2026-07-06 - 2026-07-13
Overview
19 Issues closed from 1 user
Closed
#24 --assume-cache no-ops on Nix 2.30+ — output_paths returns bare (unprefixed) store paths
Closed
#23 Calibrate the DEFAULT cost constant — it dominates real closures
Closed
#22 --cold includes the stdenv bootstrap chain; add a realistic "upstream-substitutable" mode
Closed
#21 Exclude fixed-output fetches from peak parallelism and work
Closed
#20 Report mislabels work T1 units
Closed
#19 Wire the recommendation's RAM estimate into provisioned ramGb
Closed
#18 _recommend can KeyError when peak-clamped node count isn't in node_grid
Closed
#16 Timeline/Gantt output of the simulated schedule
Closed
#15 RAM-per-core as a second constraint
Closed
#14 --provision mode: emit an EphemeralBuilder CR for the recommended shape
Closed
#13 Model per-node max-jobs (multiple concurrent builds sharing cores)
Closed
#12 History miner: extract real per-derivation build times from nix logs
Closed
#11 _recommend assumes sorted node/core grids; CLI doesn't sort user input
Closed
#10 makespan is O(n²·log n) — re-sorts the ready list on every completion
Closed
#9 critical_path drops back-pointers for zero-duration predecessors
Closed
#8 Dry-run output scraping is brittle across Nix versions/locales
Closed
#7 History prefix matching is nondeterministic with overlapping keys
Closed
#6 HEAVY table substring matching over-fires (python3.*, *-wrapper, util-linux…)
Closed
#5 to_build_set conflates "nothing to build" with "parse failure" → warm cache reported as cold
25 Issues created by 3 users
Opened
#1 Add --cold flag to estimate a from-scratch build (ignore local cache)
Opened
#2 Cost model mis-identifies the long pole: cost trivial/glue derivations ~0
Opened
#3 Mine per-derivation build times from nom / nix logs into a --history file
Opened
#4 Derive core_scaling from the derivation env, not a 35-entry name table
Opened
#5 to_build_set conflates "nothing to build" with "parse failure" → warm cache reported as cold
Opened
#6 HEAVY table substring matching over-fires (python3.*, *-wrapper, util-linux…)
Opened
#7 History prefix matching is nondeterministic with overlapping keys
Opened
#8 Dry-run output scraping is brittle across Nix versions/locales
Opened
#9 critical_path drops back-pointers for zero-duration predecessors
Opened
#10 makespan is O(n²·log n) — re-sorts the ready list on every completion
Opened
#11 _recommend assumes sorted node/core grids; CLI doesn't sort user input
Opened
#12 History miner: extract real per-derivation build times from nix logs
Opened
#13 Model per-node max-jobs (multiple concurrent builds sharing cores)
Opened
#14 --provision mode: emit an EphemeralBuilder CR for the recommended shape
Opened
#15 RAM-per-core as a second constraint
Opened
#16 Timeline/Gantt output of the simulated schedule
Opened
#17 Calibration mode: compare predicted vs actual and tune the model
Opened
#18 _recommend can KeyError when peak-clamped node count isn't in node_grid
Opened
#19 Wire the recommendation's RAM estimate into provisioned ramGb
Opened
#20 Report mislabels work T1 units
Opened
#21 Exclude fixed-output fetches from peak parallelism and work
Opened
#22 --cold includes the stdenv bootstrap chain; add a realistic "upstream-substitutable" mode
Opened
#23 Calibrate the DEFAULT cost constant — it dominates real closures
Opened
#24 --assume-cache no-ops on Nix 2.30+ — output_paths returns bare (unprefixed) store paths
Opened
#25 Stable machine-readable --json sizing contract for CI/arbitrage consumption