Group aggregates into a (suit → rank → TurnAggregate) lookup so the grid * can render in fixed positions.

Actions counted as "aggressive" for the aggression heatmap.

Pick iter cap based on (potType, sizingKey) — see parseArgs comments for * the rationale. `--iterations N` overrides via args.iterations.

Load Google Identity Services script

Look up the orbit weight for a given turn card from runout_options. * runout_options uses the same card token format ("2s", "Th", etc.).

Mean of EV values. Returns 0 for empty input (so display shows 0 rather * than NaN). Caller can detect "no EVs available" by checking the source *

Parse a combo strategy into sorted action entries

Parse a combo string like "Ad4s" into a combo index

v1.4.1: Parse an engine progress line like `[Iter 50] Exploitability: 0.00% (1771.25ms)` into an `EngineProgress`. Returns None for any line that does

Parse node locks from JSON string Format: [{"history":"Check,Bet_33","combo":"Ad4s","strategy":[0,1,0,0]}, ...]

CPU-seconds (sum across threads) inside terminal evaluation, split by terminal type. Diagnostic only — used to decide whether to attack showdown matri

v1.8.1+ per-iteration-phase timing in ms (cumulative across the whole solve). Default 0.0 — only the levelized CPU backend currently instruments these

PKCE S256 code challenge: base64url(sha256(code_verifier)).

Returns rank index 0..12 for "2","3",..."A". -1 if invalid.

Read one chart's metadata WITHOUT loading the (potentially large) per-action ranges. Used for the listing pass.

Apply per-field overrides on top of a profile.

Returns false on invalid input so main() can surface a non-zero exit code (POST_OPTIMIZATION_REVIEW Finding 4: invalid --benchmark-case used to print

v1.2.2: Run the engine with `--estimate-only` to get a pre-solve cost preview (memory + ETA). Short-lived subprocess (~50-300ms typically; monotone is

Shorten the verbose backend label for inline display. * Input: "CUDA (NVIDIA GeForce RTX 5090, 32606 MB, CC 12.0)" → "CUDA" * "CPU-DCFR" →

Shorten action labels for the mini bar display

Start the headless API server. Call this on a separate tokio task so it doesn't block the main thread.

Compute the suit-permutation group that fixes a given multiset of board cards. Returns up to 4!=24 permutations, each represented as a length-4 array

Regression for the hash-collision evaluator bug: open addressing without a stored key returned the FIRST non-empty slot, so the 197 paired-hand prime

Regression for the exact symptom the bug produced in solves: on Ks9d4h2c7s, a higher pair must out-rank a lower pair, and a set must out-rank an overp