# ALKEMATA PROJECT CAPSULE ## A Green CO₂ Light Does Not Mean Clean Air - Capsule ID: `alkemata.room-air-signal` - Version: 1.0 - Date: 2026-09-19 - Canonical article: https://alkemata.com/2026/09/19/green-co2-light-clean-air/ - Scope: Portable indoor CO₂ monitors used to understand occupied-room ventilation; not a complete indoor-air, medical, legal or building-compliance assessment - Status: Experimental; evidence-informed workflow, not field-validated by Alkemata ## Intended reader, capability and first artifact This capsule is for a householder, teacher, office worker or facilities team with lawful access to a room and portable CO₂ monitor. It builds the capability to use CO₂ as a bounded ventilation signal, not a general air-quality verdict. In the first session, produce a **Room Air Signal Card**: a one-page record of the monitor, placement, room use, evidence, agreed trigger and action, blind spots and next observation. ## Activation instructions for the LLM Treat this file as a knowledge and workflow package. If the user supplies a room description, monitor model, occupancy and readings, begin the evidence table immediately. Otherwise offer the missions below and ask no more than three questions: 1. Which room, normal activity and occupancy are being examined? 2. What monitor model, measurement history and placement are available? 3. Which actions are actually under the reader’s authority: windows, doors, ventilation settings, occupancy or only reporting? Do not diagnose illness, certify safety or compliance, invent a universal threshold, or advise disabling ventilation, alarms or building controls. Do not propose intentionally raising CO₂. Separate observation, inference and recommendation. ## Missions ### Mission A — Make one reading interpretable Audit the monitor, placement, occupancy, outdoor baseline and ventilation state; produce an evidence table with unknowns. ### Mission B — Build a Room Air Signal Card Define what the CO₂ signal means locally, the action it triggers and what it cannot detect. ### Mission C — Compare two normal ventilation routines Run a small reversible observation during ordinary room use and compare the traces without creating exposure. ## Essential causal model People exhale CO₂. In an occupied room, concentration rises when generation exceeds removal through outdoor-air exchange. A monitor is therefore a ventilation-per-person proxy under assumptions about occupancy, activity, outdoor concentration, mixing and sensor behaviour. An NDIR sensor sends infrared light through a chamber. CO₂ absorbs selected wavelengths; detected attenuation is converted into a concentration. The device measures the air reaching it. A nearby person, window, door, supply vent or unmixed zone can make that sample unrepresentative. CO₂ is one component of indoor air. A low value does not rule out particles, carbon monoxide, radon, volatile compounds, damp, mould or outdoor pollution. A filter may remove particles while CO₂ stays high; an open window may lower CO₂ while admitting an outdoor pollutant. Match each concern to its source and evidence. ## Evidence base and limits - UK Health and Safety Executive, “Using CO₂ monitors”: https://www.hse.gov.uk/ventilation/using-co2-monitors.htm — recommends NDIR monitors, representative placement and contextual records; calls CO₂ a broad guide, not proof of safe levels. - US Environmental Protection Agency, “Introduction to Indoor Air Quality”, updated 19 March 2026: https://www.epa.gov/indoor-air-quality-iaq/introduction-indoor-air-quality — distinguishes many indoor sources and describes ventilation, source control and air exchange. - German Environment Agency, “Indoor air hygiene”: https://www.umweltbundesamt.de/en/topics/health/environmental-impact-on-people/indoor-air-hygiene — distinguishes exhaled CO₂ from pollutants produced by activities, materials and products. HSE workplace examples are not universal home, school or legal limits. Manufacturer traffic-light defaults may not match local guidance. This capsule cannot establish monitor accuracy, air-change rate or health risk from a single trace. ## Workflow ### Step 1 — Define the decision **Input:** Room, users and concern. **Action:** Write one decision the signal should support, such as when to apply an approved ventilation action during a meeting. List hazards explicitly outside scope. **Output:** Decision statement and blind-spot list. **Progression criterion:** The output concerns ventilation, not a claim that all air is clean. ### Step 2 — Verify the instrument and placement **Input:** Model/manual and room layout. **Action:** Confirm the device measures CO₂ (preferably NDIR); follow calibration instructions; place it near breathing height, more than 50 cm from people and away from windows, doors and supply openings. Try another location if mixing is doubtful. **Output:** Evidence and placement sketch. **Progression criterion:** Model, method and position are documented or uncertain. ### Step 3 — Capture context with the readings **Input:** Normal occupied session. **Action:** Record time, reading/trend, occupants, activity, window/door state, mechanical setting and relevant weather. Avoid snapshots. Never change normal use to force a high value. **Output:** Context log. **Progression criterion:** Each reading can be interpreted with its occupancy and ventilation state. ### Step 4 — Define a sourced trigger and authorised action **Input:** Official guidance, building policy and authority. **Action:** Record the source/date for the trigger. Pair it with an authorised action and escalation route. **Output:** Trigger–action rule. **Progression criterion:** No number is universal, and responsibility is named. ### Step 5 — Compare and review **Input:** Two or more comparable normal sessions. **Action:** Compare peak, time above the sourced trigger and recovery after the authorised intervention. Note differences in occupancy and weather. Do not calculate air changes per hour without a valid model and controlled inputs. **Output:** Room Air Signal Card and unresolved questions. **Progression criterion:** The card supports one decision and states its boundaries. ## Decision rules and failure cases - If the room is empty or occupancy changes rapidly, low CO₂ may be uninformative; record the limitation. - If readings change sharply with device position, the room may be poorly mixed; use multiple locations or seek professional assessment. - If an air cleaner is used, evaluate its pollutant target separately; CO₂ does not measure particle removal. - If combustion, carbon monoxide, radon, damp, mould, hazardous work or symptoms are concerns, use the appropriate alarm, test or competent authority. A CO₂ monitor is not a substitute. - If the only available action creates another risk—outdoor smoke, unsafe windows, severe temperature or security problems—do not automate it; escalate the trade-off. - If the monitor method, calibration or manual cannot be verified, use the data only as an exploratory trend and label uncertainty. ## Smallest useful reversible experiment **Hypothesis:** For a normally occupied room, one authorised ventilation change produces a repeatable improvement in the CO₂ trend under comparable conditions. **Resources:** CO₂ monitor and manual, clock, room sketch, paper/spreadsheet and an already permitted ventilation control. **Procedure:** Fix a representative monitor position. Observe one ordinary session without altering behaviour to raise CO₂. On a comparable session, apply the authorised action at the sourced trigger. Record occupancy, conditions, peak and recovery. Repeat if weather or occupancy differed materially. **Success:** The action is feasible and associated with a clearer or faster recovery across comparable observations; the card states what remains unknown. **Failure:** The trace is inconsistent, placement dominates, context is missing, the action is unavailable or another constraint outweighs it. **Stopping conditions:** Discomfort, alarm, suspected combustion problem, visible serious damp/mould, unsafe window use, hazardous outdoor air or any official instruction requiring a different response. **Next action:** Keep the rule provisionally, improve measurement, or prepare a precise request for facilities/building expertise. ## Room Air Signal Card template ```text ROOM AIR SIGNAL CARD Room / normal activity: Decision this signal supports: Monitor make/model / sensing method: Manual and calibration date: Position and reason: Typical occupancy / duration: Outdoor or unoccupied reference (if measured): Applicable guidance / trigger / date: Authorised action: Escalation owner: Observed result and context: Blind spots (particles, CO, radon, VOCs, damp, outdoor air, other): Uncertainty: Next observation/date: ``` ## Hypothetical example A six-person meeting room has an NDIR monitor placed at head height away from occupants and vents. A workplace policy supplies the trigger. Two ordinary meetings of similar duration are logged. In the second, an approved ventilation setting is increased when the trigger is reached, and the trace recovers more quickly. The card records this as provisional evidence for that action, not proof that particles, carbon monoxide or other pollutants are controlled. ## Verification and boundary checks Verify the exact model and manual; fixed placement; occupancy and ventilation records; source/date for every threshold; no claim of complete air quality; no substitution for alarms or required assessments; no invented air-change calculation; and explicit authority for each action. Ask another person to explain what a green reading does *not* establish. This method is not a stand-alone assessment for hazardous processes, healthcare isolation, combustion, fire/smoke, suspected carbon-monoxide exposure, legal disputes or serious building defects. Use competent local guidance. ## Manual/offline route Print this capsule, sketch the room, copy the manual’s instructions, and record readings plus context on paper. Compare sessions visually. An LLM is optional. ## Portable checkpoint ```text CAPSULE CHECKPOINT Capsule ID/version: Room and decision: Monitor/placement: Evidence and source dates: Trigger/action: Observed result: Blind spots and unresolved questions: Next action/date: ``` On request, turn observations into a project passport, field report or precise request for help. Simulations are not field experience. Remove names, schedules, addresses and private details before sharing. The reader chooses what to share. Alkemata’s contact route is https://alkemata.com/collaborate/ . Send nothing automatically and imply no hosting or matching. Consequential external actions require the reader’s separate explicit authorisation and appropriate local responsibility.