Close the Pythia cost frontier with sourced power BOM + labeled TCO closure (N1)

[Cuuper22]

N1 of 10: Close the Pythia DGX H100 cost frontier

The project's own top priority: the sourced Pythia scenario could not resolve econ.cost.per_token (33 missing roots).

Approach

• New gpu_stack/presets/dgx_h100_tco.py: dgx_h100_node_power_bom (sourced DGX H100 component power facts: CPU, NIC, RAM, misc) + pythia_70m_dgx_h100_run_closure_assumption (every non-sourced root needed by the cost rollup, explicitly labeled as an assumption with per-field rationale — never silent defaults).
• New combined pack pythia_70m_dgx_h100_us_2024_industrial_full_tco_assumption registered in SOURCED_SCENARIO_PACKS.
• The original sourced-only energy-floor pack keeps its 33 missing inputs visible by design — sourced facts stay distinct from assumptions, per the project's provenance discipline.

Observed result (first end-to-end real-scenario cost the project has produced)

tokens_per_second = 1268976.3   ok, 21 trace steps
job_dc_power      = 10200.0 W   ok, matches DGX H100 system spec
run_power_cost    = 54.44       ok, 30 trace steps
cost_per_token    = 3.738e-9    ok, 75 trace steps, missing=0

Verification

• Full pytest: 707 passed (670 baseline + 37 new)
• Audit gate: PASS

Note: agent was interrupted by a session limit after implementation+tests; coordinator validated, committed, and pushed. Known merge overlap with #N4 branch in scenarios.py registration lines (resolved at integration).

https://claude.ai/code/session_01Eu2JVnPFgMQftwYTP3cGQZ

---

Generated by Claude Code