Tunable White 2700K–5000K: Office Daylight Curves

A tunable white office system runs a daylight curve when colour temperature and output are scheduled against the position of the sun rather than switched between fixed scenes. In Hong Kong the practical operating band runs from 2700K at the end of the day to 5000K at the morning peak, inside a luminaire whose full tuning range is wider: SUMLINK's linear range is 2700K–6500K. The curve is defined by four numbers at every node: clock time, CCT, illuminance and transition duration. Two engineering limits decide whether the result reads as daylight drift or as a sequence of visible steps: the chromatic path the engine takes between its warm and cool emitters, and the rate at which it travels it. This guide covers the band, the solar anchors, the ramp rates that stay below published detection thresholds, the Duv tolerance to specify, and the control layer that carries the schedule.

Key Facts

1. WELL v2 Feature L03 sets the daytime target in melanopic terms, not Kelvin. One point requires at least 150 EML [136 M-EDI(D65)] for at least four hours beginning no later than midday, measured on the vertical plane at the occupant's eye (the standard's table specifies 45 cm above the work-plane); three points require 240 EML [218 M-EDI(D65)]. Source: WELL v2 feature L03 threshold tables as reproduced in USGBC's 2025 WELL Light features deck and in the PNNL-32514 comparison table (which also records the 2021 Q4 revision of the three-point tier to 275 EML [250 M-EDI]).
2. Meeting those thresholds with electric light alone is harder than the Kelvin number suggests. In the Cook County / PNNL office pilot, designing for the recommended EML and circadian-stimulus thresholds produced horizontal and vertical illuminance levels roughly three times IES recommendations for visual tasks — and the thresholds could not be met at 100% of workstations in the open office with electric lighting only. Source: PNNL-32514, Lighting for Health and Wellness Recommendations in Offices (January 2023), executive summary.
3. Circadian criteria carry a measurable energy cost in interior zones, and none in daylit ones. In the SCE / LBNL FLEXLAB office test bed, annual lighting energy for interior (non-daylit) zones rose 11% to 42% above an LED baseline depending on configuration; in the daylit zone, meeting the same criteria required no extra lighting power. Tunable pendants used 31% (300 lx condition) to 21% (500 lx) more than baseline; the report notes the added energy falls between 09:00 and 13:00 and is therefore off-peak. Source: Laboratory Evaluation of Tunable White LEDs for Circadian Lighting in Commercial Offices, ET18SCE1020, April 2021, Table ES-1.
4. A well-shaped curve is below the threshold of human detection; steps applied instantly are not. Published mean detection thresholds for a chromatic change along the daylight locus, viewed over 10 seconds, range from 15 to 2 CIELUV ΔE units depending on the base colour temperature and the direction of change. The whole 5000K → 2700K journey is about 86 ΔE on the same scale; spread over an eight-hour day that is 0.03 ΔE per 10 seconds — roughly 1.5% of the lowest published threshold — while the same journey taken as ten steps and applied instantly is 8.6 ΔE per step, more than four times that threshold. Sources: Pastilha, Gupta, Gross & Hurlbert (2020), Journal of Vision 20(13):18; journey distance and rates are internal estimates computed by SUMLINK from the CIE blackbody locus approximation (§K item 7; script `scripts/insights06_cct_math.py`, output `docs/evidence/t_a17f0e2e/cct-math.txt`).
5. Within a narrow band, an idealised two-channel tunable white blend stays inside the ±0.006 Duv white box; widen the band and it does not. Computing the straight mixing chord between emitter endpoints placed exactly on the blackbody locus gives a worst-case |Duv| of 0.0036 for a 2700K → 5000K pair, 0.0063 for 2700K → 6500K (just outside the box) and 0.0107 for 2200K → 6500K, each occurring in the mid-range and each below the locus — the pink side, which is where a two-channel mix lands. The ANSI C78.377 white-light quadrangle is bounded exactly 0.006 Duv from the locus. Source: internal estimate — SUMLINK computation (§K item 7); quadrangle limit per ANSI C78.377 as described in a manufacturer product-consistency datasheet (secondary; see §K and the verification table).

1. What "daylight curve" means, and what it does not

A daylight curve is a schedule: a mapping from time of day to a pair of setpoints, colour temperature and light output, plus the transition behaviour between them. It is not a colour mode, not a scene list, and not "tunable white" on its own — a fixture can be tunable across 2700K–6500K and still be run at a single fixed 4000K all day, in which case no curve exists.

The distinction matters commercially because the curve is the part that has to survive commissioning. Hardware that tunes is a purchase; a curve that holds its nodes, ramps without visible steps, and can be re-verified a year after handover is an operating asset. The office script published under Lighting Programs is stated this way: morning activation in the 5000K–5700K region at high output, a neutral afternoon plateau around 4000K, and a gradual shift to 3000K–2700K with reduced output after hours, described on the site as after-hours protection rather than stimulation. Both deployment paths on that page — a wireless retrofit that keeps existing luminaires and changes the drivers, and a full DALI-2 build — are specified against the same time-spectrum program, which is the point: the curve is the deliverable, the gear is the carriage.

There is a second, quieter use for the same schedule. A curve that moves on known numbers gives a facilities team something to log and report, which is what turns "we installed circadian lighting" into "here is what the building delivered in June". The measurements this article asks for in §D.8 exist for that reason.

2. Why 2700K–5000K is the office operating band

Office lighting has both a tuning range and an operating band, and they are not the same number. The tuning range is what the driver and the emitters can reach; the operating band is where the schedule actually spends its hours.

SUMLINK's published linear platform, for example, is tunable white 2700K–6500K at CRI 90 (a design target on the product page). The melanopic output across that range is not linear in Kelvin — the same fixture at the same 500 lx delivers 225 melanopic EDI at 2700K and 530 at 6500K, because the melanopic ratio rises from 0.45 to 1.06 across the band:

(Published values from the BY001 product page — all design targets, not laboratory results; see §D.9 and §K item 6.)

Three practical consequences follow from that table.

First, the band's ends are chosen for different reasons. 2700K is the warm end of the platform — the setting the day finishes on, and the reason the curve can end warm without swapping hardware. 5000K is the top of the operating band because the published figure there already sits far above the WELL v2 one-point floor: the page gives 405 melanopic EDI at 500 lx, against a floor of 136 M-EDI(D65), a ratio of about three to one on the published number's own scale (see the unit note below). 6500K exists as headroom for short, deliberate peaks, and the product page says so in as many words.

One unit note belongs here, because the two families of numbers get mixed up in specification documents. WELL's tables are written in both EML and M-EDI(D65), related by the factor 1.103 (150 EML = 136 M-EDI). The BY001 page labels its column "Mel-EDI @ 500 lx" and its row "M/P Ratio", and states its range as "Melanopic DER, CIE S 026:2018" — three label families for one column. The published values are used here as published; if the column is read as EML, 405 EML corresponds to about 367 M-EDI(D65), and the ratio to the WELL floor becomes 2.7 rather than 3.0. Either reading clears the floor comfortably; the point is that a specification should name which of EML and M-EDI(D65) it is writing, and the page does not (§K, 未驗證項).

Second, a CCT schedule is not a melanopic schedule. If a curve is drawn as equal Kelvin steps, the melanopic output rises faster than the Kelvin number between 4000K and 6500K than between 2700K and 3500K. A curve that needs a defined melanopic level in the morning and a defined floor in the evening has to specify both quantities per node, not assume one follows the other. The relationship between the two quantities is the subject of the M/P Ratio guide on this site — this article assumes it and uses it.

Third, the upper bound is an operational decision, not a spectroradiometric one. Running 5000K on an office floor for six hours and running 6500K for six hours are both physically possible; they are not the same specification, and only one of them is usually intended when a client asks for "daylight" lighting. Writing the band's ends into the clause (§D.8) removes the ambiguity.

3. Anchor the curve to solar noon, not to clock noon

A daylight curve that is anchored to clock times drifts out of phase with the daylight it claims to track, because the sun does not keep office hours and Hong Kong's solar day is not centred on 12:00.

Computing solar position for Hong Kong (22.32° N, 114.17° E, UTC+8) gives the anchors below. These are computed values from the NOAA solar equations, not observations:

Three design consequences:

• Solar noon sits 16–31 minutes after clock noon on the four dates tested, so the curve's peak belongs just after 12:15 HKT, not at 12:00. On a multi-zone floor the difference is small; across a year of scheduled nodes it is the difference between a curve that tracks the sun and one that leads or lags it.
• The office day is not centred on daylight. A 09:00–18:00 occupancy window in June opens 3 h 20 min after sunrise and closes 1 h 10 min before sunset; in December it opens 2 h 02 min after sunrise and closes 16 minutes after sunset. The electric curve therefore has to do more of the morning work in winter and more of the evening work in every season than a midday-centred schedule assumes.
• Occupancy, not sunrise, should start the ramp. Starting the morning ramp at sunrise (05:40 in June) would run the schedule's highest-CCT hours before the floor is occupied. Anchoring the ramp to occupancy — with solar noon as the fixed mid-point and sunset as the evening reference — keeps the curve aligned to both the sun and the working day.

A seasonal parameter is the cleanest way to express this: the same node list, with its clock times shifted a few minutes per month so that solar noon stays at the peak. Fixed clock nodes are acceptable, but they should be described as fixed-clock and the resulting annual phase error acknowledged rather than left implicit.

4. The ramp rate is part of the specification

Daylight changes slowly. A schedule that changes CCT in visible steps is not imitating daylight, it is offering occupants a sequence of events to notice — and the literature gives a numerical line to design against.

Pastilha et al. (2020) measured the minimum detectable chromatic velocity of daylight-like illumination in an immersive environment, with adapting base lights at 13,000 K, 6500 K, 4160 K and 2000 K. Mean detection thresholds for a change viewed over 10 seconds ranged from 15 to 2 CIELUV ΔE units, depending on the base light's CCT and the direction of change, with cool changes becoming less noticeable from progressively warmer base lights and vice versa. The authors' own conclusion is relevant to scheduling: typical temporal changes in daylight chromaticity are likely to be below threshold detectability, at least where illuminance does not change with them.

Applied to an office curve:

Read together, the table makes one point sharply: what matters is the rate, not the step size. A curve spread over a working day moves about 1.5% of the lowest published threshold per 10 seconds, so the drift itself is imperceptible by a wide margin. A coarse ten-step schedule has steps of 8.6 ΔE — more than four times the lowest threshold — yet if each step is executed as a 15-minute fade the rate is 0.096 ΔE per 10 seconds, under 5% of the threshold, and equally invisible. Applied instantly, the same step is a visible event. What occupants actually notice is therefore not the curve and not the node list but the transition behaviour: a scene recall, a schedule change applied at a wall panel, a controller that jumps to the new node when it reconnects after a power cut. The specification should name the curve's rate and the minimum fade the control gear applies, and §D.8 does.

Three limitations belong with this table. The thresholds were measured for chromatic change under fixed viewing conditions with no simultaneous illuminance change; a real curve changes illuminance as well as CCT, and the perceptual interaction between the two is not captured here. ΔE(CIELUV) is a colour-difference metric, not a perceptual model of "noticing a change in the room" — it is used here because that is the metric the published thresholds are stated in. And the L* = 100 convention used to put the two sides on one scale is our own assumption: the paper reports thresholds in "CIELUV ΔE units" without restating the lightness convention, so anyone re-deriving these figures under a different reference lightness will scale them proportionally. The ratio between the curve and the threshold is what the argument rests on, not the absolute number.

The clause in §D.8 caps the rate at 0.5 ΔE*uv per 10 seconds — a quarter of the lowest measured threshold. The margin is deliberate: it leaves room for the illuminance step that accompanies a colour step, and for the fact that the lowest threshold applies to a base light and a direction of change that a real schedule will not always be in.

5. The mixing line is not the daylight curve

A two-channel tunable white engine has a warm emitter and a cool emitter, and mixing two light sources can only ever produce colours on the straight line between them. The daylight and blackbody loci are curves. A straight line through a curve is inside it in the middle — and for white light, "inside" is a visible tint.

Computing that chord for idealised emitters sitting exactly on the blackbody locus gives:

The comparison point is the white-light tolerance box: ANSI C78.377 quadrangles are bounded exactly 0.006 Duv from the Planckian locus (positive above the locus, negative below), and the common cause of a mid-range "pink" or "magenta" cast on a two-channel product is precisely the chord dropping below the locus at those CCTs. ENERGY STAR's colour-maintenance requirement — a maximum Δu′v′ shift of 0.007 over the first 6,000 hours — is the second reference worth carrying, because it sets the allowed drift over life that a maintenance schedule has to live with. Both figures are quoted from a secondary reproduction of the standards rather than the standards themselves (§K item 6), so they are used here as design references and not as a certification basis; a project that has to demonstrate compliance should work from the purchased texts.

Three conclusions for a specification:

1. A narrow operating band buys tolerance headroom. Inside a 2700K–5000K band, an idealised two-channel engine stays within the ±0.006 box; the same engine spanning 2200K–6500K does not. If the band can be kept narrow, the requirement is achievable with two channels; if it cannot, the engine needs a third (typically green or mint) channel blended against calibration data.
2. Duv has to be specified per node, not per product. The failure is a mid-range phenomenon, so a datasheet Duv quoted at the range endpoints proves little about 3500K.
3. The computed numbers above are internal estimates, and an idealisation. They assume both emitters sit exactly on the locus and that the blend runs along the straight chord. Real fixtures add endpoint offsets from binning and measurement, so measured mid-range Duv is normally worse than the table — which is why the clause asks for measured data at each node rather than a range statement.

6. The control layer: what actually carries the curve

The schedule has to live somewhere. Three carriers dominate office projects, and they differ in what they can hold.

DALI-2 colour control (device type 8). IEC 62386-209 defines a colour type called Tc, which is tunable white control expressed directly as a correlated colour temperature along the black-body line, alongside xy coordinate control and RGBWAF. A DALI-2 certified tunable-white control gear must support Tc; a single DALI short address can carry both colour temperature and brightness for a fixture, which is what makes one-address-per-luminaire scheduling practical. DALI-2 certification for tunable-white (Tc-only) products has been available since February 2020. Products that include only the xy colour type are not permitted to carry the DALI or DALI-2 trademark, which is a useful marker when reading a driver datasheet: "colour control, xy" is not the same claim as "tunable white". These statements come from IEC's public abstract page and the DALI Alliance's public technical pages; the standard text is paywalled and was not read (§K item 5), so treat them as a reading guide rather than as a verbatim quotation of the standard.

Analogue 0-10 V. One 0-10 V pair carries one controlled variable. Tunable white on analogue control therefore needs two channels — one for intensity, one for CCT — and the second channel's curve must be the one that keeps the mix on locus (§D.5). Where a retrofit keeps existing luminaires and only the driver changes, this distinction decides whether the existing wiring can carry the curve at all.

Wireless and Matter. A Matter-over-Thread retrofit path exists precisely for buildings where rewiring is not feasible: the drivers are replaced in situ and the schedule is held by the gateway rather than by new cabling. The published office retrofit path on our own site takes that form — Matter tunable-white drivers, a wall-mount gateway holding the time-spectrum program, and a presence sensor for daylight-compensated dimming and auto-off.

Whichever carrier is used, commissioning should verify four things about the curve beyond "the scene works":

1. Node resolution — the controller actually resolves the node granularity in the clause (a 15-minute curve is not reproducible on a controller that only accepts hour-and-minute scene times reliably).
2. Transition behaviour across the whole path, not only the morning and evening extremes: measure at four or five intermediate nodes, because that is where the mixing chord deviates most (§D.5).
3. Power-on and power-restore colour — a scheduled system that reboots into a default CCT will expose exactly the step-change event that §D.4 identified as the visible failure mode.
4. The fade actually applied by the driver — colour transitions have their own fade time and fade rate in the DT8 command set, and a controller that sends a new colour without a fade will produce a step regardless of what the schedule says.

7. Energy: what the measured studies actually show

Two independent measurement programmes bear directly on whether a daylight curve is affordable, and both are worth reading before promising a client either outcome.

Interior zones pay, daylit zones do not. The SCE / LBNL FLEXLAB study implemented non-tunable and tunable LED systems in a mock office and measured the energy needed to meet visual plus circadian criteria. In the interior (non-daylit) zone, annual lighting energy rose 11% to 42% above the LED baseline depending on configuration and illuminance condition:

In the daylit (perimeter) zone, daylight dimming more than covered the requirement: meeting circadian criteria required no extra lighting power. The study also notes that the incremental energy falls between 09:00 and 13:00 — off-peak for the utility service areas studied — and recommends specifying high photopic efficacy (at least 125–130 lm/W) and high melanopic daylight efficacy ratio (m-DER, at least 0.69–0.80) to keep the incremental cost down.

Meeting the thresholds is harder than buying tunable fixtures. The Cook County / PNNL pilot in Chicago designed to the recommended EML and circadian-stimulus thresholds and found that doing so pushed horizontal and vertical illuminance to roughly three times the IES recommendations for visual tasks, while still failing to meet the thresholds at every workstation in the open office from electric light alone. Glare control and optical distribution, not colour tuning, were the limiting factors.

For an office curve the operational reading is straightforward: use the morning peak as a scheduled window rather than an all-day setting, drive the perimeter with daylight-linked dimming, and treat the melanopic target as something the control system achieves with daylight plus a defined electric contribution — not as a number that can be bought from a fixture datasheet.

8. The clause to write into a specification

Copy-paste clause:

Daylight curve (tunable white scheduling). The electric lighting serving regularly occupied office areas shall be tunable white, controlled by a schedule with the following properties.

(a) Operating band. Colour temperature shall be continuously variable over 2700 K to at least 5000 K within a single fixture type, and the schedule shall not operate below 2700 K or above 5000 K during standard occupancy hours.

(b) Nodes. The schedule shall define at least eight nodes per day, each stating clock time, CCT, illuminance and transition duration. Nodes shall be anchored so that the peak-CCT node falls at local solar noon ±30 minutes, and the schedule shall support seasonal adjustment of node times.

(c) Ramp. Every transition between nodes shall be applied by the control gear as a fade of not less than 15 minutes. The chromatic rate at the light output shall not exceed 0.5 ΔE*uv(CIELUV) per 10 seconds, referenced to the adapting light at L* = 100.

(d) Colour path. At each node the delivered chromaticity shall be within |Duv| ≤ 0.004 of the Planckian locus, and within 0.004 of the CIE daylight locus where the daylight locus is defined (4000 K and above).

(e) Control. Colour temperature and intensity shall be controllable on a single address per luminaire with DALI-2 colour control (device type 8, colour type Tc) or, where a wireless control system is used, with an equivalent documented colour-temperature command path. Analogue 0-10 V control shall use a separate channel for colour temperature.

(f) Power behaviour. After a power interruption the luminaire shall resume the scheduled node, not a fixed default colour; the default colour on first power-up shall be documented.

(g) Verification. The supplier shall provide, for each node: measured CCT, measured Duv, measured illuminance at the specified plane, and melanopic EDI. Measurement conditions — instrument, plane and height, ambient temperature, drive level — shall be stated. A transition log recorded across one full scheduled day shall be provided at handover.

The clause deliberately asks for measured values at nodes. A specification that names only a range sells a capability; a specification that names values at nodes buys a schedule that can be checked.

9. What SUMLINK can and cannot state today

Two published SUMLINK sources are relevant, and both label their engineering values the same way.

The BY001 Minimalist Linear Wall Light product page states tunable white 2700K–6500K, CRI 90, and carries a melanopic table giving an M/P ratio of 0.45 at 2700K rising to 1.06 at 6500K, with melanopic EDI at 500 lx from 225 to 530. Its remaining photometric rows — percent flicker < 1%, SVM < 0.15, colour shift Δu′v′ ≤ 0.004 at 6,000 h and Ta 25 °C, SDCM ≤ 3-step, TM-30-20 Rf 88 / Rg 99, UGR < 19 — are all marked Design target on the page, with the page stating that design-target values are engineering values from the platform specification and are not yet confirmed by an independent laboratory report. The internal bill-of-materials records behind the same platform list the tunable-white light engine as 2700–6500 K at CRI 90, with R9 and SDCM still blank, and the driver rows for the retrofit path are recorded as pending supplier confirmation for model and output specification.

The published office program is the deployment side: a wireless retrofit path that keeps existing luminaires and replaces drivers, and a full DALI-2 path using a tunable-white linear run with unified dimming and tuning, both delivering the same time-spectrum program.

What that means in practice is narrow and specific. We can state the platform's tuning range, its published melanopic profile, and the schedule structure, and we can build a project-specific curve with its node table and verification records. We cannot state that any luminaire has been measured to a given Duv or melanopic EDI at a node, because no such measurement exists in our records yet; and the melanopic figures are model outputs from the platform's spectral specification, at a stated drive level and on a stated plane. Projects that need measured values get them from the per-project verification in clause (g) — which is why the clause asks for them rather than assuming them.

The wider catalogue of tunable-white linear and architectural products, with their published specifications and bill-of-materials data, is collected in the HCL linear light catalog. For the office layer above this article — the working-day narrative, the WELL alignment and the procurement framing — see Human-Centric Lighting Hong Kong: The Complete Office Guide.

Frequently Asked Questions

What colour temperature range should an office tunable white system cover?

Specify two numbers, not one. The tuning range should reach at least 2700K to 5000K continuously within a single fixture type; the operating band — what the schedule actually uses during occupancy — should be no wider than that, so the mid-range stays inside the ±0.006 Duv white-light tolerance. Wider ranges (2200K–6500K) remain useful as headroom, but a two-channel engine driven across the whole span drifts further off the Planckian locus in the middle.

How fast should an office daylight curve change?

Slowly enough that nobody notices it, which is easy to achieve and easy to specify — but the point is the rate, not the step size. Published mean detection thresholds for a chromatic change along the daylight locus are 2 to 15 CIELUV ΔE over 10 seconds, depending on the base colour temperature; the whole 5000K to 2700K journey is about 86 ΔE on the same scale, so a curve spread over a working day moves roughly 0.03 ΔE per 10 seconds, about 1.5% of the lowest threshold. Ten coarse steps of 230 K are 8.6 ΔE each, which is invisible if each step is applied as a 15-minute fade (0.096 ΔE per 10 s) and visible if applied instantly. Require a minimum fade of 15 minutes between nodes and cap the rate at 0.5 ΔE per 10 seconds.

Does a tunable white system follow the daylight curve automatically?

No. Tunable white describes what the hardware can reach; the curve is a schedule that has to be designed, commissioned and verified. Two things decide whether it behaves: the chromatic path the engine takes between its warm and cool emitters, and the control layer's transition behaviour — including what the luminaire does after a power interruption, since a system that reboots into a fixed colour produces exactly the visible step the rest of the schedule avoids.

What is Duv and why does it matter for tunable white?

Duv is the signed distance of a light's chromaticity from the Planckian locus, measured in the CIE 1960 (u, v) plane: positive above the locus (greenish), negative below (pinkish). It matters because a two-channel tunable white engine mixes two emitters along a straight line, while the daylight and blackbody loci are curves — so the mid-range of the mix can fall outside the ±0.006 Duv box that ANSI C78.377 uses for white light, producing a pink or magenta cast around 3500K. Specify Duv per node and ask for measured values there, not a single figure at the range endpoints.

Can an existing office be retrofitted with a daylight curve without rewiring?

Often, yes — if the drivers can be replaced in situ. On a 0-10 V installation the existing wiring carries intensity only, so colour temperature needs a second channel. A wireless Matter-over-Thread retrofit replaces the drivers and holds the schedule in a gateway, which is the route the published office retrofit path takes; it keeps existing luminaires and adds a wall panel plus a presence sensor for daylight-compensated dimming and auto-off. Schedule the night-time occupancy behaviour at the same time: it is the same curve, extended past the closing node.

Deploy This Strategy

The Office · Circadian Light Line script: node timeline, component specification, and the wireless retrofit and full DALI-2 deployment paths side by side. Deploy This Strategy →