Product Category Rule inventory → mapped onto the COMET ontology · 366 PCRs across 27 operators
generated 2026-06-29 12:11
27
operators (known universe)
25 open · 2 gated
366
PCRs inventoried
378 versions
323
source PDFs
w/ provenance + SHA-256
4494
requirements extracted
88% span-verified
67
clause vocab keys
v1-seed · validated bottom-up
4
programs live
EN · NO · PEF · DE
Source data — the underlying document corpus (provenance)
323
source PDFs on disk
every blob SHA-256 hashed
390 MB
total corpus size
~1.21 MB/doc avg
9,176
pages of source rules
median 29 · max 271 pp
4
document languages
English, ja, Norwegian
67%
rules currently valid
199 current · 98 expired
2026-06-17
corpus harvested since
latest 2026-06-29
Document language (of 323 PDFs)
English 216ja 76Norwegian 30German 1
Language is the document's publication language. Non-English rules (Norwegian, German) are extracted to English values with original-language verbatim quotes retained (span gate A4).
Document size (pages per PDF)
1–14 pp21
15–29 pp120
30–59 pp82
60+ pp24
9,176 pages total across 323 documents · median 29 pp · longest 271 pp. Corpus footprint ~390 MB on disk.
Age & freshness (PCR versions by valid-until year)
20202
202113
202214
20237
202414
202526
202661
202724
202870
202926
203035
20314
20321
green = still valid · red = expired (valid-until < today). 297 of 378 versions carry an explicit validity date, spanning 2020-10-14 → 2032-06-08. 98 expired versions are kept deliberately (living, versioned history).
PCRs by type
pcr 279cpcr 60pefcr 22subpcr 4epd 1
Extraction confidence (span gate = A4)
high 3461med 514low 519
3947 of 4494 requirements have a verbatim source quote located in the document.
COMET mapping ledger
unmapped 187extended 147exact 100lossy 6
exact = existing COMET class fits · extended = new class proposed · unmapped = gap → PR backlog
PCRs by operator
environdec227
sumpo76
epd-norge22
eu-ef22
us-epd12
bre3
epdhub2
ibu2
Method family
ISO14067299
EN1580445
PEF22
Extractor backend
LLM (Haiku 4.5)4141
regex353
LLM extractor is a drop-in replacement (same dict shape + span gate).
Top extracted clauses
id.standard_basis243
id.version242
scope.product_category239
id.valid_until239
id.program237
id.pcr_number231
id.pub_date225
id.operator219
id.language200
id.geography189
scope.inclusions183
id.cpc_code174
Harvest health (A8 watchdog)
environdec 247sumpo 76manual 40epd-norge 30
Latest found-count per operator adapter. Red = zero-result alert.
Coverage & representativeness — vs the global PCR universe
18.3%
of global universe (~1,586 PCRs)
range 14.1–25.8%
94.5%
of enumerated operators
307 hard-counted PCRs
7/17
segments touched
3/3 method families · 3 langs
$13,507
to complete (~1,586 PCRs)
LLM only $19
173 h
human review to complete
the binding constraint
$24
/quarter re-harvest (LLM)
living-DB compute
Indicative, not representative (why)
EPDs ≠ PCRs. ECO Platform reports ~33,009 EPDs across 22 operators — but those are governed by only ~600–750 PCRs (the rules PCRbase inventories). The global PCR universe is ~1,120–2,050 (mid ~1,586).
Current DB = 43 PCRs / 4 programs → 18.3% of universe.
But method & language diversity is fully proven: all 3 method families (ISO 14067, EN 15804, PEF) and non-English extraction (Norwegian) work end-to-end. The gap is volume & geography (no US/Asia yet), not capability.
Cost structure to complete
LLM extract19.0
Adapter eng3120
Human review10368.0
Compute is trivially cheap (~$23 for the entire universe). The cost is human review (~80%) + building ~13 operator adapters. Living maintenance ≈ $96/yr compute.
Global PCR universe — gap by segment
operator / segment
region
est. PCRs
ingested
status
Intl EPD System
Global
247–340
227
ingested
UL SPOT (USA) 🔒
Americas
150–250
—
not started
KEITI (Korea) 🔒
Asia
100–250
—
not started
JEMAI/SuMPO (Japan) 🔒
Asia
100–250
—
not started
Other ECO Platform (14 ops)
Europe
120–200
—
not started
Sector schemes + c-PCRs
Global
100–200
—
not started
US EPD (NSF/ICC-ES/SCS/PCA)
Americas
80–150
12
ingested
EPD Chile/Brazil/Mexico
Americas
50–100
—
not started
IBU (Germany)
Europe
30–50
2
ingested
EPD Hub
Europe
15–40
2
ingested
EPD Italy
Europe
20–40
—
not started
EPD Australasia
Oceania
20–40
—
not started
EU PEFCR/OEFSR
Europe
30–40
22
ingested
EPD Norway
Europe
30–30
22
ingested
INIES (France) 🔒
Europe
10–30
—
not started
PEP ecopassport
Europe
10–20
—
not started
BRE (UK)
Europe
10–20
3
ingested
🔒 = gated / non-English (metadata-only until acquired). Estimates have wide error bars; only per-operator enumeration tightens them.
How many PCRs exist globally? — statistical estimate (Monte Carlo, 100,000 trials)
Each of the 17 segments is modelled as a triangular(low, mode, high) distribution (mode = hard enumerated count where known, else band midpoint), then summed over 100,000 independent draws. Summing the segment lows/highs naively assumes every segment hits its extreme at once (perfect correlation); the independent sum is the statistically correct, tighter interval.
1,569
median (P50)
1,569
mean
1,467–1,672
90% CI (P5–P95)
distribution of the global total
▼
P5 · 1,467P50 · 1,569P95 · 1,672
Naive sum-of-bands for comparison: 1,122–2,050 (±29%). Monte-Carlo 90% CI is ±7% — uncertainty shrinks because segment errors partly cancel.
Representativeness by language — univ. (grey) vs ingested (teal)
EN866 · 31%
KO175 · 0%
JA175 · 0%
Other EU160 · 0%
ES/PT75 · 0%
DE40 · 5%
NO30 · 73%
IT30 · 0%
FR20 · 0%
FR/EN15 · 0%
~56% of the global PCR universe is English-first; the long tail is KO/JA (gated Asia), DE, IT, FR, ES/PT. PCRbase has proven non-English extraction (NO) but Asian-language coverage is 0% — the biggest representativeness gap.
Representativeness by sector — univ. (grey) vs ingested (teal)
Construction928 · 18%
Cross-sector industrial260 · 16%
Other consumer154 · 20%
Food/Agriculture150 · 36%
Electronics/HVAC94 · 0%
~59% of all PCRs are Construction (EN 15804 dominance). PCRbase's ingested set skews to EnvironDec's cross-sector mix, so Food/Agriculture coverage is high while Electronics/HVAC (PEP-dominated) is near 0%.