# R1 ineligible / not-collected providers For every provider rated in ClusterMAX 1.0 or 2.0 (excluding the "Unavailable" tier, which `scripts/retrospective.py` excludes before ever looking at incident data) that does **not** have a `retrospective/incidents/.json` file, this records why, so the exclusion is checkable. All findings below are from direct `urllib`/`curl` probes run 2026-09-23 (no WebSearch), separate from and in some cases correcting `retrospective/status-pages.json`'s discovery notes (that discovery run pre-dates the fixed Atlassian `history.json` collector and, for a few platforms, the live surface has changed since discovery ran). No severity is ever invented for a provider whose platform has none; where a provider is listed here as "no severity field," that's a statement about the platform, not a decision to guess one. ## Collected but not fully eligible (has a `.json`, may still be excluded by `retrospective.py`'s coverage-window rule) - **Lambda** (`lambda.json`, incident.io-hosted, not real Atlassian Statuspage): `/history.json` 404s -- this platform does not expose it. Only `/api/v2/incidents.json` is available, which caps at the ~25 most recent incidents with no confirmed pagination to the start of history, so `coverage_start` is correctly left `null` per SCHEMA.md. Not superseded by the new history.json collector (that collector simply doesn't apply to this platform); the file is kept as-is. ## Hyperscaler status pages (checked individually -- no blanket exclusion) - **AWS**: `status.aws.amazon.com/data.json` is real JSON but is a current/active-events feed (2 events at check time, oldest ~hours old), not an archive; no history endpoint found. `health.aws.amazon.com/health/status` requires an AWS console login. - **Azure**: `azure.status.microsoft/en-us/status/history/` is a client-rendered React app; the fetched HTML contains no incident dates, severities, or `__NEXT_DATA__`/JSON payload to parse statically. - **Google Cloud**: `status.cloud.google.com/incidents.json` is real JSON with `severity` (`low`/`medium`/`high`) and `begin`/`end` timestamps, but it only returned 6 incidents, earliest `2026-02-27` -- short of both the primary window start (2025-11-06) and the secondary window (2025-03-26..2025-09-22) by a wide margin. No further-back archive endpoint found (`/2025.json`, `/history.json`, `/incidents/2025.json` all 404). Rolling/short window, not a full archive. - **Oracle** (`ocistatus.oraclecloud.com`): client-rendered SPA shell (2459 bytes of HTML, no incident content); no JSON API found. - **IBM Cloud** (`cloud.ibm.com/status`): client-rendered SPA shell (4925 bytes of HTML, no incident content); no JSON API found. ## status.io (CoreWeave, OVHcloud) Both are on status.io. `api.status.io/1.0/status/` gives only current status (its `/incidents` and `/history` sub-resources return HTTP 403). The HTML history page (`/pages/history/?date=YYYY-MM`) **ignores the `date` query parameter entirely** -- verified by requesting `date=2024-01` and `date=2026-08` and getting byte-identical output each time, always showing only the current rolling window (for OVHcloud: a single "September 2026" `timelineMajor` block with ~20 recent events; for CoreWeave: a "No incidents in the last day" empty state). `status-pages.json`'s discovery notes ("reachable-content-present" for 6 sampled months) reflect this same always-current page, not real per-month filtering -- a false positive in that discovery pass. A path-segment variant (`/pages/history//2025-03`) 404s. No way found to reach 2025-03 depth. ## betteruptime / Better Stack (FluidStack, Together.ai, RunPod, Prime Intellect, Cudo Compute, Hydra, GPU.NET) Confirmed on 7 representative pages (status.fluidstack.io, status.together.ai, uptime.runpod.io, status.primeintellect.ai, status.cudocompute.com, status.hydrahost.com, status.gpu.net): the only static endpoint that returns real, page-specific JSON is `/index.json` (a JSON:API document), and its `included` array holds at most the single currently-active `status_report` -- not a history. Every other guessed route (`/history`, `/history.json`, `/incidents.json`, `/feed`, `/reports.json`, `/api/v2/status.json`, `/status_reports.json`) serves the same byte-identical client-rendered SPA shell for every page on the same platform (the route is handled client-side after JS loads). The authenticated API (`uptime.betterstack.com/api/v2/...`) returns `401 Unauthorized` without a bearer token we don't have. No static path to a full incident history was found for this platform. ## Trust-center pages (Vanta), not incident status pages by design **Crusoe**, **Voltage Park**, **Shadeform**, **GMI** (trust.gmicloud.ai): each resolves to a Vanta-style security/compliance trust center, which publishes controls and audit attestations, not an incident timeline. No incident data exists on these pages to collect. ## Custom platforms (each independently probed; none reaches 2025-03 with a severity or equivalent) - **Vultr**: root page loads; `/history`, `/feed` return 403; `/index.json`, `/api/v2/status.json`, `/history.json` all 404. No endpoint found. - **Vast.ai**: hand-rolled JS widget (`index.js`) that renders `maxDays = 30` of daily uptime squares from `logs/_report.log`, colored by a computed uptime threshold (`success`/`partial`/`failure`), not a provider-assigned severity. Even if scraped, the window is 30 days, far short of the 180-day study window, let alone back to 2025-03. Too shallow and no real severity field. - **Hetzner**: `/history`, `/feed`, `/index.json`, `/api/v2/status.json`, `/history.json` all 404. No endpoint found. - **deepinfra**: same -- all guessed endpoints 404. - **Salad**: `/history` returns real JSON, but it is a fixed rolling ~7-day window (`period_start`/`period_end` unix timestamps spanning ~7 days); `period_start`/`period_end` query parameters are accepted but silently ignored (verified: a request for a much wider range returned the same byte-identical 7-day payload). No per-incident severity field either (only aggregate `outages`/`downtime_secs` counters). Too shallow, no severity. - **Massed Compute**: tiny homepage (1300 bytes); no history/feed/json endpoint found. - **Hyperbolic**: `/history` returns HTML but it is an empty client-rendered SPA shell -- zero occurrences of any date, incident, or severity string in the fetched markup. - **E2E Cloud**: `/history` returns substantive HTML (mentions of "incident"/"Resolved") but zero occurrences of any 2025 date and only 8 of "2026-" (likely component/version strings, not incident dates); no discernible per-incident start/end timestamps or severity labels found in the static markup. Would need JS execution to confirm; not implemented here without that evidence. - **Neysa** (trust.neysa.ai): despite the "custom" platform label, this is effectively another trust-center-style page; `/history`, `/feed`, `/index.json`, `/history.json` all 404. - **Mithril** (flagged `atlassian-statuspage` by the original discovery, `api_summary_ok: true`): on live recheck, `/api/v2/status.json`, `/api/v2/incidents.json`, and `/history.json` all now 404, though the homepage itself loads (200KB). Whatever backed the `/api/v2` surface at discovery time is no longer reachable; no working incident API found. - **GCore** (also flagged `atlassian-statuspage` with `api_summary_ok: true` at discovery, "0 incidents"): same story as Mithril -- `/api/v2/*` and `/history.json` now all 404 despite a large (2.2MB) homepage. No working incident API found. ## No public status/incident page located at all (status-pages.json already establishes this per-provider via candidate-URL probing; verified for Hot Aisle below since it's explicitly in scope.) TensorWave, Firmus, GMO GPU Cloud, STN, Atlas Cloud, Qubrid, Verda/DataCrunch, DENVR Dataworks, Buzz HPC, IREN/Iris Energy, FarmGPU, Whitefiber, dstack, PaleBlueDot.AI, Clore.ai, Exabits, Sesterce, Aethir, Akash, Lepton AI, SMC (the `smc.statuspage.io` slug resolves but is an unconfigured demo tenant -- its only "incident" is literally titled "This is an example incident", created 2022-03-18 -- treated as no real status page). **Hot Aisle**: re-checked directly (not just via status-pages.json's candidate list). `hotaisle.xyz` (homepage) has no status/uptime link anywhere in its markup. `hotaisle.statuspage.io` resolves but to Atlassian's generic unclaimed-slug marketing page (`Real-Time Incident Communication with Statuspage | Atlassian`), confirmed by fetching `/api/v2/status.json`, `/history.json`, and `/api/v2/incidents.json` on that host and getting the identical marketing-page bytes back for all three. `status.hotaisle.xyz` and `trust.hotaisle.xyz` do not resolve (DNS failure). No public status page exists for Hot Aisle; it is excluded from collection (and, per PLAN.md, was already excluded from the primary correlation by design regardless). ## Judgment call flagged explicitly Per the coordinator's correction, hyperscaler status pages are **not** blanket-excluded as "not comparable" -- each was checked individually above on the same retrievability/coverage terms as every other provider, and each is ineligible for a stated technical reason (shallow/rolling data or a client-rendered page with no static incident content), not because it is a hyperscaler.