The problem

What did a 1987 911 Carrera cost with the sport seats? That looks like a question with one answer, and it isn't. There's nowhere authoritative to check. Porsche's configurator from that year is long gone, dealer order guides were internal documents nobody had reason to archive, and what's left — enthusiast option lists, auction listings, forum posts — disagrees and almost never says where the figures came from. Most of the work here is deciding which sources to believe, and saying so when the answer is none of them.

What each kind of document can and cannot establish

Different documents prove different things, so they aren't interchangeable. A window sticker — the Monroney label required on every new car sold in the US — is the best evidence there is: printed at the time, itemised, with prices on it. But it only ever proves a positive. Sport seats at $2,090 on one car tells you the option existed at that price that year; it tells you nothing about whether you could get it on a different model. For that you need a brochure equipment grid, the only source that proves a negative — that an option was not offered on a given trim — because it lays the whole range out at once. A brochure without a grid proves neither. EPA records are authoritative for fuel economy and nothing else. So availability and price come from different documents, and the dataset keeps them as separate facts.

Rule 1 — a sticker has to add up before it is used

Every price taken off a window sticker gets checked first: base MSRP, plus each itemised option, plus destination, has to equal the total retail price printed on the label. If the sum doesn't come out, the whole sticker goes, instead of keeping the lines that look right. That one check throws out a lot of material. Mostly it catches dealer-made reproductions that reprint the option codes without their prices, which look like the real thing and prove nothing.

Rule 2 — a blank price is not a zero

Window stickers often print an option with the price cell empty. That doesn't mean it was free. The 2011 GT3 RS label that prompted this rule leaves Bluetooth, Sport Chrono, the navigation module and the audio interface blank — and all four turn up at real money on other cars from the same period. A blank means that particular label didn't itemise the price. Only a printed zero is recorded as no-cost. Everything else stays unpriced, and shows in the interface as an em dash instead of $0.

Rule 3 — availability and price are separate facts

Three states that are easy to run together: the option wasn't offered that model year; it was offered and the price is known; it was offered and the price was never recorded. Only the middle one produces a number. That's why a build total can come back as a minimum — pick options nobody has sourced a period price for and the configurator adds up what it knows, marks the total as a floor, and tells you how many options it couldn't price.

Rule 4 — facts are compiled, wording is not copied

Specifications are facts, and facts aren't copyrightable; how they're arranged here is this project's own work. Manufacturer prose is a different matter, and none of it is reproduced. Where an option needs a description it's written from scratch rather than lifted from the brochure it turned up in, and source PDFs aren't redistributed. Contributions follow the same rule.

What is verified, and what is not

Not everything here is equally solid, and the dataset says which is which. Option availability is checked against a period brochure equipment grid for 991.2, 991.1, 997.2 — those brochures contain a real range grid that could be parsed. For 992.1, 997.1, 996, 993, 964, 930, F-series no such grid survives in the marketing material, so their option catalogues are reconstructed from documented secondary sources: internally consistent and useful, but not checked against an order guide. Base prices are firmer than per-option prices nearly everywhere. The coverage table on the front page grades every generation on each of these, and hovering a cell tells you what's behind the grade.

Why coverage is a percentage and not a checkmark

Per-option pricing is the hardest part of the project and the least complete, so the coverage figures count only prices with a document behind them. Two things are kept out deliberately: options priced at zero because they were standard equipment, which is an assumption rather than something observed, and prices inferred from a neighbouring model year. Where the configurator does borrow a nearby year's price to fill a gap, it marks that figure as approximate.

Corrections

There will be errors, and a correction with a source attached is the most useful thing anyone can send. Contributions are checked against their cited source before they go in, and the dataset is published under an open licence so the underlying figures can be checked and argued with. See the data license or write to admin@retrospec-cli.com.