Daily checks: the self-completing facility round
HomeOperationsDaily checks: the self-completing facility round
Operations · Daily checks

Daily checks: the self-completing facility round

The best daily check is the one that mostly fills itself in. Let Home Assistant confirm everything it can measure, leave the human only the physical walk-around, make the rest one tap, and the check gets done every day, honestly, with an audit trail that holds up.

Operations & SOPs7 diagramsResearch-backed · 13 sources~16 min read
In this guide
01 Why daily checks fail, and the fix02 Accuracy, self-review, and grain-of-salt notes03 The words you need04 What makes a check actually get done05 What to actually check, and when06 Auto-complete it with Home Assistant07 Make the human part almost nothing08 Getting people to do it, honestly09 The audit trail, and how to build it
Start here

Why daily checks fail, and the fix

Every facility has a daily check sheet. Most are half-fiction. Not because staff don't care, but because the sheet is slow, nobody owns it, and nothing visibly happens with it. The fix is not more discipline. It is better design: make the check fast, make most of it complete itself, and make the part a human does almost effortless.

A famous surgical study found records showing 100% checklist compliance while observers watched only 4 of 13 items actually get done[4]. A ticked box is not proof of a done check. This guide builds a daily check where the ticks are earned: the measurable items are confirmed by your sensors, and the human items are so quick there is no reason to fake them.

The whole idea in one line

Home Assistant auto-ticks every check it can measure (climate, lights, irrigation, doors, power), the grower does only the physical walk-around, one tap or an NFC scan logs the rest, and a missed or out-of-range item escalates on its own. The sheet becomes a short list of real, human-only items, on a calm green dashboard.

Who this is for

Any indoor or controlled-environment grow with sensors and an irrigation controller, scaling from a tent to a licensed medical facility. The behaviour and checklist principles hold even with no automation at all, the Home Assistant section is how you remove the busywork.

How sure is this?

Accuracy, self-review, and grain-of-salt notes

How sure is this paper?

We've gone to great lengths to keep these guides honest. One of the main ways we do that is self-review: we actively look for claims that are subjective, only lightly backed by literature, or based on grower practice rather than a controlled study — and we call those out instead of dressing them up as settled science.

Often there simply is no paper for the decision you're making. In those cases we're drawing on what other growers report and what has worked in our own rooms. That can still be useful — but it is not a lab proof. Do what works for your plants, your room, and your meters. If a table disagrees with your crop, believe the crop and log the difference.

Solid
Well supported by plant science, standards, or broad multi-source consensus
  • Checklists improve process reliability in high-risk work
Operational
What many growers and rooms actually run — start here, then tune
  • HA auto-tick + walk-around hybrid for grow ops
Grain of salt
Subjective, thin literature, single studies, or “this works for us” practice
  • Surgical mortality reductions as evidence of cannabis yield gains
  • Home Assistant logs as default validated GxP/GACP audit systems
  • Implementation-intention effect sizes as '2–3×' without base rates

See something glaringly wrong? Tell us and we'll fix it. Please open a GitHub issue with the paper name and what looks off (include a source if you have one): Report an accuracy issue. Local law, labels, and licences always override any recipe here. Inline notes labelled grain of salt flag the highest-risk over-trust points in the text.

Vocabulary

The words you need

Pause pointA named moment a short checklist runs at (lights-on, first irrigation, pre-dark), instead of one long sheet done whenever[1].
Killer itemA check that is both easy to skip and dangerous to miss. The few items a list must force, like dehumidification ramping at lights-off.
Read-do vs do-confirmTwo checklist styles. Read-do: perform each step as you read it, for setup and calibration. Do-confirm: do the round from memory, then pause to confirm nothing critical was missed, for the experienced daily walk[1].
Auto-tick (self-completing check)A check item a system marks done by itself once its telemetry confirms the condition, so the human only handles what a sensor cannot judge.
Alert vs action limitTwo thresholds per reading: a softer alert limit that warns, and an action limit that, when crossed, forces a defined response and a corrective-action note[9].
Pencil-whippingSigning off a check without doing it. A predictable response to a check that is slow, unowned, or never acted on, not simply carelessness.
Implementation intentionAn if-then plan tying a task to a fixed time and person, which makes follow-through far more likely[3].
The science

What makes a check actually get done

Adherence is a design problem, not a character problem. Decades of checklist and behaviour research point at the same handful of levers.

Why a check gets done: B = M × A × Pbehaviour fires only when all three line upMotivationwants to×Abilityeasy to×Promptreminded to→The lever is Ability: make the check easier, not the staff keener.Motivation swings day to day; a check that takes 20 seconds and one tap survives a bad day. Auto-fill what you can.
Figure 1. The Fogg behaviour model: a behaviour happens only when motivation, ability and a prompt line up at once[2]. Motivation swings day to day, so the durable lever is ability: make the check easy and fast, do not rely on staff being keen.

The levers, in order of impact

  • Cut the friction. Get each check under about 30 seconds, mobile, one tap. Every field you remove raises completion; auto-filling what the system already knows is the single biggest win[13].
  • Short lists at pause points. Five to nine killer items per list, one screen, run at a trigger moment, not one giant sheet[1].
  • Fixed time, one named owner. Tie each check to a person and a moment ('at lights-on the AM tech does X'); this if-then framing has a medium positive effect on goal attainment (meta-analytic d≈0.65)[3]. Name a backup for the owner-absent case.
  • Close the loop. Show the worker their data was used and acted on. The top cause of disengaged faking is submitting into a void where nothing ever happens.
  • Make completion visible. A team board of who did and did not do their checks lifts adherence sharply, as long as it reads as shared transparency, not surveillance.
Checks can move real outcomes

When genuinely performed, the WHO Surgical Safety Checklist cut deaths from 3.7% to 1.4% and lifted adherence to key steps from 18.6% to 50.7%[5]. The gain comes from the check actually happening, never from the mandate or the tick-rate alone.

The content

What to actually check, and when

Split the day into a few short pause-point lists, each tied to the moment it matters. Each line has a pre-set green / amber / red limit so the judgement is already made.

Not one big sheet: short lists at pause pointseach runs at its trigger moment, 60-90 secondsLights-onopening:climate,photoperiod,equipmentFirst irrigationroot-zone: VWC,EC, pH, runoffMid-daycrop walk + IPMscoutPre-darkclosing:dehumidify,security, wasteOn alarmresponsechecklistFive short pause-point lists beat one long sheet: each is tied to a moment, fits on one screen, and stays under 90 seconds.
Figure 2. Five short lists beat one long sheet: each is tied to a moment, fits on a screen, and stays under about 90 seconds[1].
A complete daily structure. Read-do for measured/calibration tasks, do-confirm for the experienced rounds.
Pause pointStyleKiller items
Lights-on / openingdo-confirmclimate in band, photoperiod on schedule, no flower light leaks, equipment online
At first irrigationread-doroot-zone VWC / EC / pH, overnight dryback, runoff EC/pH, feed vs recipe[8]
Mid-daydo-confirmcrop walk, IPM scout (type/place/severity/photo), anything off
Pre-dark / closingdo-confirmdehumidification ramped, night airflow on, doors/vault secure, waste logged
On alarmread-dothe specific response steps for that alarm
Green / amber / red, with a forced action on redpre-set the limits so the judgement is already madered: act nowamber: watchgreen: okamberredA reading drifting into red forces a corrective-action note before the check can be closed, so a problem can never be ticked away.
Figure 3. Pre-set the limits so the call is already made, and force a corrective-action note on any red so a problem can never simply be ticked away[9].
The environment killer item people miss

Confirm dehumidification ramps within minutes of lights-off and air keeps moving all night. A quiet failure here is how a clean late-flower room turns into a bud-rot room overnight.

The automation

Auto-complete it with Home Assistant

Most of a daily check is data the building already knows. Let Home Assistant confirm those items so the human never re-types what a sensor can prove.

How a check ticks itselfHome Assistant confirms the measurable items so the human does notSensorstemp, RH, CO2, doors,tank, pump, power→Templatebinary_sensoris every reading inband?→Automationif true through thewindow→To-do itemticked: 'EnvironmentOK'The operator opens the list and sees only the items a sensor cannot confirm. Everything else is already green.
Figure 4. The pattern: sensors feed a template binary_sensor that defines 'in range'; an automation calls todo.update_item to tick the item once that holds true through the check window[6][7].
Split the list: what the system knows vs what a human must seeAuto-ticked by Home Assistant✓ Temp / RH / VPD / CO2 in range✓ Lights on/off, photoperiod, DLI✓ Doors closed outside hours✓ Tank level OK, no leak/flood✓ Pump ran, runoff in band✓ Root-zone EC / pH / WC in band✓ All equipment online✓ Fridge / chiller in rangeNeeds a human (one tap / NFC)□ Visual plant & canopy inspection□ Pest / IPM scout□ Structural / leak eyeball□ Sanitation done & verified□ Consumables / stock□ Anything that smells or looks off
Figure 5. Split every item up front: telemetry-confirmable ones tick themselves; only the genuinely physical ones need a human, as a one-tap toggle or an NFC tag scan at the station.

The building blocks

  • The list: a local_todo to-do list is the daily sheet; items are ticked by automation via todo.update_item[6].
  • The 'pass' test: a template binary_sensor per item that is on only when every reading is in band (climate, photoperiod, doors-closed, tank, runoff, all-online).
  • Counters for the record: an integration (Riemann) sensor plus a daily utility_meter for DLI and runoff volume; history_stats for light and pump run-time.
  • The nag: the alert integration repeats and needs acknowledging, ideal for 'fridge out of range' or 'checks not all done by 09:00'[7].
  • Proof of presence: an NFC tag at each station ticks its walk-around item when scanned, so a physical visit is logged, not just claimed.
  • The audit trail, free: long-term statistics keep every reading, so the daily environmental and irrigation record exists with no extra logging[7].
Worked example, in words

Build binary_sensor.environment_ok = temp, RH, VPD and CO2 all in their stage bands. An automation watches it; once it has held true across the lights-on window, it calls todo.update_item to complete 'Environment in range' on today's list. The grower opens the list and that line is already green, with the numbers in the history for the record.

The interface

Make the human part almost nothing

For the items a person must do, design for the lowest possible interaction. The happy path should be nearly empty; effort appears only where there is a problem.

Default to pass, tap only for the exceptionthe happy path is almost emptyOpen listitems default to OK→Walk the routeNFC tap provespresence→All good?one 'Pass all'submit→Something off?tap it: photo + noterequired3 taps or fewer on a clean day.Only a flagged item expands to demand a photo and a note, exactly where the audit trail needs to be richest.
Figure 6. Default every item to pass and submit with one control; only a flagged exception expands to demand a photo and a note[13]. Hold every common action to three taps or fewer.
  • Default to pass. Items start OK; the operator only ever touches one to flag an exception. A passing day is a few taps end to end.
  • Exception-only evidence. A fail expands to require a photo plus a short note; passing items capture nothing. The audit trail is richest exactly where there is a problem.
  • NFC primary, QR fallback. An NFC tap auto-logs checkpoint, time and identity and is hard to spoof; print a QR on the same placard for phones without NFC.
  • Route, don't hunt. Model the walk-around as an ordered route; let the next station's card come up on arrival rather than scrolling a list.
  • Voice for the note. Spoken exception notes are the fast path when hands are gloved or on the plant, and they measurably increase how often observations get logged.
Make it stick

Getting people to do it, honestly

Two failure modes remain even with a great list: it gets skipped, or it gets faked. Design against both directly.

Miss or out-of-range escalates, tier by tierthe safety net that makes ownership realItem overdue or reading red↓Push to the named owner↓Unresolved? frontline supervisor↓Still open? senior managementDesign the owner-absent branch explicitly (PTO / sick) so an escalation never silently dead-ends. Escalate only meaningful misses to avoid alert fatigue.
Figure 7. A missed item or a red reading escalates on a tiered path: owner first, supervisor next, management only if unresolved. Design the owner-absent branch explicitly so it never silently dead-ends.

Anti-pencil-whip, by design

  • Auto-capture who, where and when. Pre-stamp timestamp, identity and station (and any sensor reading). This removes a typing step and makes fabrication detectable, the worker cannot claim a presence or a value the system contradicts.
  • Do not weaponise the tick-rate. Auditing box-count and attaching sanctions to it manufactures faking, the 100%-recorded, 4-of-13-done trap[4]. Spot-audit the reality of a sample instead.
  • Embed in the workflow. Time checks so they do not collide with the busiest moments; make the check a natural step, not a resented interruption.
  • Kill perverse incentives. No volume quotas that reward rushing; make it safe to report a real failure rather than soften or hide it.
The owner-absent branch

The most common escalation bug is no path when the owner is on leave or sick. Name a backup and route to them automatically, or the safety net quietly fails on exactly the days you need it.

Proof & build

The audit trail, and how to build it

Because the data entry is near zero but the system stamps who, where, when and what on every action, you get the easiest-to-fill check and a defensible record from the same design.

  • The record is automatic. Long-term statistics hold every environmental, light and irrigation reading as the daily log[7], the human items carry a timestamp, identity and any exception photo.
  • Roll up for compliance. Feed the daily fields into per-lot records (inputs, conditions, deviations, corrective actions) for operational seed-to-harvest discipline (regulated GACP/GMP still needs validated systems)[10][9].
  • Trend, don't just file. Set alert and action limits from your own history and review them weekly so drift is caught before it becomes a deviation[9]. Sanitation, hygiene and pest logs follow the same record discipline[12][11].
  1. 1
    Make the list
    One local_todo list = today's sheet, seeded each morning with the day's items.
  2. 2
    Define 'pass' in software
    A template binary_sensor per measurable item: the machine definition of in-range.
  3. 3
    Auto-tick
    One automation per item: when its binary_sensor holds true through the window, complete the to-do item.
  4. 4
    Add the human items
    input_boolean toggles or NFC tags for the physical walk-around; default-to-pass UI.
  5. 5
    Escalate & record
    Alert on overdue or red items; rely on statistics for the audit trail; review trends weekly.
Treat the check as living

Pilot for about two weeks, watch where it is skipped or breaks, cut non-killer items, fix wording in growers' terms, and re-issue. Re-test after any facility or automation change. A daily check is a product you maintain, not a form you laminate.

Related papers
Plant-state dashboard
From telemetry to intelligence
The closed loop
Levers, signal & plant state
Grow-room systems
The whole room as one system

References

  1. Gawande A (2009). The Checklist Manifesto: How to Get Things Right. Metropolitan Books. (read-do vs do-confirm, pause points, 5-9 killer items.) (industry/manufacturer or non-journal source) https://atulgawande.com/book/the-checklist-manifesto/
  2. Fogg BJ. The Fogg Behavior Model: B = MAP (behaviour occurs when Motivation, Ability and a Prompt converge). Stanford Behavior Design Lab. (industry/manufacturer or non-journal source) https://behaviormodel.org/
  3. Gollwitzer PM, Sheeran P (2006). Implementation intentions and goal achievement: a meta-analysis of effects and processes. Advances in Experimental Social Psychology 38:69-119 (medium-to-large effect, d=0.65). https://doi.org/10.1016/S0065-2601(06)38002-1
  4. Observational study of surgical-checklist use: high recorded compliance did not guarantee the items were actually performed (records showed 100% adherence; observers found 4 of 13 items done). PMC4484042. https://pmc.ncbi.nlm.nih.gov/articles/PMC4484042/
  5. Haynes AB, Weiser TG, Berry WR, et al. (2009). A surgical safety checklist to reduce morbidity and mortality in a global population. New England Journal of Medicine 360(5):491-499. https://www.nejm.org/doi/full/10.1056/NEJMsa0810119
  6. Home Assistant: To-do list (todo) and Local to-do (local_todo) integrations, todo entity and todo.update_item action used to auto-complete checklist items. (industry/manufacturer or non-journal source) https://www.home-assistant.io/integrations/todo/
  7. Home Assistant: Template (binary_sensor), Alert, Statistics/Recorder, Tag (NFC) and Schedule integrations used to define 'in range', escalate, log the audit trail and prove presence. (industry/manufacturer or non-journal source) https://www.home-assistant.io/integrations/template/
  8. AROYA. Crop-steering and drip-and-drain root-zone monitoring (substrate VWC, EC, pH, dryback, runoff) as the daily root-zone measurement. (industry/manufacturer or non-journal source) https://aroya.io/
  9. EU GMP Annex 1 / EMA environmental-monitoring guidance: define alert and action limits, trend routine monitoring data, require corrective action on excursions. (industry/manufacturer or non-journal source) https://health.ec.europa.eu/medicinal-products/eudralex/eudralex-volume-4_en
  10. World Health Organization (2003). WHO guidelines on good agricultural and collection practices (GACP) for medicinal plants. (industry/manufacturer or non-journal source) https://www.who.int/publications/i/item/9241546271
  11. GLOBALG.A.P. Integrated Farm Assurance, Crops / Fruit & Vegetables control points and compliance criteria (hygiene competence, maintained irrigation equipment, record retention). (industry/manufacturer or non-journal source) https://www.globalgap.org/
  12. HACCP prerequisite programmes / Sanitation Standard Operating Procedures: daily sanitation checklists, hygiene checks and pest-control logs (what, how, how-often, who-verifies). (industry/manufacturer or non-journal source) https://www.fda.gov/food/hazard-analysis-critical-control-point-haccp/haccp-principles-application-guidelines
  13. Industry analyses of digital vs paper inspection checklists: higher completion rates, faster rounds and fewer documentation errors, driven by lower friction and enforced completeness. (industry/manufacturer or non-journal source) https://www.fiixsoftware.com/cmms/mobile-cmms/

Citations marked in-text as [n] map to this list. Primary literature and official guidance except where noted. Cannabis tissue culture is strongly genotype-dependent, verify dilutions, hormone doses and local regulations against the primary sources before relying on them.