Scoring methodology · v1.0
Methodology
How Plassic computes a score, in detail. The five signal classes, their weights, data sources, and the limits of what the score can — and cannot — tell you.
Score formula
Weighted average of five signal classes
Every Plassic score is a weighted average — 0 to 100 — calculated across five independent signal classes. Higher score = lower exposure risk. The weights reflect both the evidence strength in the research literature and the practical data availability for each class at scale.
Score formula
Score = (S₁ × 0.35) + (S₂ × 0.25) + (S₃ × 0.20) + (S₄ × 0.10) + (S₅ × 0.10)
Each Sₙ is normalised 0–100 within its class before weighting. Where a class signal is absent, remaining weights are rebalanced proportionally.
| Signal class | Weight | Inputs | Source |
|---|---|---|---|
| Direct microplastic content | 35% | Microbeads, glitter particles, polyethylene (PE), polypropylene (PP), PEG derivatives, nylon-12, acrylates copolymer | INCI ingredient database · EPA Contaminant Candidate List · REACH substance registry |
| Material composition | 25% | % polyester / nylon / acrylic fibre content; PET vs glass vs stainless packaging; synthetic vs natural blend ratio | Brand material disclosures · Plassic material composition database |
| Shedding / leaching risk | 20% | Microfibre shed count per wash cycle; food-contact leaching under heat / acid / fat conditions; surface abrasion factor | Peer-reviewed research — PubMed / PMC corpus (2018–2026) |
| Manufacturing process | 10% | Single-use vs durable format; recyclability of packaging; plastic-type coding; refillable programme availability | ESG / sustainability reports · brand disclosures · GS1 product data |
| Independent testing | 10% | Published lab-measured microplastic counts or absence-of-detection in peer-reviewed or accredited third-party studies | Research corpus — Orb Media, Pew Charitable Trusts, University of Newcastle, EU JRC |
| Total | 100% | ||
Signal classes in depth
What each class measures
Direct microplastic content · 35%
The highest-weight class. We parse INCI ingredient lists and material declarations against a curated blocklist of confirmed plastic-derived ingredients: polyethylene (PE) microbeads, polypropylene (PP) granules, nylon-12, nylon-6, acrylates copolymer, PEG derivatives, and glitter formulations. Presence of any blocklisted ingredient in a functional (non-trace) concentration triggers a proportional score penalty.
Sources: INCI nomenclature database · EU REACH Annex XIV/XVII · EPA Safer Choice ingredient list
Material composition · 25%
For textiles: we parse fibre composition labels — 100% polyester scores differently from a 30% polyester / 70% GOTS-certified cotton blend. For food-contact products: we classify by material type (borosilicate glass, stainless 18/10, polypropylene, PET, melamine). Material-type risk tiers are derived from the peer-reviewed migration literature and EU Food Contact Material regulations.
Sources: Brand fibre disclosures · GS1 material attribute database · EU FCM Regulation 10/2011
Shedding / leaching risk · 20%
Separate from what a product is made of: how much of it ends up in you or the environment during normal use. For textiles: shed particle counts per wash cycle from published laundering studies (De Falco et al. 2019; Napper & Thompson 2016; Cesa et al. 2017). For food-contact materials: leaching rates under representative conditions — temperature, pH, fat content, contact duration — from migration studies (Guart et al.; Grob et al.). Where no specific product data exists, we apply category-level median shedding/leaching estimates as a conservative prior.
Sources: PubMed / PMC research corpus (2015–2026) · EU JRC technical reports · EFSA food contact opinions
Manufacturing process · 10%
Downstream-plastic-generation signals: single-use vs. durable packaging format; plastic-type coding and recyclability; whether the brand operates a certified take-back or refill programme. This class does not penalise a product for supply-chain plastic it cannot control — it penalises choices the manufacturer makes about the product's own physical form.
Sources: Brand ESG / sustainability reports · How2Recycle label data · GS1 packaging attributes
Independent testing · 10%
The rarest and most powerful signal: a published, peer-reviewed or accredited third-party lab result that directly tested this product or SKU. Confirmed detection raises risk; confirmed absence lowers it — provided the study methodology meets our inclusion criteria (sample size ≥ 3, blank controls reported, particle characterisation by FTIR or Raman spectroscopy). Studies failing our inclusion criteria are logged but excluded from scoring.
Sources: Orb Media database · Pew Charitable Trusts · University of Newcastle (Tyree & Morrison 2018) · EU JRC · Plassic partner-lab results (Phase 2)
Score bands
Four bands, three colours rearranged
The score (0–100) maps to one of four bands. The bands use only three brand colours — rearranged — so the app stays visually coherent rather than defaulting to a red/yellow/green stoplight.
80 – 100
Clean
Minimal detectable microplastic risk across all five signal classes. Often glass, stainless, certified-natural, or independently tested.
55 – 79
Low
Low aggregate risk. One or two signal classes flag minor concerns — often packaging or material composition — but no direct microplastic inputs detected.
30 – 54
Medium
Moderate risk. Multiple signal classes contribute: synthetic fibre content, known leaching conditions, or a manufacturing process relying heavily on single-use plastic.
0 – 29
High
Elevated risk across the majority of signal classes. Likely includes confirmed microplastic ingredients or materials with a high shedding / leaching profile.
Colour is never the sole signal — the verdict word and score number are always present alongside the band colour (WCAG 2.2 AA compliant).
Normalisation
How each class is scaled to 0–100
Ingredient-based classes (S₁, S₂)
Each detected ingredient carries a risk weight drawn from the regulatory and research literature. Weights are summed and normalised against the maximum theoretical burden for that product category. A product with no flagged ingredients in a class receives 100 in that class; a product fully composed of high-risk ingredients approaches 0.
Research-derived classes (S₃, S₅)
Continuous measurements (shed particle counts, migration rates) are normalised against the empirical distribution observed in Plassic's research corpus for that category, then inverted so that low shed / low leach = high score. Discrete test results (presence / absence) are mapped to fixed score deltas (±12 points per confirmed finding).
Process class (S₄)
A checklist of up to 8 binary and ordinal attributes — packaging type, recyclability, refill programme, etc. Each attribute contributes an equal share of the 100-point ceiling. Unanswered attributes are treated as neutral (50) rather than penalised, because the absence of a disclosure is not equivalent to confirmed poor practice.
Missing signals
When a class has no usable signal for a product, that class is excluded and the remaining weights are scaled to sum to 100%. Scores computed on fewer than three classes are flagged in the API as coverage: "sparse" and displayed with a reduced-confidence indicator in the app.
Limitations
What the score cannot tell you
Intellectual honesty is load-bearing for Plassic. The limitations below are not disclaimers buried in a terms page — they are part of the product.
Scores are inference, not lab measurement
Plassic does not test physical samples. Scores are computed from ingredient databases, material disclosures, and published research. A product scoring 90 may still shed fibres; a product scoring 30 is not necessarily 'safe'. The score quantifies documented risk signals — nothing more.
Database coverage varies by category
Food-contact plastics and cosmetics have the densest signal coverage (INCI and EPA ingredient data are mature). Clothing and kitchenware are expanding; scores for those categories carry a wider confidence interval. Each score record includes a coverage_confidence field — visible in the API; surfaced in the app as a data-quality indicator.
Brand reformulations may lag
Formulas change. Plassic triggers a re-score when a barcode re-scan returns a different INCI or material declaration, or when a published study references a product in our corpus. Between those events, a score reflects the most recent verified data. The score_updated_at timestamp is displayed on every result.
Scores are versioned
Scoring weights and signal-class definitions evolve as the research corpus grows. Each score record carries a score_version identifier. When a version bump changes a product's score materially (≥ 5 points), affected users receive an in-app notification. Historical scores are preserved and accessible via the API.
Independent-testing signal is sparse
Only a small fraction of the product corpus has appeared in published lab studies. When the independent-testing signal is absent, the remaining four classes are reweighted proportionally. A product that has been independently tested — in either direction — receives a more reliable score.
Data currency
How we keep scores current
Trigger 01
Barcode re-scan
When a user scans a barcode, Plassic checks whether the INCI or material declaration has changed since the last score was computed. Any change triggers a re-score.
Trigger 02
Research corpus update
When a new study enters the research corpus that references a product in our database, affected scores are queued for re-evaluation within 72 hours.
Trigger 03
Weight-version bump
When scoring weights change (a versioned update), all affected products are re-scored in bulk. Users whose product scores change by ≥ 5 points receive an in-app notification.
More reading
Need the plain-English version?
/how-it-works explains the same methodology without the formula notation — written for people who want to understand, not audit.