4 | 5 | 4 |
| 4 | Stairs → Toes → Knees | 4 | 4 | 4 | 5 |
| 5 | Toes → Knees → Stairs | 4 | 4 | 4 | 4 |
| 6 | Toes → Stairs → Knees | 4 | 4 | 4 | 4 |
Totals across groups: 17, 17, 17, 17, 16, 16. Counterbalancing reduces sequence confounding; it does not remove every order effect.
Participant stop/discomfort: terminate. Interrupted or shortened exercise, stopped collection or phone termination: mark incomplete and terminate; do not silently repeat or resume the same exercise.
Interrupted baseline: restart baseline. Interrupted recovery: continue recovery and restart its confirmation window.
A disconnected phone can continue with a deviation only while Watch collection remains active. Reliable background timing can continue with a deviation; unreliable exercise timing stops the session. Music/timer failure during exercise marks it incomplete. Device behaviour still needs physical validation.
06 THE TECHNOLOGY
Siyaa’s
PulseLab.
The research need shaped the technology.
Manual snapshots would miss the recovery curve. PulseLab links a guided protocol with timestamped acquisition, consistent calculations and graphs, while keeping the measurements distinct from interpretation.
07 DATA ARCHITECTURE
Watch sensor → local Watch buffer → WatchConnectivity → raw iPhone storage → metrics → analysis
HealthKit workout-session access and an anchored HR query collect delivered optical heart-rate samples. True sensor cadence remains system-controlled.
View supporting artifactTimestamped source samples are saved locally, deduplicated and numbered per session before ordered batch transfer.
View supporting artifactAcknowledgements and retries carry batches to iPhone. Source and phone-receipt timestamps remain separate. The final sequence manifest verifies completeness.
View supporting artifactSeparate mode directories, atomic protected writes, protocol events, deviations and immutable sample append rules. Research is never filled with synthetic samples.
View supporting artifactC1 calculations read original samples and events. Values preserve protocol version and input revision; late inputs produce new derivations.
View supporting artifactSwift Charts displays source-derived outcomes. Scientific analysis reads metrics; unapproved educational suggestions remain disabled and cannot write back to source data.
View supporting artifactSource timestamps and receipt timestamps are stored with millisecond precision. Timestamp precision does not mean a new sensor sample each second or millisecond. No fabricated values fill gaps.
08 CALCULATIONS
A typical value from the final minute, calculated from actual delivered samples, not a separate sensor reading.
Illustrative synthetic samples: 69, 70, 70, 71, 72 BPM → R = 70 BPM.
How many beats per minute HR rose from resting to the immediate endpoint.
Synthetic R = 70, I = 140 → increase = 70 BPM.
The rise relative to that participant’s resting reference.
Synthetic (140 − 70) / 70 × 100 = 100%.
The drop from immediate HR to the one- or two-minute landmark.
Synthetic I = 140, HR(60 s) = 105 → 35 BPM recovery.
The share of the initial rise that has fallen by the chosen landmark. Unavailable when I ≤ R; not capped at 100%.
Synthetic (140 − 105) / (140 − 70) × 100 = 50%.
First completed 30-second interval with delivered HR at or below 1.10 × R. A gap longer than 5 seconds resets confirmation.
Synthetic R = 70 → threshold 77 BPM. Eligible readings remain ≤77 from 80 to 110 s → T = 110 s. The next exercise still waits until at least 120 s.
R is resting reference; I is immediate post-exercise HR; HR(t) is the actual HR sample nearest the one- or two-minute target.
Landmark tolerance is ±1 second. The sample’s actual timestamp and offset remain attached. Missing baseline or endpoint samples mean an unavailable metric, not a fabricated substitute.
Threshold time includes the confirmation interval. It is distinct from total recovery phase duration, which also enforces the 120-second minimum and adult-review decision.
09 ANALYSIS DESIGN
Paired observations, exercise means and descriptive 95% Student-t intervals.
Age scatter plots and trends for each exercise and selected outcome.
Order-adjusted curves, slopes and differences from a participant random-intercept REML model.
Observed versus fixed predictions, participant residuals, variance components and residual checks.
Sex, activity level and anthropometrics stay secondary to the primary question.
Available metrics, visible sample gaps, deviations and source-to-receipt delay. These do not prove sensor accuracy.
outcome ~ centered exact age × exercise
+ categorical exercise position
+ (1 | participant)Native random-intercept REML, approximate 95% Wald intervals and no automatic significance, p-values or causal claims. More than eight usable participants and a full-rank design are numerical prerequisites, not proof of statistical power.
Assumptions, missingness, influential observations and final interpretation still need scientific review. Descriptive charts can update from the first participant.
10 PLATFORM VALIDATION
Synthetic test data is not research data.
Real Watch source. Participant controls and supported protocol required. Separate local store; unverified transfer remains provisional.
100 generated participants and accelerated development sessions. Never counted as research. Deterministic and separately stored.
A separate data domain exists, but the visitor experience is parked. It is not an active study or testing mode.
Graphs, selectable outcomes, age groups, exercise and exploratory sex/activity comparisons, recruitment counts and saved-data reopening use a complete 100-person fixture. The fixture uses production calculations on generated source samples and events.
Eleven controlled scenarios cover response/recovery extremes, missing samples away from landmarks, delayed receipt, disconnect/reconnect and interrupted recovery. Additional failure fixtures preserve unavailable endpoints and incomplete sessions.
The latest app log records a clean combined simulator build, 41 core tests and the 100-person chart UI flow. The 60-person numerical fixture agrees with an independently fitted statsmodels model within test tolerances. These are platform checks, not human-study results.
Still to validate on physical devices: HealthKit permission and sensor cadence, paired transfers, battery cost, phone lock/background behaviour and backup authentication/recovery.
11 RESPONSIBILITY
Protects: Participant privacy
No names are needed to study HR differences. Sequential IDs replace identity in research records.
View supporting artifactProtects: Participant choice
Adults consent for themselves; minors require guardian permission and their own assent. Confirmations belong to each session.
View supporting artifactProtects: Participant welfare
Requests to stop or discomfort terminate testing without a medical decision by the app. Received data is preserved.
View supporting artifactProtects: Scientific reproducibility
A record must retain the rules used. Admin changes cannot silently rewrite existing session provenance.
View supporting artifactProtects: Research integrity
Synthetic testing must never increase research participant counts. Demo is reserved and unavailable.
View supporting artifactProtects: Data integrity
Later calculations read the source rather than editing it. Duplicate packets are deduplicated, not treated as new measurements.
View supporting artifactProtects: Honest measurement
No interpolation is active. Missing endpoints remain unavailable and gaps remain visible.
View supporting artifactProtects: Interpretability
Stopping early, interrupted recovery and transfer problems must remain visible rather than looking like normal completion.
View supporting artifactProtects: Complete research inputs
Protocol completion alone cannot prove every Watch packet arrived. Unverified research metrics remain provisional.
View supporting artifactProtects: Local confidentiality & durability
iOS file protection and atomic writes reduce exposure and partial-save errors. Failed saves roll back memory before acknowledgement.
View supporting artifactProtects: Export confidentiality
AES-GCM protects the app backup. Device authentication is required. Replacement-device key recovery remains unfinished.
View supporting artifactProtects: Research-data privacy
The documented handling rule keeps research records out of code repositories, prompts and external AI tools. This site includes only selected simulator images and summaries.
View supporting artifactProtects: Participant control over continued use
Continuing use must remain consented and needed for the approved project. Annual review is documented; final duration and deletion audit remain TBD.
View supporting artifactProtects: Scientific scope
Recovery thresholds organize the experiment; they are not medical cutoffs or exercise clearance. Educational suggestion rules remain disabled.
View supporting artifactGuiding-teacher approval was reported on October 3, 2026. The workspace still requires the specific supervisor identity, role and date in the Youth Science Canada information/consent document. No additional regional approval is invented here.
The plan defers separate exercise permission; the current app requires it. That discrepancy is visible for Siyaa’s review before live research, rather than silently changing either document or code.
12 MEASURED WORK
Files ending .swift under App/ only; generated outputs, tests, website and historical AI tools excluded.
Counting sourcePhysical lines in those App Swift files, including blank lines and comments. Not a measure of hours or original authorship.
Counting sourceXCTest method definitions in PulseLabCoreTests.swift; definitions are distinct from executed/passed cases.
Counting sourceXCTest method definitions in PulseLabUITests.swift; not number of repeated runs.
Counting sourceExplicit (age, sex) assignments in the Test Mode generator; not real participants.
Counting sourceCases declared in DerivedMetricKind, not separate sensors or all possible future formulas.
Counting sourceSelectable cases in AnalysisOutcome; not a claim about filter counts.
Counting sourcegit rev-list --count HEAD at the source snapshot. Uncommitted website work is not counted.
Counting sourceReviewed safeguard entries in this site, including labelled policies and parked limitations; not a count of completed controls.
Counting sourceSelected actual simulator PNG captures included in the gallery; not fabricated mockups.
Counting sourceUnique captured build operation IDs ending in success across project history, from the corrected development evidence summary. Includes tooling checks as well as app work. Repeated runs count separately; this is not a test-case count, and unrecorded activity is unknown.
Counting sourceUnique captured test operation IDs ending in success across project history, from the corrected development evidence summary. Includes tooling checks as well as app work. Repeated runs count separately; this is not a test-case count, and unrecorded activity is unknown.
Counting sourceBuild-time snapshot: 2026-10-05 UTC. No hours or uncaptured activities are estimated. Successful app checks are reported as dated evidence, not inferred from test definitions.
Download public-safe provenance13 DOCUMENTATION LIBRARY
Research, protocol, technology, calculations, tests and governance—organized as readable, source-linked artifacts.
13 artifacts
The question hierarchy, recruitment design and scientific limits.
Documented study decisionsFive-minute rest, counterbalanced exercise and adaptive recovery.
Documented and implemented; device validation remainsApproved fields, unit normalization and permissions.
Implemented; permissions are operator confirmationsHealthKit acquisition, durable buffering and sequence checks.
Implemented; paired-device validation remainsSource-derived formulas and explicit missing-value behaviour.
Implemented and covered by automated testsPaired comparisons, exact age and repeated-measures analysis.
Implemented with documented inference limitsDeterministic generation, complete metric coverage and failure fixtures.
Synthetic / Test Mode onlyThe latest documented core suite, UI flow and independent numerical reference.
Recorded October 4, 2026Human controls, teacher status and the specific remaining documentation.
Teacher approval reported; specific records and alignment remainData minimization, separate stores and authenticated export.
Implemented protection with explicit recovery limitsThreshold gaps, background callbacks and chart readability.
Documented corrections; some device checks remainA transparent record of implementation assistance and ownership.
Documented assistance policyA reproducible snapshot of app source, tests, versions and Git history.
Recomputed by website buildNo matching artifacts. Try another category or clear the search.
These are curated public-safe summaries of real workspace artifacts, not unredacted source-document exports. Each opens a document with source locations and hashes. Raw prompts, consent forms, developer logs and study databases are excluded.
14 DEVELOPMENT JOURNEY
Documented investigations, corrections and remaining checks. Technical takeaways are not invented personal reflections.
15 SCIENTIFIC INTEGRITY
Original delivered HR samples, actual source/receipt times and recorded protocol events.
Versioned transformations of source data. They can be recalculated without editing measurements.
Comparisons, model estimates and diagnostics. Interpretation requires real evidence and reviewed assumptions.
Separate, rule-based and non-medical. Suggestions remain disabled until rules are approved.
Synthetic generator patterns validate software. Observations require actual measurements; scientific conclusions require collected evidence and reviewed analysis. A graph alone cannot establish causation or a medical conclusion.
View supporting artifact16 ACTUAL APP CAPTURES
Every image is an existing simulator capture. Synthetic records and charts demonstrate the app, not findings from people.
Earlier Test Mode Home, before synthetic records were loaded. The app and website share the charcoal, coral and teal palette.
Actual Test Mode simulator setup: anonymous ID, exact age, activity categories and unit controls. Shown fields are synthetic, not participant data.
Accelerated Test Mode baseline with live synthetic HR and visible stop controls. This short countdown is not the five-minute research baseline.
Watch simulator waiting for the phone. No sensor reading is present; this image does not demonstrate live physical transmission.
Actual 100-person synthetic Insights: available-case means and paired observations. These are platform-validation patterns, not research findings.
Actual synthetic adjusted curves and slope intervals. Approximate uncertainty belongs to the fitted means, not individual medical predictions.
Source-derived values from a synthetic session, including confirmed threshold time. These numbers are not observations from a real participant.
17 PROJECT TIMELINE
Earlier work asked whether age and exercise changed HR. This project extends that question to recovery, individual variation and the engineering needed to observe it.
Private evidence capture and reproducible metric tooling were added to this repository.
View supporting artifactBeyond the Beat and PulseLab were distinguished, with a scientific question hierarchy and explicit ownership boundaries.
View supporting artifactTeacher approval was reported; the study plan documents continuous acquisition and adaptive recovery. Specific supervisor records remain required.
View supporting artifactThe first app build introduced shared models, a dark design system, a guided protocol and separate Test Mode.
View supporting artifactA reproducible 100-person Test dataset, complete recovery metrics, storage safeguards and callback regressions were documented.
View supporting artifactRepeated-measures analysis, adjusted comparisons and data-quality graphs were validated numerically and through simulator UI checks.
View supporting artifactDates come from current-repository commits and dated project records. The earlier heart-rate project is described in PROJECT.md, but its date is not documented here; no earlier workspace was accessed.
18 TECHNICAL NOTEBOOK
Validates mode, protocol and participant confirmations.
Anonymous fields, units and assigned exercise order.
State machine, instructions, countdown, music and interruptions.
HealthKit, Watch buffer and WatchConnectivity.
Raw samples, events, deviations and versioned records.
Read-only source derivation into revisioned metrics.
Reads available metrics; does not change source data.
Displays outcomes; unapproved suggestions stay disabled.
Readiness, modes, settings, authenticated backup and restore.
P1-draft is the development protocol identifier; P1 is the registered research identifier. C1 identifies current metric calculations; random-intercept-reml-v1 identifies the analysis fit. These are identifiers in code, not proof that a study has been activated.
Governance + participant setup → session protocol → acquisition → raw storage → C1 metrics → analysis → results. The Watch stores samples before transport. Each stored session retains a governance snapshot and protocol/calculation provenance.
Newly arrived raw samples advance the input revision. New derived values are appended for that revision; previous derivations remain. Analysis excludes incomplete/provisional data and does not pool incompatible versions. Educational output has no write path to raw data.
19 REAL RESEARCH
RESEARCH RESULTS · AWAITING EVIDENCE
No real participant findings are included in the project artifacts reviewed for this site. This section will be populated after research collection and scientific review. Synthetic charts appear only as Test Data / Platform Validation.
No enrolment total or research finding is inferred from Test Mode counts. Publication of future results requires an aggregate, privacy-reviewed dataset and scientific interpretation.
20 WHAT I LEARNED
What changed in your thinking? Which decision was hardest? What would you test differently next time?
Private local draft only. Text stays in this browser; it is not published or sent to a server. Do not enter participant details. Download your draft to keep a separate copy.