Conversation
Verified-facts pass over the major clusters, sourced per entry: Abilene to live reality (4/8 buildings, 421MW IT, ~202k GPUs vs the 450k target; 600MW expansion scrapped); Lordstown 1500->300 per Epoch's site assessment; Milam 1500->1200 per the primary releases; Hyperion notes corrected (3 Entergy CCGTs approved, 7 pending the LPSC vote - the '10 plants' claim was premature) + $50B/full ~2032; Colossus 1 TVA grid now ~300MW + CAA litigation noted; Ellendale leases 400MW w/ building 2 delivered; Helios approved past 1.6GW; IREN Horizon 1 accepted by Microsoft 8/13. Merged the duplicate Santa Teresa entries (stargate-dona-ana + STACK Project Jupiter were the same campus, double-counting 1GW). Added 4 verified sites: OpenAI Camellia (3.2GW, Effingham GA), Microsoft Kilby campus (Pecos), AWS/STACK Caddo-Bossier ($18B, MW undisclosed, 0 booked), Savannah River (Amentum, selection stage). 238 clusters, planned 129.1GW; GPU total honestly deflates 1.58M->1.33M. 484 tests green. Majors-focused pass, not all 238 re-verified; full sweep needs a fresh search-budget session.
…s age datacenters.json is a bare array whose consumers expect that shape, so the ingester now maintains a sidecar envelope (datacenters-freshness.json) with the news-scanner semantics: lastChecked moves on every completed scan, lastRefreshed only when an approval actually changes the approved dataset. The freshness registry reads the sidecar instead of kind:none, closing the audit gap where this dataset reported 'unknown' forever and could never alarm. Sidecar seeded 2026-05-19 = the last committed change of datacenters.json (git-verified), so the monitor tells the truth from day one: that dataset is stale and its in-process 6h refresher needs watching. stampPath is optional so path-injecting tests keep working. 486 tests (2 new), tsc, build green.
… live Cross-validated all 238 entries against Epoch AI's satellite-verified dataset (86 sites, CC-BY, epoch.ai/data/ai-data-centers): 56 Epoch rows matched 58 of ours, and the marquee sites agree (Abilene byte-identical at 421MW IT). Fixed what it exposed: FIVE internal duplicate pairs merged (Fairwater Atlanta, Goodnight, Susquehanna, Project Jade, and the Kilby pair this branch itself introduced - mea culpa); ~3.1GW of satellite-visible live IT power was invisible because phased campuses sat at construction/ rated-0 (Prometheus 560MW, Pryor Creek 370, five AWS/Meta campuses, Helios 130...) - flipped to operational-with-build-continuing per the IREN precedent; stale figures raised (Colossus 2 350->950MW IT, Fairwater Atlanta ->636, Google New Albany ->453); full-campus figures stored as rated cut to measured live (Eagle Mountain 520->150, three more Meta campuses ~3x over); Ellendale corrected down to 68MW live (delivered != energized); Denton GPU count withdrawn as implausible; Dona Ana planned ->1,750MW IT. Four verified missed Google sites added (Columbus 303MW, Bristow 238, Omaha 237, Cedar Rapids u/c). Every calibrated field flagged estimated + sourced to Epoch (CC-BY credit in entry sources and dataset methodology). Net: 237 clusters, operational 65->80 @ 11.2->14.9GW live. Follow-up (PR body): 15 further missed US sites incl. the QTS Richmond trio need per-site placement. Full gate green (test/tsc/build, exit 0).
aurph
changed the title
Two advisories published upstream since the last merge (browserslist unbounded memory growth, HIGH; postcss-selector-parser AST recursion, low) trip CI's npm-audit gate on every new push, blocking the whole open PR train. npm audit fix; 0 vulnerabilities after; tests/tsc/build green. Lockfile-only.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Two builds, one theme: the clusters dataset stops rotting in public, and the last freshness blind spot closes. Independent of PR #25 (branched off main; merge in either order — no conflicts).
Commit 1: clusters majors refresh (was 62 days stale)
clusters.json last moved June 26. This is a majors-focused, verified-facts pass — every change sourced, mostly primaries — not a re-verification of all 238 entries (that wants a fresh research session; this one's search budget was spent on the deals work).
Corrections:
stargate-dona-anaand the STACK "Project Jupiter" entry are the same campus, silently double-counting 1 GW planned. Merged (STACK's precise site coordinates kept, sources unioned).Updates: Colossus 1 TVA grid power ~300 MW + Clean Air Act litigation noted; Colossus 2 unpermitted-turbine record + NAACP suit noted; Ellendale/Polaris Forge leases now 400 MW with building 2 delivered July 1; Helios approved capacity past 1.6 GW; IREN Horizon 1 accepted by Microsoft Aug 13; Stargate Michigan's DTE supply-agreement details.
Added (4): OpenAI Project Camellia (3.2 GW, Effingham GA, Georgia Power); Microsoft's Kilby campus at Pecos (2 GW, Chevron); AWS/STACK Caddo-Bossier ($18B, MW undisclosed so zero booked, Amazon); Savannah River Site (selection stage, DOE/NNSA).
Net: 238 clusters, planned 129.1 GW, and the GPU total honestly deflates 1.58M → 1.33M (Abilene's target number replaced by its live count). New deal-row links (Kilby, Camellia) are deliberately not wired via
linkedDealyet — those ids live on PR #25; link after both merge.Commit 2: datacenters freshness stamping (audit gap)
datacenters.jsonreported "unknown" freshness forever — it's a bare array with no date field, so the monitor could never alarm on it. The ingester now maintains a sidecar (datacenters-freshness.json) with the news-scanner semantics:lastCheckedmoves on every completed scan,lastRefreshedonly when an approval changes the approved dataset. The registry reads the sidecar; the data file's shape is untouched.Seeded 2026-05-19 — the last committed change of datacenters.json per git — so the monitor tells the truth immediately: that dataset is stale and its in-process 6h refresher (long suspected unreliable on autoscale) is now actually being watched.
Commit 3 (Sept 1): Epoch AI cross-validation — dedup, calibrate, recover 3.7 GW of live power
Cross-validated the full registry against Epoch AI's satellite-verified dataset (86 sites, CC-BY; direct CSV at
epoch.ai/data/data_centers/data_centers.csv). 56 Epoch rows matched 58 of ours; marquee sites agree (Abilene byte-identical at 421 MW IT). What it exposed, now fixed:estimatedand carries the Epoch source; CC-BY attribution added to the dataset methodology.Net: 237 clusters; operational 65 → 80 at 11.2 → 14.9 GW live. Follow-up for a later pass: 15 further Epoch-verified US sites (~1.5 GW, incl. the QTS Richmond trio whose overlap with
meta-henrico-vaneeds manual resolution) and the 11 non-US sites if the registry ever goes global.Verification
Commits 1-2: 486/486 tests (2 new), tsc clean, build green. Commit 3: full gate re-run green (exit 0). Changed: clusters.json, datacenter-ingester.ts, freshness-registry.ts, the new sidecar, and the ingester tests.
🤖 Generated with Claude Code
https://claude.ai/code/session_01FY4ziPhfu8VBPFVy5cDMNy