FAQ and support

You can still start with one photo. The optional summary of three to five records requires usable photo-equality evidence, compatible analysis methods, and your confirmation that you took the photos separately with the setup unchanged. Re-uploading or reanalyzing the same photo is not a retake. Legacy records and photo-deleted records remain readable, but cannot form this summary. For before/after, choose two photos and identify the single intended change. Your statements do not verify actual capture conditions or statistical independence. Matching bytes detect duplicates; different files do not prove separate photos. Median and range describe only the selected posture scores, not measurement accuracy, CdA, watt savings or a confidence interval. Changing the selection resets your review.

No. Only Japanese and English dashboard responses allow same-origin camera access. Public pages, the sign-in-free trial, and API responses deny camera access; microphone, geolocation, browser payment, and USB are denied everywhere. Iris does not start a live video stream and processes only the still image you deliberately select or capture.

Every Japanese and English dashboard and API response managed by Iris, including an anonymous dashboard redirect, is private, no-store, no-cache, noindex, nofollow, and noarchive. The Sites authentication dispatcher owns sign-in initiation, sign-out, and callback before the Iris Worker, so Iris does not claim to change those response headers. Robots rules and the sitemap exclude all three dispatcher paths.

No. Iris does not accept numbers or dates inside exercise names, record names, labels, notes, guidance, or other free text as evidence for an external-AI numeric claim. Only typed numeric and date fields, the correct evidence-category count, and validated structured aero target ranges are eligible for matching.

No. Iris normalizes full-width digits and requires every quantity to match a retained value plus an authorized unit, explicit unitless metric label, exact retained range, or correct evidence-count category. Non-rendering control formatting, unsupported numeral scripts, unitless physical quantities, prose calculations, or ranges assembled from unrelated retained values cause the whole provider-written answer to be discarded for the local rule-based answer.

Choose a .FIT file from Training. FIT format, CRC, and metrics are decoded in the browser, then Iris stores only extracted activity metrics. The original FIT file and filename are not uploaded or retained; the server does not recheck the metrics against the FIT binary. Only the duration-and-IF fallback load is recomputed by the server before storage; existing saved filenames use an anonymous label.

No. AI Coach activity context marks FIT metrics as browser-decoded and not rechecked against the binary by the server. Missing heart rate, power, cadence, speed, normalized-power, and intensity-factor zero sentinels are omitted. A training-load number is sent only with current provenance that confirms FIT session TSS or duration × IF and a positive value. Unavailable and legacy loads stay absent; Iris does not infer their value or source.

Not necessarily. Evidence counts and confidence use only records retained after the external-AI boundary checks for dates, ranges, provenance, and consistency. Strength total volume is recomputed from valid exercises and a missing duration is not zero minutes. An unset race target is omitted, while temperature and humidity remain athlete-entered scenarios. Aero values require matching stored and analysis method, view, confidence, and overall score; a retake-required result cannot support posture or fit. Excluded saved rows are not counted as answer evidence. Iris also matches numbers and dates in provider-written text to retained values and recognized units. A new value, unit mismatch, generated conversion, calculation, range, or numerical prescription causes the whole provider answer to be discarded in favor of the local rule-based answer.

Iris binds named average, maximum, normalized, and FTP power; heart-rate fields; body weight, exercise weight, and strength volume; the six recovery factors; and duration, distance, environment, aero, and record-coverage metrics to their exact structured source fields. The binding survives explanatory wording and an unambiguous metric name after its value in the same sentence. Aero score always means overall aero score, while a bare-number overall, quality, or other named aero score remains bound to that exact component. For individual aero metrics, only the 18 fixed product keys with their expected units are admitted; server-defined Japanese and English labels bind each measurement and named score to its own metric, and numeric deviations are withheld from external AI. Aero records are checked newest first, so a latest, previous, or dated value cannot be borrowed from another record. The qualifier covers every aero value in the same uninterrupted clause whether placed before or after it, preventing a trailing-date bypass or a correct first value from masking a later cross-record substitution; claims that mix multiple dates or both latest and previous are rejected. Ambiguous mappings of multiple same-unit labels and multiple values are rejected in either order, and another field cannot substitute merely because its value or unit matches. Iris discards a mismatched provider-written answer and returns the local rule-based answer.

The browser regenerates the selected photo as a JPEG and replaces its upload name with an application-generated name. Iris does not send, save, return, or export the original photo filename. Before storage, the server also rejects a JPEG that retains embedded personal metadata such as EXIF, XMP, IPTC, or a JPEG comment; existing aero-photo names are replaced with an anonymous name.

The browser center-crops the photo, regenerates it as a 768 by 768 pixel JPEG, and removes its embedded source metadata and original filename. Before storage, the server also rejects a JPEG that retains embedded personal metadata such as EXIF, XMP, IPTC, or a JPEG comment. It uploads the result under an application-generated name to storage private to your account. You can change or delete it from Account settings.

No. Today, yesterday, a weekday, last analysis, recent, and other unresolved relative times are rejected for aero numbers because retained dates do not provide a unique calendar interpretation. Iris must use an exact retained date, latest, or previous; otherwise it stays qualitative. Relative wording in a separate non-aero clause does not relabel an independently grounded aero value.

No. FIT activities are ordered newest first, and latest, previous, or an exact retained date binds duration, distance, power, heart rate, cadence, speed, elevation gain, intensity factor, and training load to that same activity. The qualifier covers every FIT metric in the same uninterrupted activity clause. Mixed dates, latest-plus-previous wording, and unresolved relative times such as today, yesterday, a weekday, or recent are rejected.

No. Each dated history is ordered newest first, and latest, previous, or an exact retained date binds every metric from that category in the same uninterrupted clause to one retained record. Recovery signals, sleep duration, HRV, resting heart rate, body weight, strength duration, RPE, volume, exercise weight, race distance, goal time, scenario temperature, and humidity cannot be borrowed from another date. Mixed dates, latest-plus-previous wording, and unresolved relative times such as today, yesterday, a weekday, last workout, last check-in, last weigh-in, last race, or recent are rejected.

No. Latest, previous, an exact date, or an unresolved relative time may accompany a quantity only when that same claim names the exact recognized metric. Activity, record, check-in, workout, event, analysis, output, value, score, and load are category words, not metric names. Iris rejects category-only or bare numeric wording instead of matching it to any retained field with the same value.

Iris sets a 500-Unicode-character limit for each text item in the strict schema sent to external AI, then rechecks the exact required keys, declared types, array limits, unique canonical evidence labels, nonblank text, and the same server-side limit. It never keeps the rest by silently deleting, repairing, or truncating a malformed part. Any mismatch discards the whole provider-written answer and returns the checked local rule-based answer.

The trial profile and up to 20 extracted FIT activities stay only in this browser's local storage. After 30 days from the latest save, Iris deletes them the next time you open the trial; it cannot delete local storage while the browser is not running. You can remove them immediately with the trial's delete button. Trial photos, conversations, original FIT files, and original FIT filenames are not stored, and trial data is not sent to Iris or your production account.

Your current question and the needed non-heart-rate training, race, strength, and aero context are sent for the answer. Exact activity and aero-analysis times are reduced to calendar dates; profile time zone, event names and course notes, strength-session titles, and duplicate confirmation-card copy are not sent. Saved recovery, HRV, resting heart rate, body measurements and weight, prior question trends, and record-map coverage recalculated from current records stay excluded unless you select the per-answer checkbox. It resets to off after sending. When valid dated weight history is available, Iris sends that history as the sole weight authority and omits the rounded undated profile weight; the profile value is only a fallback when no valid dated record remains. IDs, filenames, photo keys, pose landmarks, free-text health notes, and earlier question text are never included in external-AI input context. For abuse prevention, request metadata contains only a stable pseudonymous identifier created under a separate secret, distinct from Iris's internal owner ID; it contains no email address or Sites user ID.

No. With health context enabled, Iris sends only complete recovery rows whose six subjective values have a recognized version and valid 1–10 range. Partial or uninterpretable rows are excluded. Missing sleep duration, HRV, and resting heart rate are omitted rather than sent as zero measurements. An incomplete or inconsistent current summary exposes no factor values or conclusion text and must be treated as missing or uninterpretable.

Iris does not add that question, response, or response-context summary to chat history, so they disappear after reload. The current question and minimized context are still sent to external AI, and the request still counts against plan and abuse-prevention limits. Feedback is not stored for an answer-only response. A confirmation-card record is stored only if you press its save button. The choice resets to off after sending.

Use an explicit save verb, for example: ‘Log today’s body weight as 150 lb’ or ‘Log squat 60 kg for 10 reps x 3 sets, RPE 7.’ Iris supports kg or lb, converts stored values to kilograms, and shows a confirmation card. Nothing is saved as a body-weight, strength, or recovery record until you press Review and save record. The server confirmation is valid for 30 minutes and rejects changed values or use from another account; ask Iris to create a new card after it expires. Recovery candidates require sleep quality, energy, stress, motivation, muscle soreness, and overall feeling so missing scores are never invented. Repeating the same confirmation does not create a duplicate record. A body-weight or recovery card never overwrites another record already saved for the same date; review or edit that record in Conditioning.

No. A conservative Iris server-side safety classifier covers chest pain, fainting, severe breathing difficulty, self-harm or suicidal intent, overdose, sudden stroke warning signs, danger signs after head, neck, or body trauma, and suspected-concussion support before plan and request limits. It separates emergency and protected-support guidance with category-matched fixed references. It does not use external AI or saved athlete records, does not consume the daily Iris allowance, and does not save the question or response. If external AI flags another safety risk, Iris discards its normal advice and returns only server-defined safety guidance, kept general and without guessed references. This is a precautionary route, not a diagnosis or emergency service.

Iris treats confusion, disorientation, loss of coordination, or a seizure near explicit heat exposure as an urgent warning. It tells you to stop exercise, call local emergency services, move to a cool area, remove outer clothing, cool rapidly while help is coming, and avoid forcing drinks when mental status is altered. This conservative rule runs locally on the Iris server before external AI and does not diagnose the cause.

A direct head hit is not required. After a crash, fall, collision, or blow or jolt to the head, neck, or body, serious danger signs select fixed emergency guidance before external AI, saved records, or plan limits. Without those danger signs, post-trauma headache, nausea or vomiting not described as repeated, dizziness, balance difficulty, fogginess, visual, light or noise disturbance, memory gaps, neck pain, an explicit current concern about concussion, or a request for post-concussion return-to-sport steps selects separate fixed support: stop now, do not return to riding, training, or competition that day, seek prompt assessment by a qualified healthcare professional, and do not drive yourself. For the first 24–48 hours, use relative rather than strict rest: ordinary daily activity and a short walk may be appropriate only if symptoms worsen no more than mildly and briefly; reduce or stop if they worsen more. Begin the six-step return only with healthcare-professional approval and supervision, usually spending at least 24 hours at each step: regular activities; light aerobic activity such as 5–10 minutes of walking or stationary cycling; moderate activity; heavy non-contact activity such as high-intensity stationary cycling; medically cleared controlled practice; and competition. Do not substitute outdoor riding for stationary cycling in the early steps. Fall- or head-impact-risk activity needs explicit medical clearance after concussion-related symptoms, cognitive abnormalities, and examination findings resolve. New or worsening symptoms mean stop, contact the healthcare professional, and return to the previous step only as directed after recovery. Students complete return to learning before unrestricted sport. If a danger sign develops, call local emergency services. This conservative boundary follows the Japan Sport Council sports-accident handbook, CDC concussion danger signs, CDC six-step return to sports, and Amsterdam international consensus on concussion in sport. Iris does not diagnose concussion.

Iris tells you to call local emergency services for immediate danger, not stay alone, move away from anything you could use for harm, and ask a trusted person to stay. It links Japan’s MHLW crisis-support directory and identifies 988 as a United States-only call or text service. This route runs before external AI, plan limits, saved records, or chat storage.

Not by itself. Iris refuses to create a weight-loss or compensatory-exercise plan and routes you toward a current clinician or specialist service, with the Japan national center and NIMH guidance as fixed references. Fainting, chest pain, severe breathing difficulty, or immediate danger of self-harm still takes the corresponding emergency route.

Free includes three aero analyses per calendar month, five Iris chat round trips per day, today’s calendar records, today’s aero archive and Iris conversations, and FIT import. The FIT list shows records from the 24 hours after their server import, independent of your profile time zone. Friend features, full FIT history, past aero archives, past Iris conversations and question-trend review, and historical calendar review require Pro. Older saved data is not deleted and remains included in your complete account JSON export.

No. Pro removes the Free plan’s three-aero-per-month and five-Iris-round-trips-per-day caps, but technical anti-abuse, security, and service-capacity limits still apply. Its main paid value is full FIT, aero, and Iris history with its evidence trail, question trends, friends, and month/year calendar review.

Iris enables Pro only after a signed live Stripe event confirms that the plan identity, currency, monthly amount, quantity, renewal interval, and current approved sales-country evidence match the offer. If verification fails or is unavailable, Iris does not grant Pro by mistake; it keeps safe Free access and shows billing management. Contact support under Billing in Settings.

The Free / Pro section in Account shows the current period end from the Stripe subscription Iris verified, formatted in your profile time zone. An ongoing subscription labels it as the next scheduled renewal; a subscription set to cancel at period end labels it as the scheduled end of Pro access. Iris shows no guessed date when the period is unavailable or the product or current sales-country contract is not verified.

Account explains Stripe-confirmed subscription states with fixed English copy. Past-due, unpaid, incomplete, paused, or ended subscriptions pause Pro and keep billing management available so you can review an invoice or payment method. Iris never echoes an unknown internal state; it shows safe general guidance and does not activate Pro.

Free costs US$0. English Pro is planned at US$5.99 per month, including applicable taxes. The first payment is charged at sign-up and then renewed monthly until you cancel, with no minimum term. The United States, United Kingdom, and EU member states are the planned initial target markets, but English Pro has no approved sales launch date. Sign-up and live payment remain disabled until each target country’s tax, consumer, privacy, health-data, minor-use, and accessibility requirements receive external approval, the virtual-office and contact arrangements are confirmed, and the Stripe production configuration is verified. Iris will never start a paid subscription without your explicit action.

Both editions use only the Stripe Product and Price fixed for that release. Immediately before creating Checkout, Iris re-fetches the selected Price from Stripe and requires the exact Product, currency, Japanese ¥980 or English US$5.99 amount, monthly interval, and live/test mode. A signed event must satisfy the same fixed contract before Pro is enabled. Missing, mismatched, or temporarily unavailable provider evidence fails closed, and the fixed IDs are not shown on public pages or in general logs.

Yes. Records used lists only categories the answer actually cited, and Iris rejects an external-AI answer that cites an unavailable category. Scientific and safety method references are selected from a fixed server allow-list for topics present in the answer; external AI cannot add or replace their URLs. They describe methods and limits and do not validate an individual recommendation. These references and the separate external-AI sharing boundary remain with saved chat history and the portable JSON export.

Iris does not create a diagnosis, individually validated prediction, or recovery score. It shows today’s six self-reports separately and counts low values of 1–3 for sleep quality, energy, motivation, or overall feeling and high values of 8–10 for stress or soreness as Iris-specific, unvalidated caution bands. A value without a recognized scale version and valid source range stays unavailable instead of being clamped and is excluded from the caution summary and progression templates. Sleep duration, HRV, resting heart rate, activity load, and the workload ratio remain context only. No caution band does not guarantee readiness. Speak to a qualified professional if concerning symptoms continue.

No. It is the proportion of 15 explicit checklist items that have a record: activity in four recent seven-day windows, recovery check-ins on seven recent days, FTP and weekly hours, and two aero analyses for repeat comparison. Missing data never means poor condition or ability.

Open Training, then Training plan. Each upcoming session can be created, edited, marked completed or reopened, and deleted. The optional target-load value is your own planning label; Iris does not calculate or validate it and does not turn the plan into a readiness, adaptation, or injury-risk score. A stable request identifier prevents a network retry from duplicating the same new planned session.

Open Training, then Training calendar. Free shows today's saved records; Pro can move through the loaded month and year history and jump to a date. FIT activities, aero summaries, recovery, body weight, strength, races, record-map updates, and planned sessions appear together without being sent to external AI. The calendar counts records only; it does not score ability, readiness, adaptation, or injury risk. Older FIT, aero, conditioning, and plan pages load only when you request them.

Open Conditioning, then Strength training log. Manual records and records saved from an Iris confirmation card appear together. You can edit or delete each record. A stable request identifier prevents the same new-record request from creating a duplicate if a network retry repeats it.

A systematic review and meta-analysis published online in 2025 (2026 journal volume) included 17 studies and 262 cyclists, only 60 women. Heavy strength training may improve cycling efficiency, anaerobic power, and cycling performance, but the certainty of evidence was low and did not support robust recommendations for optimal implementation. Iris therefore does not promise a benefit and prioritizes recording exercise, load, reps, sets, RPE, and next-morning response for adjustment.

Iris does not prescribe a universal mL/hour or mg/hour rate. The dashboard sweat-rate trial estimates an observation from pre/post body mass, intake and urine entirely in the current browser; it sends and stores nothing. The 30–240 minute product range warns that short trials amplify hourly-scaling error and long trials accumulate non-sweat mass change. The simple field estimate also cannot correct respiratory or metabolic mass loss or sweat retained in clothing. Repeat it in comparable training, then practise a plan that avoids body-mass gain during the event. Individualise sodium for duration, sweat loss, salty-sweat history and the exact product used. Temperature and humidity alone are not WBGT, so follow the organiser’s current heat policy.

It combines alignment with product-configured comparison bands and capture quality for repeat-photo comparison. Before calibration, the product rule weights metric alignment 84%, capture quality 10%, and metric reliability 6%, then applies an evidence ceiling when quality is insufficient. These are not individually validated scientific weights. The bands are not validated elite norms or your personal optimum. The score is not ability, measured CdA, a wind-tunnel result, or a bike-fit prescription. The supporting measurement comparison included 11 cyclists; the torso-angle study included 19 trained men and measured frontal area rather than CdA. A comparison report separates the overall score from all three components, presents any incompatibility that withholds numbers first, and retains every other detected limitation on screen and in print. Missing components in an older record keep the result provisional; an explicitly stored component outside 0–100 or with a nonnumeric value withholds the comparison numbers. Static and moving joint angles can differ, so compare the same setup and verify changes dynamically or with a qualified fitter.

No. Your selected helmet and apparel categories, still-air calculation speed, and a CdA with its self-reported measurement method when provided are stored with the private aero analysis as comparison context. Iris does not infer CdA, equipment-specific ΔCdA, watt savings, or time savings from a photo or category, and the categories contribute zero points to the posture score. Only when you enter both a CdA for the current rider, bike, position, helmet, and apparel and its measurement method does Iris assume still air (so ground speed equals relative air speed) and standard air density ρ=1.225 kg/m³, then use your still-air calculation speed as v to approximate aerodynamic drag with F=0.5ρCdAv² and aerodynamic power with P=Fv. Iris does not validate the selected method's accuracy or repeatability. Wind, yaw, gradient, rolling resistance, and drivetrain losses are excluded. The displayed N and W are totals for your supplied current-setup CdA, not an item effect, saving, total cycling power, or predicted time.

Yes. In aero history, you can delete only the private photo while keeping its analysis, metrics, and history. This is available on Free and stops retrieval immediately. Records without photos can still contribute numbers to comparisons, but Iris shows a caution because camera distance and height, equipment, and clothing cannot be visually rechecked. A pending state appears if storage deletion needs a retry. You can delete all Iris data, including conversations, photos, and friend connections, from Settings. Full deletion immediately stops new saves and sharing, then confirms cancellation of any tracked Stripe subscription before final erasure. Once that deletion is confirmed, the current browser removes Iris-owned trial and session keys and iris-named app caches, notifies other same-origin Iris tabs, and other open Iris dashboards normally recheck the minimal deletion state once per minute while the browser is running. Those tabs then clear their Iris state and sign out; ChatGPT authentication cookies and unrelated browser data are not targeted.

Use the feedback form in Settings after signing in, or email akira2940tri@gmail.com if you cannot sign in or need the seller’s telephone number. For a useful report, include your device, browser, time of the issue, steps, expected result, actual result, and a screenshot with personal information removed.