{"version":"1.0.0","asOf":"2026-09-14","title":"Agent Provider Depot — scoring methodology","citation":"Agent Provider Depot (2026). Scoring methodology v1.0.0. GET /api/methodology · human: /methodology · research: /research/methodology.md","goals":["Help an agent (or human) pick a default provider per job-to-be-done (JTBD) quickly.","Surface tradeoffs marketing pages hide: lock-in, privacy defaults, reliability under load, DX for programmatic agents.","Prefer real URLs and honest confidence over fake precision."],"apiScale":{"min":0,"max":10,"note":"HTTP API and TypeScript types use 0–10. Research seed JSON (research/*.json) uses 0–100; the store normalizes on ingest."},"dimensions":{"capabilityFit":{"label":"Capability fit","higherIs":"better","meaning":"How well the provider fits agent/production workloads in its category (quality, features, model/HW coverage)."},"pricePerformance":{"label":"Price / performance","higherIs":"better","meaning":"Value delivered per dollar for typical agent workloads (not absolute cheapest)."},"latencyReliability":{"label":"Latency / reliability","higherIs":"better","meaning":"Uptime, latency consistency, rate-limit predictability, SLA maturity."},"developerDx":{"label":"Developer & agent DX","higherIs":"better","meaning":"Docs, SDKs, OpenAI-compat, CLI, dashboards, time-to-first-call — for agents, not just humans."},"privacyPosture":{"label":"Privacy / data retention","higherIs":"better","meaning":"Data retention defaults, training opt-out, regional controls, SOC2/HIPAA/BAA, ZDR options."},"lockInRisk":{"label":"Lock-in risk (higher = worse)","higherIs":"worse (more lock-in)","meaning":"Matches research: higher = more lock-in / harder exit. Overall formula contributes (10 - lockInRisk) × weight."},"maturity":{"label":"Maturity / track record","higherIs":"better","meaning":"Track record, funding/stability, production references, API stability."}},"defaultWeights":{"capabilityFit":0.25,"pricePerformance":0.18,"latencyReliability":0.15,"developerDx":0.15,"privacyPosture":0.1,"lockInRisk":0.07,"maturity":0.1},"overallFormula":"overall = 0.25*capabilityFit + 0.18*pricePerformance + 0.15*latencyReliability + 0.15*developerDx + 0.10*privacyPosture + 0.07*(10-lockInRisk) + 0.10*maturity. Raw lockInRisk stays higher=worse; only the aggregate inverts.","scoreKinds":{"curated_estimate":"Opinionated research estimate from public docs/pricing. Default for this catalog.","measured":"Backed by first-party benches or production telemetry the Depot has seen.","hybrid":"Curated estimate adjusted by measured signals or accepted agent insights."},"confidence":{"high":"Official pricing/docs checked; widely used; stable product.","medium":"Public info good but sales-gated pricing, rapid model churn, or thin SLA data.","low":"Marketplace volatility, limited sources, or discontinued status."},"jtbdPicks":{"rule":"Each JTBD has depotPickProviderId — take it unless whenNotTopPick matches your constraints.","runnersUp":"Strong alternatives when the pick is blocked by region, budget, model, or existing vendor.","reweighting":"GET /api/leaderboards?jtbd= reweights provider scores with that JTBD's weights. Per-provider jtbdScores overrides win when present.","audience":"Picks optimize for agent builders shipping products, not hyperscaler enterprise RFPs (those may prefer Bedrock/Azure/Vertex even when Depot picks Anthropic/OpenAI direct)."},"researchMapping":{"note":"research/*.json uses 0–100 and different field names. API store maps them.","fields":{"capability":"capabilityFit / 10","pricePerf":"pricePerformance / 10","reliability":"latencyReliability / 10","dx":"developerDx / 10","privacy":"privacyPosture / 10","lockInRisk":"Scale only: api.lockInRisk = research.lockInRisk / 10 (polarity preserved: higher=worse)","maturity":"maturity / 10"},"categoryIds":"Research categories use slugs (inference-apis). API categories use ids (cat-inference-apis) plus slugs (inference-apis). Lookups accept either."},"sourcesPolicy":["Prefer official product / pricing pages.","Then primary company announcements.","Reputable comparisons as secondary only.","Prices move weekly in 2026 LLM/GPU markets — treat pricingNotes as directional; re-verify before procurement."],"reviewCadence":"Frontier LLM and GPU clouds monthly; auth/email/storage quarterly. Seed lastReviewed is 2026-09-14.","whatWeDoNotClaim":["We do not publish synthetic latency benches as gospel.","We do not score ethics beyond privacy/compliance posture.","We do not auto-scrape live prices into this seed; seeds are curated snapshots.","Library-only tools (Playwright, Puppeteer) may appear as tags; they are not hosted providers."],"humanUrl":"/methodology","machineUrl":"/api/methodology","researchUrl":"research/methodology.md","researchSemantics":{"version":"1.0.0","asOf":"2026-09-14","scale":"0-100 for all dimensions","dimensions":{"capability":{"meaning":"How well the provider fits agent/production workloads in its category (quality, features, model/HW coverage).","higherIs":"better"},"pricePerf":{"meaning":"Value delivered per dollar for typical agent workloads (not absolute cheapest).","higherIs":"better"},"reliability":{"meaning":"Uptime, latency consistency, rate-limit predictability, SLA maturity.","higherIs":"better"},"dx":{"meaning":"Developer and agent DX: docs, SDKs, OpenAI-compat, CLI, dashboards, time-to-first-call.","higherIs":"better"},"privacy":{"meaning":"Data retention defaults, training opt-out, regional controls, compliance (SOC2/HIPAA/BAA), ZDR options.","higherIs":"better"},"lockInRisk":{"meaning":"Difficulty of exiting or multi-homing. Proprietary APIs, unique data formats, committed capacity, closed ecosystems raise this score.","higherIs":"WORSE — more lock-in risk. Unlike other dimensions, a high lockInRisk is a negative signal.","note":"IMPORTANT semantic difference vs some app TypeScript types that invert this (higher = easier exit). Research seed files use higher = MORE lock-in (worse). Builders must map/invert if ingesting into inverted schemas."},"maturity":{"meaning":"Track record, funding/stability, production references, API stability, company longevity.","higherIs":"better"}},"overall":{"meaning":"Opinionated weighted blend for general agent-stack use. Not a pure average — capability and reliability weigh more; lockInRisk is inverted in the blend (high lock-in lowers overall).","defaultWeights":{"capability":0.25,"pricePerf":0.18,"reliability":0.15,"dx":0.15,"privacy":0.1,"lockInRiskInverted":0.07,"maturity":0.1},"lockInInvertFormula":"lockInContribution = (100 - lockInRisk) / 100 * weight"},"confidence":{"high":"Official docs/pricing verified recently; widely discussed production use.","medium":"Public info solid but pricing opaque or product changing fast.","low":"Thin public data, marketplace volatility, or historical/sunset status."}}}