Model Intelligence Framework
One repeatable machine per model: what a car could be, what a specific VIN actually is, and how rare that makes it — every fact carrying its source, region, and confidence. The 997 is the first run.
Layer 1 · Master Set
what could exist
Variants, option M-codes, colors, transmissions, production, timeline, impossible-combo rules.
Layer 2 · Instance
one real VIN
Reconciled against the Master Set. Options only from a build sticker or COA — never guessed from photos.
Derived · Pop Report
how rare
Print run, observed pop, 1-of-N by config, rarity tier. Kept separate from value.
Trust ladder — a fact is canonical only with 2+ independent A/B sources
| Tier | Class |
|---|---|
| A | Porsche primary |
| B | Books & registries |
| C | Forum consensus |
| D | Single anecdote |
Independence is checked by source lineage — a book and a forum post can be the same Porsche bulletin laundered twice.
The era dial — sourcing & confidence shift with age
Modern · 2005+Digital build data, PET, DVLA/MOT, salvage feeds, listings. Tight.
Middle · 1990–2004Option stickers, COA, club registries, books. Mixed.
Classic · 1970s–80sKardex/COA, factory literature, registrars. Scarce.
Vintage · pre-warPhysical archives, historians, expert authentication. Wide.
Older cars: the spine flips from spec to provenance, authentication becomes mandatory, and the coverage gap-list becomes a value-ranked archival hunt.
Ships in v1 — internal
- 997 Master Set: variants, colors, take-rates, impossibility rules
- Instances reconciled from real listings — no photo-guessed options
- 3-layer identity: VIN instance · source observation · aggregate cohort
- Pop Report: observed + 1-of-N; every fact sourced & region-scoped
Deferred — needs proof first
- Survivor-population modeling — until backtested
- Public / commercial release — legal review (internal for now)
- Cross-jurisdiction survivorship merge (registries are caps, not per-car)
- Backward expansion to pre-2005 eras
See it live → 997 Carrera census