Skip to content

Claude/api key review setup 79sajy - #98

Merged
jakechinitz merged 3 commits into
mainfrom
claude/api-key-review-setup-79sajy
Jul 30, 2026
Merged

Claude/api key review setup 79sajy#98
jakechinitz merged 3 commits into
mainfrom
claude/api-key-review-setup-79sajy

Conversation

@jakechinitz

Copy link
Copy Markdown
Owner

No description provided.

claude added 3 commits July 30, 2026 00:40
The request body used Chat Completions conventions that the /v1/responses
endpoint rejects: response_format instead of text.format, content parts
typed 'text' instead of plain strings, and strict:true alongside
additionalProperties:true. Any run with a valid key would have failed
with a 400 before this fix.

Also harden response parsing (raw REST responses have no output_text
convenience field), clarify the missing-key error message, and bump
workflow actions off deprecated Node 20 runtimes.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01GM6nhbnf2bYhQQWzAh2Bwm
- Normalize assumption overrides per time block (year1..years16_20) so
  block-keyed JSON overrides get numeric shorthand wrapped into {value}
  objects and merge onto paths the engine reads; unknown block keys warn
  instead of merging into dead paths.
- Route expansionRates and dcInferenceShare reads through resolveGrowthRate
  so plain-number overrides no longer silently fall back to hardcoded
  defaults.
- Add NODE_GROUP_MAP and use it in ChartsTab, SupplyEngineTab, and
  MarketClearingTab, which indexed the NODE_GROUPS array by group letter
  and always got undefined colors/names.
- Remove dead tight2026.demandAssumptions field and its helper (never
  consumed by App or the engine; values were identical to base).
- Add supply expansion ceiling editing table to AssumptionsTab (engine read
  these rates but the UI had no way to change them).
- Render the monthly assumption update log in AssumptionsTab (was written
  by the updater but displayed nowhere).
- Harden update-assumptions.mjs: explicit output shape instructions in the
  prompt, wrap unwrapped node overrides, protect updateLog history from
  being overwritten by model output.
- Guard chart tooltips against null series values; fix hook-order violation
  in GrowthRatesTab; stop ChartsTab node selector dropping nodes past 20.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01GM6nhbnf2bYhQQWzAh2Bwm
- Reconnect gpu_inference fab capacity: it was defined (350K/mo) and grown
  but never consumed, silently deleting ~37% of accelerator supply. Both
  fab lines now feed the shared GPU pool; forecasts, tightness,
  utilization, and reported capacity use the combined figure.
- Make MW gating sequential: built datacenter capacity is no longer a
  substitute for power hookups. Only grid_interconnect and off_grid_power
  pool (they are genuine alternatives); datacenter_mw gates separately,
  and grid/off-grid split MW demand by share instead of each seeing the
  full load (~3x double-count on the power chain).
- Tie infrastructure to net fleet growth: buildings, hookups,
  transformers, construction, and ops staffing are reused when a GPU is
  replaced, so their demand tracks deployments minus retirements, and
  gating credits retired capacity. Previously a steady-state fleet
  generated perpetual phantom construction demand.
- Fix glut/contraction: tightness now compares demand flow to production
  capacity (dividing by inventory+capacity made every balanced stock
  market read as hard glut); organic expansion pauses in glut instead of
  compounding into guaranteed overcapacity; deeply idle capacity contracts
  up to 10%/yr (floored at half); the demand forecast ratio can go below
  1 so planning can see declines.
- Align duplicate intensities: cowos/advanced_wafers inputIntensity in
  nodes.js matched to the TRANSLATION_INTENSITIES values the engine uses;
  cpu_server fallback matched to the node value.

Verified in-browser: no non-finite values in any series across base and
slowdown scenarios; slowdown now shows real capacity rationalization
(HBM 10.1M -> 8.3M, CoWoS 222K -> 135K) which was previously impossible;
base case no longer reports balanced markets as gluts.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01GM6nhbnf2bYhQQWzAh2Bwm
@jakechinitz
jakechinitz merged commit e85afaa into main Jul 30, 2026
1 check passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants