How GemID Works

The approach: property-based matching

GemID identifies gemstones the same way a trained gemologist does: by measuring physical and optical properties, then eliminating every species whose documented properties are inconsistent with those measurements.

This is deliberate. A photograph captures appearance — color, shape, surface reflection — which varies with lighting, cut quality, camera settings, and surface condition. Refractive index (RI), specific gravity (SG), optic character, and fluorescence are intrinsic physical constants of the crystal structure. They do not change between photographs.

GemID does not use machine learning or image recognition. Every elimination decision is deterministic and explainable: the app shows exactly which property caused a gem to be removed from consideration, and what the documented value is.

The reference database

The engine matches against a database of 130 gem species and varieties, including natural stones, synthetics, and common simulants. Each entry stores up to 25 properties per gem.

Properties stored per gem

• Optical constants: RI minimum and maximum, birefringence, optic character (SR / DR uniaxial± / DR biaxial± / aggregate), anomalous double refraction (ADR), dispersion
• Density: SG minimum, maximum, and typical value
• Visual: color categories, transparency levels, luster types
• Phenomena: asterism (+ ray count), chatoyancy (+ sharpness and milk-and-honey effect), adularescence, labradorescence, play-of-color, aventurescence
• Instrument responses: LW fluorescence (365 nm), SW fluorescence (254 nm), Chelsea filter response, Hanneman filter response, thermal conductivity category, spectroscope absorption patterns, dichroscope pleochroism type
• Physical: Mohs hardness range, cleavage quality, fracture type, crystal habits, magnetism level, facet doubling presence and strength
• Inclusions: loupe-observable types (10×), darkfield microscope types
• Treatment and origin: known treatments (heating, fracture-filling, dyeing, beryllium diffusion, etc.) and natural vs. synthetic determination protocols
• Simulants: documented imitants with detection notes

Data sources

Property values are compiled from primary gemological references: Gems by Webster & Read (6th ed.), Gemology by Hurlbut & Kammerling, GIA course materials, Gübelin & Koivula's Photoatlas of Inclusions in Gemstones, and peer-reviewed publications from Gems & Gemology. Where published ranges conflict, the broader range is used to avoid false eliminations.

The identification filter chain

When you enter an observation, the engine runs 26 independent matchers against every gem in the database simultaneously. A gem must pass all applicable matchers to remain a candidate. Matchers are evaluated in this order — visual first, instruments later — but all 26 run on every evaluation cycle:

Short-circuit evaluation

For candidate gems (those still in contention), the engine stops at the first failing matcher — there is no need to evaluate the rest. This is fast and produces a clean primary reason for elimination ("RI 1.543 outside 1.762–1.770 ± 0.020"). For eliminated gems, all 26 matchers run to produce a fail count, which is used to sort the eliminated list — gems that failed only one or two filters appear first as "closest misses," which is useful when a reading may be imprecise.

Measurement tolerances

No instrument is perfectly precise, and no two stones of the same species have identical properties. GemID applies published tolerances that reflect real-world instrument accuracy:

Why ±0.020 for RI

The GIA Gem Identification Lab Manual states that standard gemological refractometers are accurate to approximately ±0.003 under ideal conditions, but shadow-edge estimation, contact liquid variation, temperature, and reading angle combine to produce typical field uncertainty of ±0.003 to ±0.010 in practice. The ±0.020 tolerance extends conservatively beyond that to account for instruments at the edge of calibration and stones with surface irregularities that prevent a clean reading. The tradeoff is that species with closely overlapping RI ranges (e.g., aquamarine and blue topaz) may both remain candidates until SG or optic character separates them — which is the correct behavior.

Why ±0.10 for SG

Hydrostatic weighing accuracy depends on scale resolution, wire thickness, surface tension of water, and whether the stone is mounted. Published species ranges themselves span ±0.05 to ±0.20 for most gems due to natural compositional variation (e.g., garnet solid solutions). The ±0.10 tolerance reflects that SG is a useful filter when combined with RI, not a definitive standalone identifier.

Over-the-limit RI readings

Most gemological refractometers have an upper limit of approximately 1.81. When you enter an OTL (over-the-limit) reading, the engine retains only gems whose documented minimum RI exceeds 1.81 — primarily demantoid garnet (1.880–1.889), high-type zircon (1.810–1.984), cassiterite (1.997–2.093), and synthetic YAG (1.834). The other 120+ species are eliminated immediately.

Conservative matching: when in doubt, don't eliminate

Every matcher follows one rule: if the user has not entered a value for a property, that property does not eliminate any gem. A nil observation is not the same as a negative observation.

Several matchers extend this conservatism further:

• Fluorescence "variable": Gems with highly variable fluorescence (e.g., opal, pearl) are never eliminated by a fluorescence observation — their documented response genuinely covers the full range.
• Loupe inclusions "clean": Most gems can appear loupe-clean; entering "clean" does not eliminate gems that document inclusions. Only a positive identification of a specific inclusion type triggers filtering.
• Spectroscope "none visible": Failure to see an absorption band may reflect instrument limitations or observer skill. This observation does not eliminate gems that document spectroscope bands.
• Crystal habit — empty gem list: If a gem's crystal habit field is not populated in the database, that gem always passes the crystal habit matcher. The absence of documented data is not evidence of absence.
• Magnetism, Chelsea, Hanneman — gem marked "not applicable": Where a property is not a meaningful test for a given gem, it is flagged not-applicable and always passes regardless of observation.

Test recommendation and ranking

When more than one candidate remains, GemID recommends the next most useful test. The recommendation is not a fixed sequence — it is computed dynamically from the current candidate set using a scoring formula.

The test with the highest score is recommended first. Tests with zero discrimination (i.e., all remaining candidates would give the same result) are never recommended — they add no information.

Discrimination

Discrimination measures how well a test would split the current candidate set if performed. Three formulas are used depending on whether the test produces boolean, categorical, or continuous results:

• Boolean tests (e.g., shows color-change or not): 1 − (largerGroup / total). Ranges from 0 (all candidates agree) to 0.5 (perfect split).
• Categorical tests (e.g., Chelsea filter: red / inert / green): 1 − (largestCategory / total). Approaches 1 when candidates are evenly spread across all possible outcomes.
• Range tests (RI, SG, hardness): spread / (spread + avgWidth), where spread is the range of candidate midpoints and avgWidth is the average individual species range. Approaches 1 when candidates have very different values and tight individual ranges — meaning a precise reading would cleanly separate them.

Difficulty

Difficulty scores the practical barrier to performing each test — equipment required, skill, time, and risk to the specimen. Lower difficulty means the test is recommended more aggressively relative to its discrimination:

Cabochon adjustment

When the stone form is set to cabochon, optical phenomenon tests receive a difficulty bonus (halved to 0.5) because phenomena are the primary diagnostic for cabochon stones — they should rank above color and transparency in that context.

Set stone adjustment

When the stone is set (mounted in jewelry), RI and SG tests are removed from recommendations entirely — both require access to a flat facet or the ability to weigh the stone in water, which is not possible in a setting.

Natural vs. synthetic and treatment protocols

Once a gem species is identified, many professional workflows require determining origin (natural vs. lab-grown) and treatment status (heated, fracture-filled, beryllium-diffused, etc.). GemID includes interactive two-phase protocols for the gem species where these determinations are feasible in the field.

Why physical properties usually can't separate natural from synthetic

Lab-grown gemstones are grown from the same mineral compound as natural stones. The RI, SG, optic character, and fluorescence of synthetic ruby are essentially identical to natural ruby — they are the same crystal, grown in a different environment. Physical property filters cannot make this distinction.

What differs is growth features — the microscopic record of how the crystal grew. Flame-fusion synthetic corundum shows curved growth lines and gas bubble inclusions. Hydrothermal synthetic emerald shows chevron growth patterns and characteristic two-phase inclusions. Natural rubies may show silk (rutile needles), fingerprint inclusions, and color zoning consistent with metamorphic or magmatic growth. These require magnification and informed examination.

How the protocols work

Each protocol is a branching decision tree: each step presents a test to perform, the observations to look for, and a branching path based on what you see. The sequence is designed so that the most accessible and diagnostic test comes first. Steps terminate in a conclusion: Natural, Synthetic (with growth method), Treatment Present, Treatment Not Detected, or Inconclusive.

Conclusions are described as "consistent with" a specific origin or treatment status — not certified. The protocols follow the same methodological framework used in gemological laboratory practice; they do not replace laboratory-grade analysis with advanced instrumentation (photoluminescence spectroscopy, UV-Vis spectrophotometry, EDXRF).

Result language and epistemic honesty

GemID uses specific language throughout that reflects the actual confidence level of a field identification:

• "Candidates include" — multiple gems remain consistent with the entered observations. More testing is needed.
• "All inputs are consistent with [gem]" — a single candidate remains after all applied filters. This is the strongest conclusion the engine produces.
• "Consistent with" — the measured properties do not exclude this identification. This is not a certification.
• "Indicators observed" — a specific observation (inclusion, fluorescence pattern) is consistent with a known treatment or origin indicator, but is not definitive in isolation.

The engine never states that a stone is a particular gem — only that it is consistent with one given the measurements taken. This is the correct framing for field identification in any system, including manual gemological assessment.

What GemID cannot determine

GemID is a field identification tool. There are categories of determination that are outside the scope of instrument-based field testing:

• Origin by geographic locality — Kashmir sapphire vs. Sri Lankan sapphire, Colombian vs. Zambian emerald. Locality determination requires trace element analysis (EDXRF, LA-ICP-MS) and is beyond field instruments.
• Subtle or advanced treatments — beryllium lattice diffusion in corundum, lead glass fracture filling in ruby, HPHT color treatment in diamond. Some indicators are visible in the field; definitive determination often requires laboratory analysis.
• Natural vs. synthetic when inclusions are absent — A loupe-clean synthetic stone grown by a process that does not leave characteristic inclusions cannot be distinguished from a natural stone by field methods alone.
• Appraisal or valuation — Market price guidance in GemID is approximate context based on published ranges, not a USPAP-compliant appraisal. GemID is not an appraisal tool.
• Certification — A GemID field identification is not equivalent to a laboratory report from GIA, AGL, Gübelin, SSEF, or other accredited laboratories. Those reports involve additional analytical instruments, professional certification, and chain-of-custody documentation that GemID does not and cannot replicate.