SWGOH Tools

Recent changes

• v1.38 Site: Search optimization based on a fresh look at how the site shows up on Google. Character page titles now lead with "Best Mods for [Character] (SWGOH)" instead of "[Character] Mods Guide" — the new pattern matches what every top-ranking competitor (swgoh.gg, gaming-fans, allclash) uses for the same long-tail queries, and the audit showed roughly fifty distinct "[character] mods" searches we were appearing on but not getting clicked. Mod Analyzer title now includes the word "Optimizer" explicitly since that's the actual searched term that other tools dominate; the differentiator is scoring rather than auto-optimization. ROTE Planner title leads with "Plan Your Rise of the Empire" to anchor it as SWGOH-specific when Google would otherwise interpret "rote" as "route". News articles now show a tool callout both above and below the body — most readers scroll past the top placement, and the bottom placement catches engaged readers right when they're deciding what to do next
• v1.37 Advisor: Step back, swap the engine. The advisor chat now runs on a stronger reasoning model with hidden reasoning enabled instead of the speed-tuned model it had been using. That model swap is the single biggest quality lever — most of the prior failure patterns (invented percentages, doubling down on wrong relic claims, mixing up zeta and omicron gates) were the speed model under load on a long, rule-heavy prompt. Output budget raised from ~600 words to ~2000+ so composite-ranked zeta and omicron lists plus cross-comparison answers stop silently truncating mid-list. Tone rewritten as a knowledgeable SWGOH friend rather than a corporate help bot — the prior "be direct and terse" framing was actively fighting the rule that says to apologize and concede when corrected, and the model kept choosing terse over honest. Numbers rule trimmed from a list of eight forbidden percentage examples down to one short sentence; the long version was a textbook case of "the more you forbid it, the more you remind the model the option exists." Signal tags on each zeta and omicron line had one last numeric leak ("tier 1", "tier 2") that's now replaced with a qualitative bucket
• v1.36 Advisor: Three fixes from a real chat where the advisor was still inventing percentages and pretending to rerank. Root cause for the percentages: the composite ranker was emitting strings like 'popularity 67%' in the signal tag, giving the model a template to fabricate 'leads at 75% offensive win rate' style numbers downstream. Signal tags now use qualitative buckets ('popular meta leader', 'you use this character often') with zero percentages anywhere in synthesized content. Numbers rule in the prompt got extended with the new fabricated examples (75% offensive, 70% offensive, 44% of teams, 10% of teams) for direct pattern-match avoidance. New rerank rule: the composite ranking is fixed at load time and the model is now told this explicitly — it can FILTER the list (drop entries not matching a constraint like 'only GAC') but cannot reorder. When asked to rerank, the honest answer is 'the ranking is fixed; here's the filtered list'. To make 'exclude TW omicrons' an actionable filter for zetas, each zeta that's a precursor for an unlearned omicron now carries the precursor's mode list — 'unlocks an unlearned omicron on the same ability in [GAC + TW]' — so the model can act on mode-based filter requests

Full changelog