CRI 97 measures eight colour samples and says nothing about R9, nothing about the spectrum after two years on a case. A supplier-facing specification guide to Rf, Rg, Rcs,h1 and the two colour-shift statements a jewellery counter needs.
A jewellery counter is specified by three numbers that measure three different things, and most quotes carry only the first. CRI 97 says the average of eight colour samples is faithfully rendered; it says nothing about R9, the ninth sample, which decides whether yellow gold looks warm or dull, and nothing about whether the spectrum shifts after two years on a case. TM-30 answers the second question with Rf, Rg and Rcs,h1, and Δu′v′ answers the third by measuring how far the light itself has moved on the CIE 1976 (u′,v′) diagram. This article states what each metric covers, gives the threshold pairs that WELL v2 and LEED v5 assessors read in 2026, and sets out the specification clauses and the handover checks that close the gap between a CRI 97 claim and a display that still reads the same colour next season. Every SUMLINK figure below is quoted with the label its own page carries.
Terms and units used below. CRI = colour rendering index; Ra = the CIE General Colour Rendering Index (the average colour difference over eight colour samples); R9 = the index for the ninth sample, excluded from the Ra average and added to address red hues. Rf = fidelity index (average colour fidelity, 100 = identical to the reference); Rg = gamut index (100 = neutral chroma relative to the reference); Rcs,h1 = local chroma shift for hue bin 1, the red region; Rf,h1 = local fidelity for the same bin. TM-30 = ANSI/IES TM-30, IES Method for Evaluating Light Source Color Rendition. CCT = correlated colour temperature, in kelvin. Δu′v′ = the Euclidean distance between two chromaticity points on the CIE 1976 (u′,v′) diagram — the metric used for both initial colour tolerance and colour maintenance. SDCM = standard deviation of colour matching, quoted in MacAdam steps. lx = lux; Ta = ambient temperature; SPD = spectral power distribution. The expansions are ours; the symbols are as the standards and the cited pages write them.
A CRI figure is one number produced from eight colour samples. Ra is defined in CIE 013.3-1995 as the average colour difference over those eight samples, and R9 — the ninth sample, used to address the appearance of red hues — sits outside that average. That single structural fact decides most jewellery disputes, because the metals and stones a counter carries are dominated by exactly the red end of the spectrum: yellow gold, rose gold, rubies, garnet, and the warm tones of coloured diamonds.
CIE's own position on the metric has moved. CIE PS 002:2025, second edition, records that Ra was first published in 1965 and last improved in 1974, and names three technical issues: the outdated 1960 CIE (u,v) colour space, the small number of test-colour samples, and a conceptual limit — Ra is a fidelity metric only. The same document states that the CIE now believes the time has come to start adopting the General Colour Fidelity Index Rf, defined in CIE 224:2017, and recommends that both metrics be reported in parallel while Rf is established in regulations and specifications.
Two further findings in that document matter to a display specification. CIE reports that Ra does not agree well with perceived colour rendering for LED sources that contain narrow-band spectral components. And it records preference experiments in which subjects preferred illumination that slightly enhanced the saturation of the objects they viewed, even though the chosen sources had lower Ra values.
So a CRI 97 claim on a jewellery quote should be read as: the eight samples were rendered at 97. It is not a statement about red, not a statement about saturation, and not a statement about the spectrum the customer sees in year three.
The gem trade answers this question with practice rather than with a metric, and the practice is worth reading before the specification.
GIA, whose D-to-Z colour scale is the industry standard, states that diamonds are colour-graded by comparing them to stones of known colour under controlled lighting and precise viewing conditions. Colour grading does not take place in ambient shop light; it takes place under a defined source in a controlled surround.
ISO 24016:2020, Jewellery and precious metals — Grading polished diamonds — Terminology, classification and test methods, goes further and specifies the lighting itself: colour temperature, intensity, and the distance between the stone and the light source, with reflections and visual distractions from the environment excluded, which is why grading cabinets are normally painted a neutral grey. (We quote the standard's scope as the SSEF, the Swiss Gemmological Institute, describes it in a published article on its LED daylight source; we have not read the standard text.)
That same SSEF article states the mechanism plainly. Conventional LEDs emit large amounts of blue and yellow-green light with a gap in the blue-green, while natural daylight has a more continuous spectrum, and the article adds that this discrepancy can lead to differences in perceived colour when such light is used in judging the colour of gemstones. The International Gem Society reaches the same conclusion from the buying side: a continuous spectrum, without any gaps, is ideal for viewing gemstones.
Industry design guidance divides by material rather than by a single number. Fagerhult's jewellery light guide recommends roughly 3000 K for gold, on the grounds that warm light suits the metal, and natural white or 4000 K and above for silver, white gold, chrome and crystal, where a cooler light suits the material. The same guide warns against glare, recommends narrow-beam accent rather than flat general light, and notes one reflection problem specific to jewellery: polished gold behaves as a mirror, so several small light sources aimed at a piece make it sparkle as stars, which the guide calls disturbing, and wide-distribution luminaires are its suggested remedy.
Three conclusions follow for a specification, and none of them is visible in a CRI figure.
TM-30 exists because a single average hides too much. The tutorial on the method by Michael P. Royer — 52 pages, published in LEUKOS (18:2, 2022) and archived by the U.S. Department of Energy, written about ANSI/IES TM-30-20 — records that the standard specifies 149 distinct numerical output measures and one graphic, and that describing it as "just Rf and Rg" is an error. It also sets out the values a specifier can actually use:
| Measure | What it characterises | How to read it | Typical range |
|---|---|---|---|
| Fidelity index Rf | Overall average similarity of colour appearance to the reference | Closer to 100 = greater similarity | 70 to 100 |
| Gamut index Rg | Approximate average change in chroma | Above 100 = increased chroma; below 100 = decreased chroma | 80 to 120 |
| Local chroma shift Rcs,h1 (red bin) | Average relative change in chroma for the red hue region | Above 0% = more saturated reds; below 0% = less | Roughly −20% to 20% |
| Local fidelity Rf,h1 | Average similarity for the red hue region | Closer to 100 = greater similarity | 60 to 100 |
Three reading rules come with those numbers, and each one is a trap in a tender conversation.
Rf is not a better CRI; it is a different measurement with a known limit. The tutorial states that Rf averages the magnitude of the difference for all 99 colour evaluation samples, and that two sources can reach the same Rf through opposite hue shifts, because the average does not carry direction. It calls this a substantial limitation of all average colour fidelity measures.
Rf,h1 is the TM-30 counterpart of R9. The tutorial states that Rf,h1 shares its intended purpose with R9 from the CIE 13.3 method. For a counter, that is the number to ask for when yellow gold is on display — and, unlike R9, it is an average over a small set of similar samples rather than a single sample, so it is less sensitive to one unusual reflectance curve.
Rf and Rg together say what a single index cannot. The tutorial's own guidance for architectural applications is that four values — Rf, Rg, Rcs,h1 and Rf,h1 — provide sufficient information to predict colour preference, colour vividness and colour fidelity, and that for a visual match of objects in a polychromatic scene at least Rf, Rg and Rcs,h1 should be compared. It also warns that no universal threshold defines when a difference in Rf or Rg becomes noticeable, precisely because both are averages.
TM-30 also fixes where the thresholds come from. The standard carries an Annex E of recommended specification criteria for three named design intents — Colour Preference, Colour Vividness and Colour Fidelity — at three levels of restrictiveness, stated for interior light levels of 200 to 700 lx where the space contains a variety of colours. The tutorial notes that Colour Vividness is the intent that fits display and retail lighting, and that Colour Preference could dominate in retail generally. It adds a light-level dependence that is easy to miss: preference for increased saturation falls as illuminance rises, and above 5,000 lx no chroma increase over the reference is preferred, while at very low levels — a museum case, for example — a greater increase is needed. A jewellery counter at 600–800 lx sits between those extremes, which is the region where a modest Rg above 100 is defensible and a large one is not.
One version note, because it appears in every quotation of a threshold. The current edition is ANSI/IES TM-30-24, listed free of charge by the IES with its Excel calculation tool, and its Annex E carries the specification recommendations. SUMLINK pages publish figures under the name TM-30-20, which is the edition those figures were calculated against. Where a specification or rating system names an edition, quote the edition it names; do not treat the year suffix as interchangeable.
Two voluntary systems carry TM-30 numbers into projects today, and both were read for this article from their own documents.
| System | Clause | Colour requirement as written |
|---|---|---|
| WELL v2, Feature L08 Part 1 (1 point) | "Enhance Color Rendering Quality" | At least one of: CRI (Ra) ≥ 90; or CRI (Ra) ≥ 80 with R9 ≥ 50; or IES TM-30 P1 — that is, Rf ≥ 78, Rg ≥ 95, −1% ≤ Rcs,h1 ≤ 15%. Where tunable white is used, the requirement is met at 1,000 K intervals from the lower end (minimum 2700 K) to the higher end (maximum 5000 K) |
| LEED v5, EQc2 Option 5 Path 2 (1 point) | "Quality Electric Lighting", colour rendering | Luminaires with a colour rendering index of at least 90, or meeting Table 1, "Color Rendering Requirements Using IES TM-30-20": fidelity index Rf 78 or higher, gamut index Rg 95 or higher, red local chroma shift Rcs,h1 |
The WELL v2 text is quoted from the captured feature text held in this repository, and it cites IES TM-30-24 as its reference. The LEED v5 rows for Rf and Rg are quoted from the USGBC reference guide, July 2026 edition; the numeric requirement in the Rcs,h1 row of that table did not land in our text extraction, so we state the row exists and do not quote a value for it — the WELL v2 figure (−1% to 15%) is the one we can quote verbatim.
Two consequences for a jewellery specification, and they are the reason this section exists.
Name a pair, not a number. The TM-30 routes in both systems are two-part (Rf with Rg) or three-part (with Rcs,h1); a supplier quoting Rf alone satisfies neither. Rf 90 with Rg 85 is not a compliant TM-30 P1 claim; it is a different product behaviour — high fidelity with a saturated-colour deficit, which on a counter shows up as flat reds.
The CRI route has a red floor in one system and no red floor in the other. WELL v2's CRI route admits Ra ≥ 80 only with R9 ≥ 50. LEED v5's CRI route is written as "at least 90" with no R9 condition. A jewellery project that relies on LEED alone can therefore be specified with a high Ra and a poor R9. The red region is the gap that decides how yellow gold and rubies read (§D.1), and the WELL v2 route closes it by attaching R9 ≥ 50 to its lower Ra floor, while both TM-30 routes carry Rcs,h1 for the same region.
Δu′v′ is a distance on the CIE 1976 (u′,v′) diagram. It is used for three separate claims, and a specification that does not say which one is not a specification.
| Claim | What it fixes | Typical published limit |
|---|---|---|
| Initial chromaticity tolerance | How far a product's colour sits from the target colour at first measurement, and how far two units of the same model may sit from each other | Follows ANSI C78.377: ±0.006 from the target; the tier table's separate four-step quadrangle block is 0.0033 |
| Colour spatial uniformity | How much the colour varies across the beam angle of one luminaire | Within 0.007 of the weighted average point on the CIE 1976 (u′,v′) diagram, in the IEA 4E SSL Annex tier table |
| Colour maintenance | How far the colour moves as the product ages | Δu′v′ < 0.007 at 6,000 h for tiers 0 to 2 of that table, and < 0.004 at 6,000 h for the strictest tier |
The tier figures above are quoted from the IEA 4E SSL Annex performance tiers for downlight luminaires, which is a published international framework; the entry also notes that the measurement method is defined by CIE 15:2004 and that the target colour temperatures and their tolerances follow ANSI C78.377. We quote the framework, not the standards it cites — we have not read ANSI C78.377 or CIE 15:2004 for this article.
Colour maintenance is the claim that decides a jewellery case, and it is usually the missing one. A counter runs 10 to 12 hours a day at high output in a warm enclosure. The two failure modes are different: chromatic drift with drive level appears as soon as the system dims, because a two-channel tunable engine mixes two emitters and the mix point moves; chromatic drift with age accumulates over thousands of hours through LED and phosphor degradation, and it makes a replaced unit visibly different from the units beside it.
Neither is a CRI claim, and neither is detectable from a photometric file. Both have to be stated. Three published examples from the SUMLINK catalogue show the scopes side by side, and they are not the same scope:
Only the ageing scope is published on the BY001 page; the 20# and Linan pages name both scopes on the dimming-range figure.
All three are marked Design target on their pages, and that label carries a defined meaning on those pages: engineering design values from the driver and LED platform specification, not yet confirmed by an independent laboratory report. A jewellery project should ask for both statements, in one document, for the luminaire it is buying.
SDCM appears in every retail tender and does not measure the same thing as Δu′v′. An SDCM figure describes how tightly a production batch is binned around its target point — it is an initial, unit-to-unit statement. A colour-maintenance figure describes how one unit's colour moves over time. A product can be binned at ≤ 2 steps and still drift beyond that distance in two years, which is why both have to appear.
The published catalogue draws the distinction clearly, and the range across lines is wide enough to specify against:
| Published on | SDCM, initial | Δu′v′, ageing or dimming |
|---|---|---|
| BY001 series page | ≤ 3-step MacAdam ellipse, initial, ANSI C78.377 | ≤ 0.004 at 6,000 h, Ta 25 °C |
| 20# recessed magnetic track page | ≤ 1 (COB) / ≤ 3 (SMD) | < 0.003 over the full dimming range, at 6,000 h, Ta 25 °C |
| Linan 2.0 page | ≤ 2 / ≤ 1 depending on chip | < 0.002 over the full dimming range, at 6,000 h, Ta 25 °C |
All six values are marked Design target on their own pages. For a jewellery counter the practical requirement is the tighter end: accent modules that read the same as the module beside them, and that still read the same after the first lamp change. Note also what the 20# numbers imply for a two-tier design: the COB accent module and the SMD ambient module in the same system carry different binning, so a specification that quotes one SDCM for "the system" has quoted the wrong one for half of it.
The chart below is the live SUMLINK jewellery program page, captured 2026-10-15. It is the delivery layer this article sits under, and it is stated as a five-node day with counter and ambient levels separately named.
| Time | CCT | Illuminance | M/P ratio | Mode |
|---|---|---|---|---|
| 10:00 | 5000 K | 800 lx (counter) | 0.82 | Diamond Mode / Maximum Fire |
| 12:00 | 3500 K | 600 lx (counter) | 0.48 | Gold Mode / Warm Depth |
| 14:00 | 4000 K | 700 lx (counter) | 0.60 | Jade Mode / True Colour |
| 17:00 | 3000 K | 500 lx (counter) | 0.38 | VIP Preview / Intimate |
| 20:00 | 2700 K | 350 lx (ambient) | — | Closing / Night Colour Protection |
Four commitments in that table are worth isolating, because each one maps to a metric in the sections above.
Counter accent runs 600–800 lx against 250–350 lx of ambient, which the page states as a 3:1 contrast ratio. The contrast is the design decision; the absolute lux is secondary. That is the same logic Fagerhult's guide describes when it recommends narrow-beam accent over flat general light, and it is why the program's ambient layer is a separate CRI 90 line rather than the same accent module turned down.
The CCT arc runs 5000 K to 2700 K across the operating day, and the mode names follow the material. The programme hub states the intent for each mode: Diamond Mode at ultra-high CCT cool white for fire and brilliance, Jade / Gemstone Mode at a high-CRI neutral ~4000 K, and one-tap switching between Diamond, Gold and Jade modes on a counter accent module its Jewellery section publishes as Ra > 95 with R9 > 60 and a 2700 K–6500 K CCT range. The jewellery program page linked above publishes CRI 97+ for that same counter accent module; the hub carries CRI 97+ only for its office-path accent module, and we record the difference in §K.3. That is the same material-divided logic Fagerhult's guidance reaches from the other direction: warm light for gold, neutral or cool for silver, white gold and crystal.
The reported M/P ratio falls with the CCT arc, from 0.82 at 10:00 to below 0.35 in the evening. The M/P ratio is a melanopic-to-photopic ratio; it is the metric this series uses for the non-visual side of a schedule, and it is not a colour-quality metric. It appears here because the program publishes it per node, and because the closing node uses it as the stated basis for colour protection.
One claim in that row is a published claim, not a verified one. The page states that a closing M/P ratio below 0.40 "prevents blue-light colour shift on gemstones left on display". M/P is a ratio of melanopic to photopic output; it is not a measure of a stone's photostability, and we hold no record of a stone having been measured before and after a display season under this program. We report the claim as the page states it and mark it 未驗證 in §K.3. What the closing node does do, and what can be checked, is reduce illuminance to 350 lx and shift the spectrum to 2700 K — a lower dose and a lower proportion of short-wavelength output than the 10:00 node.
The programme hub also states one number that differs from the program page: it describes Diamond Mode as ultra-high CCT cool white (~6000K+), while the program page's Diamond node is 5000 K and its header reads "Morning Light 5000K". We flag the difference in §K.3 rather than reconcile it, because the operating point is the thing a specification has to name, and only the project can settle which node it wants.
Copy the following into a jewellery lighting specification or a purchase order. Each clause closes a gap that the metrics above leave open. Clause 1 and clause 4 are the two that matter most, because they are the two that most quotes omit.
Clause 4 is the one that changes outcomes at the second year. A display case that was specified with a bin figure alone has documented the day of installation and nothing after it. The fields these figures sit in — colour shift Δu′v′, SDCM, Rf and Rg beside the standard photometric columns — are the ones the HCL specification library leads with, and they are the fields to copy into the request.
This section states our own position, including the parts that are not flattering.
What is published without asking. The catalogue index leads with the human-factor fields — M/P ratio, flicker, SVM, Δu′v′ — beside the standard photometric columns, and the about page states the catalogue standard as tunable white at CRI 90 with R9 ≥ 50. The jewellery program page publishes its five-node table, its mode names, and the CRI 97+ and R9 > 60 figures for the counter accent module. The BY001 series page publishes Δu′v′ ≤ 0.004 at 6,000 h and Ta 25 °C, SDCM ≤ 3-step, TM-30-20 Rf 88 / Rg 99, CRI Ra 90, UGR < 19, SVM < 0.15 and percent flicker < 1%.
What the labels say. Every one of those figures on the series pages is marked Design target, and the pages define the term: values marked Design target are engineering design values from the driver and LED platform specification, not yet confirmed by an independent laboratory report; Verified values are backed by a test report available on request. Read against §D.8, that places the whole colour-metric block — Rf, Rg, Δu′v′, SDCM — in the design-target category today.
What is unlabelled, and therefore not evidence. Two figures on live pages carry no label. The about page's metric strip reads "Colour Fidelity CRI ≥97 · SDCM <2", and the catalogue entry for the 35# surface-mounted track describes it as directing "CRI 97+ spectra at any surface". Neither figure is marked Design target or verified, and neither is attached to a named model with the measurement condition stated. They sit above the catalogue standard (CRI 90, R9 ≥ 50) and describe the platform's best-performing lines rather than the standard. A specification should not adopt them as written; it should ask for the figure with its label, its model and its measurement condition.
What has to be requested. The Downloads rows for datasheets, IES / LDT photometric files, test reports and compliance documents are marked "available on request" on the product pages, while the about page states that full IES files are downloadable without registration. Both statements are live and we have not resolved which governs; ask early and expect to ask. For a jewellery project the request list is short and specific: per-mode colour metrics (clause 1), spectral data (clause 6), the two colour-shift statements (clause 4), per-module binning (clause 5), and the handover measurement method (clause 8).
Four limits, stated so that the article cannot be read as more than it is.
No independent laboratory report is attached to any figure quoted here. Every SUMLINK number above is either a Design target or an unlabelled catalogue figure, as each subsection says. The colour metrics we quote are engineering values from platform specifications.
No project has been measured against this specification. We hold no record of a jewellery installation in which per-node Δu′v′, TM-30 values or spectral data were measured and accepted. The clauses in §D.8 are a proposed procurement requirement, not a report of a completed verification.
The gemmological sources are quoted for what they say, not as lighting standards. GIA's grading practice, ISO 24016's grading conditions as SSEF describes them, and the SSEF and IGS statements about spectral continuity are gemmological sources. None of them sets a display-lighting requirement, and none of them was written about retail lighting.
The mode table is a program, not a measurement. The five nodes, their CCTs and their illuminances are the published program. Whether a specific installation holds those values depends on the luminaires, the driver, the control gear and the commissioning — which is what clause 8 exists to check.
If a supplier cannot state, per mode, the colour metrics, the spectral basis and the two colour-shift figures, the project is buying luminaires with good CRI. That is a different purchase from a display strategy whose colour can be checked at handover and again in year two.
No. CRI 97 is an average of eight colour samples, and R9 — the ninth sample, which decides how gold, rubies and warm-toned gemstones read — is outside that average. CIE now recommends adopting the General Colour Fidelity Index Rf and reporting Rf alongside Ra during the transition, and both WELL v2 and LEED v5 accept a TM-30 route whose thresholds are Rf ≥ 78 with Rg ≥ 95. Specify CRI 97 together with R9, Rf, Rg and Rcs,h1, and state the edition of TM-30 the figures were calculated against.
Rf is average colour fidelity: closer to 100 means the 99 evaluation samples look closer to their reference appearance. Rg is gamut: above 100 means the light increases chroma relative to the reference, below 100 means it reduces it, so an Rg of 99 is essentially neutral. Rcs,h1 is the local chroma shift in the red hue region, which is the number that matters for gold and rubies; the TM-30 tutorial's own guidance is that Rf, Rg, Rcs,h1 and Rf,h1 together predict preference, vividness and fidelity in most architectural applications. A published pair of Rf 88 with Rg 99 is high fidelity with no chroma boost.
Δu′v′ is the distance between two chromaticity points on the CIE 1976 (u′,v′) diagram. It is used for three different claims: initial chromaticity tolerance, which follows ANSI C78.377 at ±0.006 and 0.0033 for the four-step quadrangle; colour spatial uniformity, within 0.007 across the beam; and colour maintenance, which is the one that decides a display case. The IEA 4E SSL Annex downlight tier table requires Δu′v′ below 0.004 at 6,000 h for its strictest tier. Require both an ageing figure at 6,000 h and a dimming-range figure, at a stated ambient temperature, with the projection basis named.
Two reasons, and only one of them is about the customer. Colour temperature changes how the metal and stone read: Fagerhult's jewellery guidance recommends roughly 3000 K for gold and natural white or 4000 K and above for silver, white gold, chrome and crystal, so a single fixed CCT cannot serve both well. The second reason is exposure: the published SUMLINK jewellery program closes at 20:00 with 2700 K at 350 lx instead of the 10:00 node's 5000 K at 800 lx, which lowers both the illuminance and the proportion of short-wavelength output. The program page further states that a closing M/P ratio below 0.40 protects displayed gemstones from colour shift; we report that as the page's claim, since M/P measures melanopic output rather than a stone's photostability.
Five, and they are all short. Per-mode colour metrics — CRI Ra, R9, and TM-30 Rf, Rg, Rcs,h1 with the edition named. Colour data at 1,000 K intervals across the tuning range, not only at the endpoints. Two colour-shift statements: Δu′v′ at 6,000 h with the ambient temperature and projection basis, and Δu′v′ across the full dimming range. Binning in MacAdam steps (SDCM) for each module type supplied, since accent and ambient modules are often binned differently. And the spectral power distribution per light engine, with the CCT and drive condition it was measured at — because spectral continuity is what a gemstone sees, while indices only summarise it.
The jewellery program page carries the five-node time-spectrum table, the Diamond / Gold / Jade / VIP modes, the counter-to-ambient ratio, and the pre-configured DALI-2 or Casambi scene file. Tell us which stones and metals the counter carries and we will issue the colour-metric and colour-shift request list against the clauses in the article above. Deploy This Strategy →