[dayfine]

Depends on #183 (base frlg/rng-callable-steps; the diff shows only this change).

AutoStory's RNG plan (#182) stops twice and leaves SID Helper and Starter RNG to the user. A new option, "Run Starter RNG in AutoStory" (off by default; off keeps today's stops), runs both in the program. At 00.01, when the RNG plan's SID is empty, it types the name, runs SID Helper's steps from the keyboard and reads the TID off the Trainer Card, then goes on at 00.01b. At the end of 00.04, saved under Squirtle's ball, it runs the plan's attempts with StarterRngAttempts and feeds each outcome to RngPlanTracker. A hit that isn't shiny goes on to the next attempt, since each attempt resets to the save itself. A miss is retried, at most 2 times, then the loop moves on to the next attempt still shiny for a SID left; it stops only when none is left. A shiny is identified against the target: only a shiny on the target narrows the SIDs. One off the target confirms nothing and logs outcome=shiny_off_target next=stop. Either shiny goes on to the rival's battle (00.04b from the rival's trigger), saves after it as the manual path does, and runs on to the end checkpoint. If Starter RNG refuses an attempt's timings, the loop logs it and skips to the next attempt (starter_rng=setup_error). Anything else, including a setup error such as "No sound detected.", stops the run with the save still under the ball. Each attempt has a 3-hour cap and the loop a 12-hour cap. Every attempt logs an RngPlan: result line, and a shiny that leaves one SID logs RngPlan: confirmed ... candidate=k of=N.

Reading the log: every line of the loop starts with RngPlan:, in this order. First tid= (where the TID came from), then sid_helper (SID Helper's candidates), then one attempt= line per planned attempt (seed, advances and the SIDs it is shiny for). For each try of an attempt, a state line at its start and after every reset gives the attempt and try counts, resets, seed, advances, the attempt's SIDs, the SIDs ruled out, the try's and the loop's time against their caps, and the calibration. The state lines also show on the overlay log. After each try, a result line gives the outcome, Starter RNG's own outcome (attempt_deadline, out_of_resets, no_matches and wild_shiny are told apart, though all count as a miss; setup_error is an attempt that couldn't run), the nature, IVs and PID when known, and next=retry|next_attempt|confirm|stop. Last, confirmed with candidate=k of=N, when a shiny leaves one SID. The exact formats are in the comment above rng_log_tid() in RngManipulation/PokemonFRLG_RngPlanLog.h.

• Starter RNG now starts from the RNG plan's calibration (the console's seed and CSF offsets from FRLG AutoStory: plan Starter RNG attempts from SID Helper's candidates #182, plus its own in-game offset, STARTER_INGAME) instead of 0/0/0, and saves each reset's calibration back to it, so a stopped run keeps what it learned; tested in PokemonFRLG::AutoStoryRngLoop.

Tested with the new PokemonFRLG::StarterRngLoop suite (the loop's decisions on scripted outcomes and a fake clock) and PokemonFRLG::AutoStoryRngLoop (end messages and the rows after a shiny), run with a test runner linked against a local full build alongside RngSteps, RngTargetSearch, the RngPlan suites, AutoStory and AutoStorySimulation (82 passed, before the move below). Code health shows 0 failures. Not run on a Switch yet.

The loop moved unchanged from AutoStory/PokemonFRLG_AutoStoryRngLoop into RngManipulation/PokemonFRLG_StarterRngLoop and PokemonFRLG_StarterRngConsoleSteps in 9f01402; AutoStoryRngLoop keeps the story glue (StarterRngPlanned, the end message, rows_after_starter_rng): git diff --color-moved=dimmed-zebra --color-moved-ws=allow-indentation-change 8ff380d27 9f014021d.

• One short line per reset on the stream's overlay log, e.g. Starter RNG try 27/50: seed e66e (+1 frame late), advance 1677 (target 1670, +7). Missed. (rng_reset_line(), tested in PokemonFRLG::StarterRngLoop); the long RngPlan: state lines stay in the file log only.

🤖 Generated with Claude Code

https://claude.ai/code/session_01AKYzhrMnPK61XHWEch4a6E

[dayfine]

Superseded by #214 (FRLG stack consolidated into stage PRs; will be resplit for upstream).