Sundr Decision Model
Changelog
Every change to the scoring algorithm, data sources, and methodology is logged here. The current model version is v1.3, stamped on every API response.
v1.3
Some discontinued devices had a marketplace listing asking far more than the thing cost new, and that listing was setting the bar for whether a repair was worth it. Those now fall back to the launch price.
- FixedA third-party listing well above what a device originally sold for is scarcity pricing, not what replacing it costs. An $89 Nokia 8110 listed at $538 made a $183 repair look like a third of the replacement cost, when it is really twice the launch price. Listings more than 15% above what a device originally cost are no longer used as the replacement figure.
- ImprovedWhen that happens the launch price stands in, because it is a real price for that exact device, and the result says where the number came from instead of calling it an Amazon marketplace price.
v1.2
Canadian repair costs were being weighed against American replacement prices. Both sides of the comparison are now in the same currency, and devices with no Canadian price get their own US price converted instead of a category average.
- FixedA repair cost in Canadian dollars is no longer divided by a replacement cost in US dollars. When the replacement figure fell back to a category average, that average stayed in USD while the repair cost had already been converted, which inflated the comparison by roughly the exchange rate and pushed Canadian verdicts toward replacing.
- ImprovedTwo thirds of the devices we track have no Canadian price. Those now use the real US price for that exact device, converted at the daily rate, instead of an average for the whole category. A real price for your device beats a guess for its category.
- AddedWhen a replacement figure comes from US pricing, the result says so under the number rather than presenting it as a local price.
- FixedThe repair-if-under threshold in the API is labelled with the currency it is actually in. It used to say CAD for everyone, including US callers reading US dollars.
v1.1
The decision data layer now returns three things the engine had been asking for and never getting, and community repair data reaches a dozen more categories.
- FixedDevices that are still sold new are no longer described as no longer sold new. The retail-window flag never made it out of the database, so every matched model was treated as retired and some replacement costs came from used or renewed listings instead of the current new price.
- FixedCommunity repair data (Open Repair Alliance) now reaches vacuums, TVs, printers, headphones, earbuds, kettles, toasters, coffee makers, food processors, sewing machines, irons, cameras, smartwatches and air conditioners. Two category names in the lookup matched no records at all, and most categories were missing from it.
- AddedBrand-level repairability averages and brand repair outcomes are used as a fallback when a specific model has no iFixit score of its own.
v1.0
Initial versioned release of the Sundr decision model.
- AddedDecision model version field stamped on every API response (model_version: "v1.0").
- AddedFrance Indice de Réparabilité data integrated as parts ecosystem signal for supported product categories.
- AddedDecision threshold rules: concrete repair-if-under / replace-if-over cost cutoffs based on category and repair cost.
- AddedData source transparency layer. AI consumers can inspect which data sources were used and where gaps exist.
- AddedAsymmetric French repairability scoring: low scores (< 5) are weighted as strong replace signals; high scores (> 9) are discounted for manufacturer self-reporting inflation.
- ImprovedBlended confidence reporting when both iFixit and French repairability data are available, weighted 70/30 to match the score blend.
- ImprovedRegional currency applied to threshold rules so non-CAD users see the correct currency code in structured output.
- AddedFive-factor weighted scoring: repair cost, remaining lifespan, environmental impact, parts availability, and repair success rate.
- AddedORA (Open Repair Alliance) community repair success data integrated as the repair success rate factor.
- AddediFixit repairability scores and device-specific repair guides surfaced in response.
- AddedConfigurable weight presets: balanced, budget, eco, longevity, and convenience. Each shifts factor weights to match user priorities.