Weighing Strategy

The Goal: Fewer Useful Measurements

Frequent early weighing helps establish how a particular plant, pot, and medium behave. Once that evidence becomes useful, it should reduce routine work. The long-term aim is the fewest measurements that still support timely decisions and keep the learned pattern honest.

Four to six pots is an example of a lighter day, not a quota, floor, or cap. One, two, or zero can be appropriate. A larger group can also be due together. Do not add measurements to fill a list, postpone important readings to make the count look good, or water on staggered days merely to spread the weighing workload.

Reviewing every active container's existing data each morning does not require physically weighing every container. A shared planter counts as one weighed assembly even when several plants live in it.

The active roster has 34 tracked containers: P01–P32, P35 and P36. Shared containers receive one whole-pot weight; separate botanical profiles do not create additional weigh-ins.

What Exists Today

The daily report uses this adaptive policy alongside the validity, correction, watering, and measurement rules below.

What Makes a Weight Comparable

Use the same assembly and a stable scale surface. Keep the same pot, plant, medium, labels, and any normally included accessories together. Record the actual grams and observation time. Empty free-standing drainage water before a post-drain reference; do not weigh an accidentally included pool of runoff as if it were retained pot mass.

Pot setup means the complete weighed configuration, not pot diameter. Repotting or a substantive medium/configuration change can start a new setup even when the physical pot is reused. Moving a plant to another light or changing a dimmer does not by itself change the setup number, although it can change drying behavior.

Pruning, spilled medium, removed labels, or changed saucers can shift mass independently of drying. Log what actually happened and treat the comparison cautiously. Clean, Prune, Check, and Other notes do not automatically subtract that mass from the model or create a new setup. Do not fabricate a Repot just to make a graph line up; use the appropriate entry and a reviewed setup correction when needed.

The ledger's Date is when the observation happened. Recorded is when it was saved. Updated from Google is when a view was fetched. A newly refreshed screen can still be showing an old measured weight.

Recorded Dry Top-Dressing Adjustment

Dry top dressing added September 16, 2026 increased the measured assembly mass by 108 g for Money Tree and 150.5 g for Kiwi aeonium. Earlier History weights include these offsets so they remain comparable with the dressed pots.

Post-dressing readings are the actual scale values. Original readings remain in weight-cell notes; observation times, watering events and pot setup numbers are unchanged.

Enter future weights directly from the scale. Do not add or subtract these offsets again. Wet/dry references, charts, and other derived values recalculate from the adjusted history. Earlier losses and wet-to-dry capacities stay the same; this preserves comparable mass history without claiming that future drying conditions will be identical. Dated reports, photographs, and older repository observations remain evidence of what was recorded at that time.

This adjustment applies only to the measured dry additions above. Unknown mass changes still need a setup review; the logger does not normalize them automatically. The workbook runbook retains the detailed record.

Useful Points in a Cycle

The five-day wet-anchor recovery allowance does not mean that any reading five days later is as informative as a post-drain reading. Water never implies that a weight was taken. A new low in an open cycle also does not automatically become a completed Dry endpoint. See reference meanings.

How the Model Learns

The current implementation selects up to five recent, usable completed cycles for the same plant and setup, with endpoints within 180 days of the current watering. It does not borrow another plant's cycle because the species or pot looks similar.

A usable training cycle needs a Wet anchor, a later eligible endpoint before the following watering, at least four distinct dated readings spanning three days, a descending log-linear fit with R² of at least 0.60, and meaningful observed loss. Its bounding waterings must be compatible with full-cycle forecasting. Partial/Spot cycles are excluded from that full-cycle training; legacy blank application values retain their compatibility treatment. These are software evidence criteria, not biological guarantees.

Recent, better-fitting cycles receive more influence, including a 60-day recency half-life. A supported historical estimate can be used before the current cycle has four new readings. Two or three current readings gradually adjust the historical estimate; four across three days can support a blended basis; six usable readings allow the supported current curve to take over. Other validity and fit checks still apply.

Forecast windows widen with weak or variable evidence. They are planning ranges, not statistical confidence intervals or promises that nothing can happen earlier. The model withholds very distant forecasts beyond its implemented 90-day horizon from the last observation; a displayed window still does not justify waiting that long without useful intermediate evidence.

Keep Learning Without Repeating the Initial Work Forever

Not every later cycle needs intensive daily sampling. A reliable historical estimate is the benefit of the earlier work. Preserve occasional intermediate points and periodically obtain a well-sampled completed cycle so usable history can remain recent.

There is a tradeoff: a cycle with too few measurements may help the current care decision but fail the training criteria. Do not label it a newly learned cycle just because a lighter schedule was desired. If qualifying evidence ages out, the pot's setup changes, or current behavior contradicts history, increase sampling for that pot temporarily.

Choosing Today's Small Subset

1. Read the current evidence for every active container. Include new waterings, corrections, exact weight times, current setup, model support, and unresolved alerts. Use current records rather than a previous day's report.
2. Remove already-satisfied work. A confirmed eligible weight is different from a queued draft. Do not ask for a duplicate merely because a filter or date changed.
3. Prioritize necessary follow-ups. Missing post-water anchors, credible measurement problems, genuinely overdue observations, and pots approaching or at a relevant window come first.
4. Choose a few flexible learning samples only when useful. Favor the pot for which a reading would reduce meaningful uncertainty or prevent its evidence becoming stale. An unfamiliar curve is a reason to plan learning, not an automatic lifelong daily assignment.
5. Spread flexible work within defensible windows. Use each pot's last actual reading and current evidence. Do not use fixed groups that can put a rapidly changing pot on the wrong day.
6. Explain exceptions briefly. If many pots really need weights today, include them. If none needs one, the report can say “No weigh-ins due” after a complete review.

For any request, the useful question is: What will this weight help decide or validate? If missing leaf-cycle, drainage, or other plant-specific confirmation is the only obstacle to watering, another gram reading may not resolve it. The daily report should retain any plant-specific Check-only exception, distinguish it from Water candidates, and avoid repeated weights that cannot resolve the missing evidence. A dry-reference hit alone stays in the hold/monitor category until a plateau is confirmed; another same-evening weight should not be requested merely to force that signal.

Partial-Drying Houseplants

P31 / #7 Peperomia Bicolor and P32 / #8 Tricolor oyster plant need manual partial-drying checks rather than a cactus-style whole-pot plateau. Peperomia should partially dry without prolonged drought; the oyster plant uses an upper 1–2-inch check scaled to its actual root-ball depth once measured. The watering exceptions link the supporting species guidance.

Request a weight when it can establish a real baseline, clarify a changing trend, check a credible measurement problem, or support a justified follow-up. Missing root-zone moisture or plant-condition evidence calls for that physical check; repeated weights cannot supply it. Keep their Check-only decisions separate from main Water candidates, and do not prolong drought to obtain a flat curve. Start their weight history with actual measurements during ordinary care.

Leaf-Cycle Plants

The same evidence rule applies to P35 / #9 shared Lithops, P36 / #10 split rock, and P28 / G3 Royal Flush: another weight cannot establish the leaf-replacement stage. Their manual watering exceptions take priority over a weight-only forecast. P35 is now one shared pot and P36 a separate pot, both repotted into setup 2 on September 23. Their initial setup-2 Routine weights are 1247 g for P35 and 713.5 g for P36, with Dry medium and firm plants recorded separately. Neither is a verified dry baseline, and neither repot batch records a Water event. Keep nursery/setup-1 observations outside the new setup's comparisons; weigh the entire P35 container once, not its four visible heads separately.

Avoid Indefinite Deferral

Anchor a proposed next observation to actual records and a plausible measurement window. Do not recalculate “today plus three days” every morning while no new evidence arrives. That would continually move the appointment away.

Carry genuinely overdue evidence-based follow-ups into today. An old every-other-day flag alone need not make an otherwise predictable pot urgent under the adaptive policy, but every deferral still needs a reason. When a longer gap cannot be justified, use the repository's conservative cadence temporarily for the affected pot.

Examples of the Policy

Use these examples to choose which measurement would help next.

Understanding the Weight Metrics

Use positive elapsed time between distinct eligible readings from the same watering cycle and setup. The current implementation resolves timestamp ties using correction-aware record ordering and excludes unusable observations.

For an illustrative sequence of 400 g on day 0, 390 g on day 2, and 387 g on day 3, all measured at the same time of day:

The first interval lost 5 g/day and the second 3 g/day. Their simple mean, 4 g/day, would give unequal-duration intervals equal influence. The total-change/total-time value is the relevant elapsed-time-weighted rate here. A mean weight has units of grams, not grams per day.

Negative signed change means the weight fell. Positive loss rate means drying-related mass loss; a negative loss rate means a gain. Neither proves its cause. Blank metrics mean insufficient usable evidence, not zero loss. A modeled cycle-loss estimate is also distinct from an observed two- or three-reading rate.

Care Days, Saved Readings, and Retries

The workbook uses America/New_York and a 4 a.m. local care-day boundary, including daylight-saving changes. A Monday 3 a.m. reading belongs to Sunday's care-day progress; Monday begins at 4 a.m. The observation keeps its actual Monday timestamp. The rollover uses local wall time, not an assumed fixed UTC offset.

Rollover alone is not a reason to repeat a recent reading at 9 a.m. The daily task should use the actual elapsed time and adaptive need. A new watering after that reading or a genuine measurement problem can justify a further weight.

Queued means the device is holding an entry. Saved today means the relevant observation has been confirmed in History for that care day. Send or recover the existing request before entering it again; a retry should retain its original identity. See saving and corrections.

What Causes the Schedule to Tighten Again

• A real repot or changed weighing setup invalidates comparison with old-setup anchors.
• Changed light, airflow, temperature, or plant activity may make the earlier drying pattern less predictive without changing the setup number.
• Unexpected gains, rapid loss, conflicting fits, partial watering, or stale evidence can make a closer observation useful.
• Sparse or aged-out training may require another informative completed cycle.

The model does not automatically interpret an Other note about lighting, prune away an unexplained mass change, or know that a plant is resting. The reviewer must use the recorded context and avoid inventing confidence. Any temporary increase should have an evidence-based reason and a path back to less work.

Sources and Implementation References

• Daily task prompt — the owner's adaptive workload policy and report restrictions.
• Dry-down learning — eligible cycles, forecast basis, and learning limits.
• Daily care and Integrity — the retired calendar and maintained integrity checks.
• Recent-weight metrics — signed changes and elapsed-time rates. See the current detector rules for plateau evidence.
• Logger source — recentWeightMetrics_, learnedDryDownCurves_, and care-day handling.
• Practical care notes — collection-specific interpretation and the limits of scale evidence.