Reading the forms

These poems place recorded behavior beside my retained references to state or care. They do not assign clinical episodes, detect mood or estimate treatment effects. The curves between measured endpoints are visual constructions, not inferred trajectories.

Stride

Blue and yellow paired filaments encode daily sample medians of walking speed and stride length. Only records with exactly matching source, start and end timestamps enter the paired input. The retained export supplies 29,297 pairs across 1,196 source/date groups; one unpaired speed record is excluded. Each source has a separate passage; observed dates advance left to right. Blue upward depth and yellow downward depth use separate linear zero-to-archive-maximum scales. Curve width and seam undulation compose the form. Anonymous source labels, pair counts and exact medians remain in the reader. Device changes are not interpreted as behavioral changes.

Sleep Texture

538 observed noon-to-noon windows become nested, broken elliptical contours. Radius follows observed-date order, not elapsed calendar time. Clockwise angular position starts at noon and advances through the actual 23, 24 or 25 elapsed hours. Five-minute cells are drawn when their midpoints fall within a retained asleep interval. Blue and yellow alternate adjacent contours; red highlights the selected date. Quantization can lose short observations or merge short breaks. The reader distinguishes original recorded duration from rounded silhouette duration. Open space does not establish wakefulness; visible runs are not awakenings. Raw interval endpoints stay private.

Undertow

538 unsmoothed contours preserve hourly asleep durations in noon-to-noon windows. Centers advance with actual calendar dates, with a small aesthetic diagonal drift. Radius at each elapsed-hour vertex equals a fixed baseline times (1 + 0.75 × that bin’s recorded asleep hours). Yellow solid lines use S1; blue dashed lines use S2. Red dots expose selected vertices. Curves do not reconstruct raw sleep intervals; missing dates have no contour and empty bins do not establish waking. Daylight-saving windows retain their actual number of elapsed bins.

Rift

57 care-window dates become paired feathers: yellow sleep triangles on the right, blue step squares on the left, and smaller red sent/received filaments. Chronological angular placement composes the wings. Extension beyond a fixed inner margin uses separate raw sleep and step scales and a shared Messages scale. Missing sleep has a gray dashed stub and slash, while zero remains at the inner margin. The white seam mark identifies the retrospective care date. The shape does not label clinical phases or treatment effects.

Lacuna

63 fixed-shape open lenses enclose runs of 1–14 missing sleep-record dates between observed windows, totaling 162 missing dates. Seven longer recording breaks are omitted from this view; area is rescaled across the remaining runs and stays proportional to missing-date count. Yellow triangle and blue square boundaries denote the preceding and following observations. Positions are packed for composition, not placed on a time axis. Gaps do not estimate sleep duration or identify episodes.

Noon Rings

One chronological spiral gathers all calendar dates in the recorded asleep span, from oldest outside to latest inside. A central enlarged ring displays the selected date and initially shows the latest recorded date. The earlier five-year spiral layout remains available in settings. Small yellow arcs encode recorded durations within noon-to-noon elapsed-hour bins; their total angular fraction equals recorded hours divided by the actual 23, 24 or 25 elapsed hours. Arc placement is illustrative. Source S2 has a blue square, missing dates have gray slashes, and empty sectors do not establish waking. The exact reader supplies local clock labels.

Held Date

One column per date around the retrospectively stated January 6, 2021 care reference: 28 days before, the reference date, and 28 days after. Yellow sleep triangles rise above the baseline on a raw 0–25-hour scale; blue step squares fall below on a separate raw scale covering the case. The smaller red footer shows sent above and received below, using one count scale. Sleep uses previous-noon-to-noon windows; steps and Messages use calendar dates. The case has 46 recorded sleep windows and 57 step dates. Messages counts are defined for all 57 dates; ten have no retained eligible rows and are shown as zero. This selected retrospective case is not evidence of treatment effects or verified onset/recovery.

Rest / Motion

538 paired dates form a woven field of loops. Yellow bends upward with recorded asleep hours and blue downward with steps. Additional bend depth beyond a fixed small margin is linear on separate full-data scales. Seam positions follow calendar dates, with a compositional undulation and shared loop width. Curves are illustrative, and missing dates have no loop. The earlier sleep-versus-steps constellation remains available in settings. Source filters and reader ordering retain full-data scales and positions. Optional red paths join only consecutive eligible dates and break across gaps. Sleep windows and step dates partly coincide rather than sharing identical boundaries.

Body sources

The retained Apple Health export supplies 2,546 observed step dates from October 31, 2016 through October 26, 2023, and 538 recorded asleep windows from February 12, 2018 through August 15, 2022. Sleep has no overlap with Bluesky; steps overlap Bluesky on 176 observed dates. These are recording bounds, not complete physiological coverage. Sleep retains asleep categories, removes exact duplicate signatures, unions intervals within each source, and selects one source per window by overall observed-date coverage. Steps likewise remove exact duplicates and select a source by date coverage rather than adding sources together. Anonymous source labels remain inspectable. The current Messages snapshot contains 314,261 eligible rows, excludes reactions and other associated/system rows, and never supplies message contents.

Want

One burst marks one retained Amazon shopping date/hour. Horizontal position follows the calendar; vertical position follows the local hour. Each spoke is a search row, marked by color and endpoint shape for its furthest recorded outcome: search only, clicked without adding, added without purchasing, or purchase-linked. Burst radius follows square-root count, making the enclosing disk area proportional to the count. Angular spacing is aesthetic and does not reconstruct event sequence. Filters preserve the full calendar, hour and radius scales.

The retained December 2025 search export has 680 rows: 437 click-linked, 180 cart-linked and 125 purchase-linked. Converting UTC timestamps to America/Los_Angeles yields 262 hourly buckets over 155 dates, June 29, 2024 to December 15, 2025. These are search-linked flags, not unique purchases, order totals or spending. Missing hours mean no retained search record, not confirmed inactivity. No product, search term, price, identifier or IP address is published.

A separate rail marks point-aligned care, high/mixed and low references on the same calendar. Report-day proxies and explicitly dated references can appear; uncertain ranges remain excluded. The rail supplies context without assigning mood to shopping. The exact reader identifies retained references on a selected date; association does not establish cause.

Counterpoint

Each thread connects before and after medians around one retained reference group with a single alignment date. Categories describe references to care, high/mixed state or low state; they do not label the surrounding days. Exact stated dates align to those dates, current references to their report dates. Uncertain recent/month/year intervals are excluded from this point-aligned form. Repeated category/date-bound references are consolidated; groups are not episode counts.

The left semicircle is before, the right semicircle after; values increase from bottom to top. Both arcs show the same labeled observed-range scale; zero may lie outside that range. Each side uses 7, 14 or 21 days, excludes the anchor and needs five eligible days for its median. Counts and words use log(1 + value) within the observed minimum-to-maximum range; shares use a linear observed range. Full and held-out eligible comparisons across categories define that range. Filters and held-out views retain it, and exact values are unchanged. Red triangles, blue circles and yellow diamonds distinguish high/mixed, low and care references without relying on color alone. Night means 22:00–05:59 in America/Los_Angeles and needs five records per day; words and reply share need one. Holding out health-coded posts recalculates denominators, probing reuse of label-defining records without creating independent clinical labels. Messages sent and received come from a separate snapshot, excluding reactions and system/associated rows. The hold-out control does not apply to Messages.

Return Paths

A strand marks a gap of at least 12 hours between consecutive retained authored records, with paired 24-hour windows inside archive bounds. The lower endpoint carries the pre-gap measure; the upper endpoint carries the post-gap measure. Each window includes its boundary record. Calendar mode places the lower endpoint at the departure timestamp and the upper endpoint at the return timestamp on one linear calendar scale. Horizontal span is elapsed time; height is activity, not duration. Sub-day gaps occupy less than a pixel at the multi-year scale. Expanded mode preserves the earlier deliberately enlarged logarithmic widths and endpoint offsets, removes the calendar axis, and labels that geometry explicitly. Raw vertical height is linear by default, so large differences retain their size. Compressed height restores the original log(1 + value) scale for counts and words; shares always use 0–100%. No moving average or neighboring-strand smoothing is used. Each endpoint still summarizes its own paired 24-hour window. Thresholds of 12, 24 and 48 hours expose sensitivity; filters and held-out views retain fixed scales. Missing endpoints remain open.

Primary colors and endpoint symbols reflect available now/recent return-day references, prioritizing care, high/mixed and low; the exact reader retains all categories. Annotation availability describes coding coverage among existing candidates, not validated recall. Unlabeled dates remain unknown. Gaps are not measured sleep or episode transitions. Windows overlap and reference selection is not independent; no causal or significance claim follows from these paired observations.

Coverage and publication

The retained authored archive contains 69,043 records from May 4, 2023 through October 8, 2026. Newer retained snapshots replace older values for the same record; previously retained deleted records remain observations. Archive bounds describe observed coverage, not proof of exhaustive capture. Times use America/Los_Angeles.

This collection publishes only the measurements, date bounds, anonymous source/group labels and retained reference categories needed by these artworks. Source text, evidence excerpts, coding ledgers, record identifiers, contacts and coordinates stay private. The information button in each poem exposes exact values and source qualifications. Reference categories are unconfirmed self-reports, not clinical ground truth.