Technical White Paper · v1.2 · May 2026

The PackIndex Methodology:
Proprietary Data, Contributor Intelligence & ML‑Powered Prediction

How the PackIndex three-layer cascade, global contributor network, machine learning prediction engine, and AI analytical layer combine to deliver the most accurate real-time packaging market intelligence available.

Issued byOpen Packaging Network / Polimex Trade Inc.
CoverageL1 Raw Materials · L2 Semi-Finished · L3 Finished Units
Update CadenceDaily · Weekly · Monthly (by tier)
Geographic ScopeNorth America · Europe · Asia-Pacific
Abstract

PackIndex combines four mutually reinforcing systems to produce packaging market intelligence unavailable anywhere else. A proprietary three-layer price cascade ingests and normalises over 140 raw material price series across global markets. A structured global contributor network — packaging buyers, procurement leads, and category managers across North America, Europe, and Asia-Pacific — feeds the Pulse Diffusion Index with real-time market sentiment that quantitative models alone cannot observe. A purpose-built machine learning prediction engine, trained on the full PackIndex dataset and continuously retrained on new data, achieves the highest forward-price accuracy of any packaging benchmark. And an AI intelligence layer — PackIndex Advisor and PACKIQ’s 100+ specialist digital experts — translates this data infrastructure into actionable procurement intelligence. This paper describes all four systems, their interactions, and the governance framework that keeps them accurate, independent, and commercially trusted.

Contents
Section 01

Why Packaging Needs a Purpose-Built Intelligence System


Packaging procurement sits at the intersection of commodity markets — fibre, aluminium, polymers, energy — yet no single intelligence platform has historically served it with the rigour applied to financial or energy markets. Buyers have relied on trade publication surveys (slow, subscription-gated, survey-biased), supplier-issued price notices (self-serving, lagged), or internal cost models built on stale reference points.

The fundamental problem is structural: packaging cost is a derived quantity that varies by region, specification, and market cycle. A corrugated case price is not a commodity — it is a function of recovered fibre, starch, energy, liner board conversion, and plant utilisation, across multiple geographies simultaneously. No legacy platform tracks this end-to-end. PackIndex was designed from the ground up to solve this — combining proprietary data collection, a global contributor network, machine learning prediction, and AI-powered analysis into a single integrated system.

Section 02

The Three-Layer Cascade Architecture


All PackIndex data is organised into three mutually dependent layers. Changes in Layer 1 propagate through Layer 2 into Layer 3 on each update cycle — ensuring finished unit indices always reflect current input economics, not last quarter’s survey responses, across every market we cover.

L1 · Layer 1
Raw Materials
  • Virgin & recovered fibre (global)
  • Aluminium ingot & sheet (LME/SHFE)
  • Polyolefins & specialty polymers
  • Natural gas & grid electricity
  • Starch, coatings & adhesives
  • Bioplastics (PLA, PHA, PBAT)
L2 · Layer 2
Semi-Finished Grades
  • Kraftliner / Testliner board
  • Corrugating fluting medium
  • Folding boxboard (FBB / SBB)
  • BOPP, BOPET & CPP films
  • Metallised & barrier films
  • White-lined chipboard (WLC)
L3 · Layer 3
Finished Unit Indices
  • RSC corrugated cases (A–C flute)
  • Stand-up & flow-wrap pouches
  • PET & HDPE bottles
  • Rigid & luxury folding cartons
  • Labels, sleeves & mailers
  • Pallets, caps & dispensers
Pulse Contributor Overlay — Anonymised global contributor sentiment (weekly Pulse Diffusion Index, −100 to +100) applied at L3, blended with the quantitative model at a proprietary weighting reviewed quarterly.
ML Prediction Engine — Continuously retrained machine learning models generate forward price estimates across all three layers, with confidence intervals updated weekly on new data.
Section 03

Layer 1: Global Raw Material Coverage


PackIndex maintains a proprietary monitoring network covering over 140 distinct raw material price series across the global packaging supply chain. All data is continuously normalised and ingested into the PackIndex calculation engine, with each series assigned an update frequency, geographic scope, and confidence tier. Prices are expressed in the user’s selected currency, converted at daily reference exchange rates from the Fed and ECB. The index base is January 2025 = 100, with validated historical records extending to January 2023.

Americas
North America
  • US OCC #11 recovered fibre
  • RISI corrugated board grades
  • Henry Hub natural gas
  • Domestic polymer spot (OPIS)
  • US diesel & energy indices
  • USDA starch & ag inputs
EMEA
Europe
  • EU OCC, ONP recovered grades
  • NBSK / BHKP pulp grades
  • TTF natural gas (Amsterdam)
  • LME metals (aluminium, tin)
  • EU polymer spot markets
  • Eurostat energy & labour
Asia-Pacific
APAC
  • Asian OCC & mixed paper
  • China polymer production indices
  • Asian aluminium premium (SHFE)
  • Regional energy benchmarks
  • APAC bioplastic grades
  • Trans-Pacific freight indices
CategoryKey SeriesUpdate Freq.Geo ScopeConfidence
Virgin FibreNBSK, BHKP, SBSK pulp gradesWeeklyGlobalHigh
Recovered FibreOCC (US #11 & EU), ONP, Mixed PaperWeeklyUS / EUHigh
MetalsLME Aluminium, Tin, Copper, SHFEDailyGlobalHigh
PolymersLDPE, LLDPE, HDPE, PP, PET, PLA, EPSWeeklyUS / EU spotHigh
Specialty PolymersABS, HIPS, GPPS, PC, PA/Nylon, EVAWeeklyGlobalMedium
EnergyHenry Hub, TTF Gas, EPEX, DieselDailyUS / EUHigh
Process InputsStarch (USDA), Caustic Soda, CoatingsMonthlyGlobalMedium
BioplasticsPLA, PHA, PBAT gradesWeeklyGlobalMedium
Source attribution policy All prices published on the PackIndex platform are PackIndex data. Market inputs across all geographies feed exclusively into the PackIndex calculation engine. Users interact with PackIndex indices — normalised, validated, and expressed in their chosen currency — ensuring analytical consistency across regions, time periods, and market cycles.
Section 04

Layer 2: Semi-Finished Grade Derivation


Layer 2 indices are derived from Layer 1 raw material prices using proprietary bill-of-materials conversion weights — the material inputs consumed per unit of semi-finished output. These coefficients are based on industry-standard conversion benchmarks and reviewed annually, with interim revisions triggered when Pulse contributor signals indicate significant structural divergence from the model.

Layer 2 currently covers 14 board grades — including Kraftliner, Testliner, Corrugating Fluting, FBB, SBB, WLC, Greyboard, and speciality variants — and 9 film grades including BOPP, BOPET, CPP, Metallised PET, and EVOH. Each grade carries its own geographic price variant where market data supports regional differentiation. New grades are added quarterly as contributor coverage and market relevance justify the addition.

The derivation methodology ensures that when a major pulp market moves — whether in North America, Scandinavia, or Southeast Asia — the effect propagates through to the relevant board and film grades within the same weekly update cycle, without manual editorial intervention.

Section 05

Layer 3: Finished Unit Index Construction


Layer 3 is the headline output of PackIndex — the cost to produce a standard packaging unit at a moderately efficient converter, for medium production volumes, in the relevant market. Each L3 index is defined by a canonical specification: a fixed format, material grade, weight, print complexity, and volume assumption, published alongside every index to support direct comparability against individual SKUs.

Finished unit indices are constructed by applying material weight coefficients to the relevant L2 grades, adding conversion costs (energy, labour, and overhead), and applying the Pulse adjustment where sufficient contributor data exists. The multi-input cascade creates natural smoothing — no single commodity reading can produce an anomalous finished unit output. Conversion costs are updated monthly using US BLS, ONS (UK), and Eurostat labour and energy indices.

Regional variants of L3 indices reflect the different input cost structures of US, EU, and APAC converters, allowing buyers to benchmark against the market most relevant to their supply base.

Data integrity controls Every L3 index passes through automated anomaly detection on each update cycle. Values outside defined tolerance bands are held for editorial review before publication, ensuring the data reaching commercial users is clean, consistent, and auditable.
Section 06

The Global Contributor Network & Pulse Diffusion Index


The PackIndex Contributor Network is one of the platform’s most strategically significant assets — and the primary reason the Pulse Diffusion Index captures market turning points weeks ahead of published trade data. Contributors are active packaging procurement professionals: buyers, category managers, and supply chain leads who participate in the Pulse survey in exchange for platform credits and early access to PackIndex intelligence.

The network is structured to ensure signal quality, not just signal volume. Contributors are verified by professional role before onboarding. Each submission is weighted by the contributor’s declared category expertise, submission consistency score, and historical accuracy against subsequent published index moves. A category specialist with a strong track record carries more weight in the PDI than a new or inconsistent respondent — creating a meritocratic, self-improving signal engine.

140+
Raw material price series monitored globally
3
Major market regions with dedicated contributor cohorts
28d
Rolling aggregation window for the PDI
1wk
Pulse update cadence — fastest packaging sentiment signal available
Contributor Onboarding
Verified Professionals Only
Contributors are onboarded through a role-verification process. Packaging buyers, procurement leads, and category managers confirm their active market participation before their signals enter the PDI calculation. This structural gate prevents noise from non-market participants from contaminating the index.
Signal Weighting
Accuracy-Weighted Aggregation
Each contributor carries a dynamic weight reflecting their category expertise, submission consistency, and historical signal accuracy. Contributors whose past signals have accurately predicted subsequent index moves carry disproportionate influence — the model updates monthly, compounding accuracy over time.
The Pulse Diffusion Index
−100 to +100 Weekly Signal
The PDI aggregates directional signals on price, lead time, and supplier sentiment across packaging categories. At +100, all contributors signal rising prices and tightening supply. At −100, the reverse. The PDI consistently leads published commodity moves — contributors see and feel market dynamics before they register in formal price publications.
Contributor Incentive
Credits, Intelligence & Access
Contributors earn PackIndex platform credits for each verified Pulse submission, redeemable for PACKIQ agent runs, report access, and premium index history. This incentive structure creates a sustained, growing contributor base with strong participation rates — the network compounds in value as it grows.
📡
Why the contributor network matters Published commodity prices reflect what the market paid. The Pulse network reflects what the market expects to pay next. This distinction — between lagging and leading intelligence — is the single most commercially valuable feature of PackIndex for procurement teams operating in volatile markets.
Section 07

The ML Prediction Engine


PackIndex has built a purpose-designed machine learning prediction engine to achieve the highest forward-price accuracy possible for packaging commodity indices. The engine is trained on the full PackIndex historical dataset — spanning all L1, L2, and L3 series across all geographies — and is continuously retrained as new data enters the system. This ensures models remain calibrated to current market dynamics rather than drifting on static historical patterns.

The prediction engine is not a single model. It is an ensemble architecture in which multiple model families are fitted independently, their outputs compared against held-out validation data, and their predictions blended using a weighting scheme that favours whichever model family is currently performing best for a given series. This adaptive blending is the core innovation that allows the engine to outperform any single modelling approach across the diversity of packaging commodity series.

📈
Time-Series Decomposition & Seasonal Fitting
Each price series is decomposed into trend, seasonal, and residual components. Seasonal factors are fitted independently per commodity category — corrugated inputs follow agricultural cycles distinct from polymer or metal series — allowing the engine to model category-specific seasonality rather than applying generic assumptions.
🧠
Gradient Boosting & Ensemble Regression
Gradient-boosted tree models are trained on feature sets combining lagged price history, cross-commodity correlation signals, energy price inputs, and Pulse contributor data. Ensemble outputs are validated weekly against held-out recent data before being published as forward estimates, ensuring live accuracy monitoring on every prediction cycle.
🔄
Continuous Retraining & Accuracy Tracking
Models are retrained on new data every week. Each retraining cycle logs accuracy metrics — mean absolute percentage error (MAPE) by series, by layer, and by geography — against a rolling validation window. This creates a transparent, auditable accuracy record that improves over time as the PackIndex dataset grows.
👥
Pulse-Augmented Forward Estimates
The Pulse Diffusion Index is incorporated as a live feature in the prediction models. When the PDI moves sharply in a category — signalling a market turn before it registers in commodity data — the ML engine adjusts its forward estimates accordingly. This integration of human market intelligence with quantitative modelling is unique to PackIndex.
Why continuous retraining matters Packaging commodity markets exhibit structural breaks — trade policy shifts, supply chain disruptions, energy price shocks — that invalidate models trained on historical averages. By retraining weekly and monitoring accuracy continuously, the PackIndex engine adapts to structural breaks faster than any static modelling approach.
Section 08

AI Intelligence: PackIndex Advisor & PACKIQ


Data and prediction are necessary but not sufficient for commercial decision-making. The final layer of the PackIndex intelligence system is AI — two distinct products that translate the data infrastructure into language, analysis, and recommendations that procurement professionals can act on directly.

PackIndex Advisor
Live Market Intelligence, On Demand
The PackIndex Advisor is an AI assistant embedded directly in the platform, with real-time access to the full PackIndex dataset: live index values, top movers, Pulse scores, ML forward estimates, and published Intelligence articles. Users ask natural-language questions about current market conditions and receive answers grounded in live data — in any language. The Advisor operates without credit deduction and is available to all subscribers.
Live index access Pulse scores Multi-language ML forward estimates No credits required
PACKIQ
100+ Specialist Digital Experts
PACKIQ is a network of over 100 AI specialist agents, each trained on a specific packaging discipline: sustainability compliance, regulatory mapping, cost engineering, supplier risk, procurement strategy, material science, and more. Each agent combines the PackIndex data layer with deep domain knowledge — producing analysis that previously required a team of consultants. Agents run on a credit system and are accessible to all PackIndex subscribers.
100+ specialist agents 9 disciplines PFAS / regulatory Cost engineering Supplier risk Live data grounded

The AI layer is architecturally integrated: every agent and advisor response is grounded in live PackIndex data, ensuring that AI-generated analysis is always anchored to current market reality rather than generalised training knowledge. This integration — live proprietary data feeding AI reasoning — is what separates PACKIQ from generic AI tools applied to packaging questions.

Section 09

Confidence Classification Framework


PackIndex classifies every L1, L2, and L3 series into one of three confidence tiers based on data source quality, update frequency, and cross-validation coverage. Confidence tier is displayed transparently alongside each index value, giving commercial users a clear signal for how to weight each data point in procurement and budgeting decisions.

HIGH
Price sourced from a verified, authoritative market reference updated at or above weekly cadence, cross-validated against at least one independent reference. Full historical chart and ML forward estimate published.Examples: pulp grades, LME metals, grid electricity, Fed and ECB reference FX rates
MEDIUM
Price derived from the documented cascade methodology applied to High-confidence inputs, or sourced from a reliable trade reference updated monthly. Derivation assumptions available to enterprise subscribers on request.Examples: board grade indices, major polymer grades, process chemical inputs
MODELLED
Price estimated using the L1/L2 cascade and ML models where direct market data is limited. Clearly marked in the platform UI. Directional signal suitable for forward planning and scenario modelling.Examples: bioplastic grades, niche specialty films, emerging market regional variants

PackIndex commits to publishing the confidence tier for every series and to upgrading tiers as data quality improves. Enterprise subscribers have access to the full methodology notes for any series on request.

Section 10

How PackIndex Compares to Alternatives


CapabilityPackIndexTrade SurveysCommodity TerminalsBroker Letters
Finished unit benchmarks✓ Full L3 coverage△ Selected grades✕ Raw only△ Qualitative
Real-time / weekly updates✓ Daily–weekly✕ Monthly✓ Daily✕ Ad hoc
Transparent methodology✓ Published framework✕ Black box△ Exchange rules✕ None
Multi-layer cascade (L1→L3)✓ Native
Global contributor network (PDI)✓ Verified & weighted△ Qualitative only
ML forward price prediction✓ Continuous retraining△ Basic models
Global geographic coverage✓ US · EU · APAC△ Regional△ Commodity-level✕ Local only
AI-powered analysis (PACKIQ)✓ 100+ specialist agents△ Limited tools
Packaging-specific scope✓ Packaging-native△ Broad categories✕ Commodity-focused△ Partial
Section 11

Governance, Review & Editorial Standards


PackIndex operates under a quarterly data governance cycle. Bill-of-materials conversion weights, L3 canonical specifications, ML model accuracy thresholds, and the PDI calibration model are reviewed and, where warranted, revised at each cycle. Revisions are announced in the OPN Intelligence editorial feed with a minimum 14-day notice period before implementation.

Anomaly detection runs automatically on every update cycle across data ingestion, model outputs, and published indices. Values outside defined tolerance bands are held pending editorial review, completed within one business day. PackIndex does not accept commercial arrangements that influence index values — the methodology applies uniformly across all subscribers and geographies.

Discrepancy reporting If a published PackIndex value diverges from your procurement experience, we want to know. Verified discrepancies drive methodology improvements that benefit all users. Contact: packiq@opnplatform.com — reviewed within one business day.
Section 12

Commercial Applications


PackIndex is built for commercial decision-making at every level of the packaging supply chain. Procurement teams use L3 indices as live benchmarks in supplier negotiations, replacing reliance on supplier-issued price notices with data-backed market positions. Category managers use the three-layer cascade to understand which upstream commodities are driving cost pressure — and how quickly that pressure will reach converter pricing.

Finance and supply chain teams use PackIndex ML forward estimates for budget modelling and cost forecasting, updated weekly on real market dynamics. The Pulse Diffusion Index gives procurement leads a leading indicator layer — contributors in the network consistently see market turns earlier than teams relying on published trade surveys. PACKIQ’s 100+ specialist agents translate this intelligence into direct action: PFAS compliance mapping, USMCA origin audits, supplier price lag detection, recycled content auditing, and procurement strategy recommendations.

📈
Built for procurement, finance, and supply chain PackIndex and PACKIQ are commercial decision-support tools for packaging professionals. Every index, every ML model, every contributor signal, and every AI agent exists to turn market complexity into a competitive advantage for the buyers and teams who use them.