How StarGuide Calculates Tier Scores, Team Ratings & Character Rankings

Where does StarGuide's character data come from?

StarGuide is built in two layers. The research layer draws on public HSR community analysis, cross-referenced across multiple sources: base kit info, eidolon and light cone mechanical effects, character-specific strategic context, starting meta positioning, relic build recommendations, and official patch notes. Sources consulted during research include Prydwen, Game8, KQM, Guoba Certified, community spreadsheets, and official HoYo patch notes. The underlying mechanical truth about characters is shared community knowledge, and every write-up on StarGuide is cross-referenced against multiple of these sources.

On top of this research foundation, StarGuide calibrates, authors, and builds: teammate synergy grades (the S/A/B decisions between every DPS and their candidate teammates), composition archetypes (which teams count as canonical prebuilts), every numeric point value (eidolon and LC score bonuses calibrated to the scoring scale), team generator and fit-gated realization formulas, relic grading math and threshold calibration, every personalization algorithm on the site, and curated multi-character team investment paths. The research layer is shared community knowledge; the calibration and algorithmic layer is what makes StarGuide's numbers its own. If you ever find something on StarGuide that reads like a copy-paste from another site, use the feedback button (visible in the corner of every page) and it will be fixed. No exceptions.

What features are free on StarGuide?

Anything that can be learned from reading public community sources is free on StarGuide, permanently, with no paywall. This includes character pages, tier lists, team recommendations, kit analysis, eidolon and light cone guides, pull advice, banner info, investment paths, and per-character relic grading. Per-character relic grading is free on StarGuide and free on tools like Fribbels.

Supporter and Trailblazer passes fund the systems built on top of that research: the Endgame Planner (non-overlapping team pairs optimized per stage), the Next Best Action simulator (every possible upgrade ranked by efficiency), Relic Insights (farming domain priority ranked by your roster's actual needs plus team readiness checks), and the Strategy Coach (tool-orchestrated roster advisor). These systems require roster-aware computation that does not exist elsewhere.

How does StarGuide calculate character investment scores?

Each character starts with a headline tier rating (T-1 through T5) based on community consensus for each endgame mode (Memory of Chaos, Pure Fiction, Apocalyptic Shadow). StarGuide deducts the assumed best-in-slot S1 light cone bonus from this headline tier to reveal the true E0 + no LC baseline. Eidolon bonuses and light cone bonuses are then added on top. The formula is: baseline = canonicalScore(headlineTier) minus bisSignatureBonus; finalScore = baseline + sum(ownedEidolonBonuses) + lcBonus.

Eidolon bonuses are direct score points calibrated by mechanical impact: kit-transforming effects (new mechanic, role change, major scaling) default to 3.3 points, significant upgrades (notable multiplier or utility) default to 1.7, modest gains (real upside, limited scope) default to 0.9, and trace-level stat boosts default to 0.6. Each value has an override range (typically plus or minus 0.5 to 3 points) to reflect actual mechanical gaps on a per-character basis.

Light cone bonuses: a best-in-slot signature LC at S1 is worth approximately 10 points (one full tier step). Strong alternatives are approximately 7.2 points, decent alternatives approximately 5.4, and budget 4-star options approximately 4.3. LC values at S2 through S4 are linearly interpolated between S1 and S5. The methodology page includes a live investment score calculator where you can pick any character, set their eidolon level and light cone, and watch the score update in real time.

What are StarGuide's tier thresholds?

Tiers run in uniform 10-point gaps with thresholds at the midpoint between adjacent canonical scores. The full table: T-1 (peak, canonical 125): score above 120. T-0.5 (excellent, canonical 115): above 110. T0 (great, canonical 105): above 100. T0.5 (very good, canonical 95): above 90. T1 (good, canonical 85): above 80. T1.5 (decent, canonical 75): above 70. T2 (fair, canonical 65): above 60. T3 (functional, canonical 55): above 50. T4 (weak, canonical 45): above 40. T5 (weakest, canonical 35): 40 or below. These thresholds are uniform across every character with no per-character overrides. The methodology page includes an interactive tier ladder with a draggable slider.

How are team scores calculated in StarGuide?

Team scoring is a weighted average of each member's realized effective score. Each character has an effective score (their tier baseline plus eidolon and light cone investment), and each team member realizes a fraction of that effective score based on how well they fit the rest of the team. The formula: realized_i = effective_i × realization(fit_i); teamScore = Σ(realized_i × weight_i) / Σ(weight_i).

Realization factors by fit grade: S+ = 1.00 (full power), S = 0.90, A = 0.83, B = 0.70, C = 0.52, D = 0.32. The gaps between grades are deliberately larger than a single tier step on a typical effective score, so synergy priority beats raw tier: a T1 DPS running their perfect S+ teammates contributes more to the team score than a T0 DPS running A-fit teammates. This matches how Honkai Star Rail actually plays, where tier ratings assume best-case teams and a character taken out of their ideal shell is no longer operating at their rated strength.

Weights: every DPS-slot member gets weight 1.5 (the lone carry in a hypercarry team, both carries in a dual-carry team, all three in a triple-carry DoT team). Supports get weight 1.0. DPS slots are realized symmetrically: a single-carry team's lone DPS realizes at 1.00 (no mutual fit to compute), while in a multi-carry team every DPS realizes at its best mutual fit with any other DPS in the team, taking the higher of the two reverse-lookup views per pair. A team where every DPS is S+ mutual with at least one other DPS realizes every DPS in full; a mismatched multi-carry pairing hits every DPS according to their best partner, symmetrically. This is what makes the score focal-invariant: the same four characters always produce the same score regardless of which DPS the caller treats as focal.

Supports are realized by whichever DPS values them most. The formula takes the max realization factor across all DPS views for every non-DPS teammate, so a support that one carry loves and another ignores still contributes their full value. The dual-carry pool union during generation also merges the authored teammate pools of both potential carries, so a support listed by one carry is considered by both.

S+ prebuilt boost: every teammate is independently checked against the authored prebuilt team shells for the DPS in the team. If the teammate appears in a prebuilt, their fit is promoted to S+ regardless of their individual base rating. The check is per-teammate, not all-or-nothing, so partial matches still reward you. This is the single biggest reason authored prebuilt teams outscore manually assembled ones. The methodology page includes a live team score calculator that breaks down every component of this formula.

Key property: a weighted average of realized scores is mathematically bounded above by a weighted average of effective scores, so there is no path by which investing in a universal eidolon (Trailblazer E6, 4-star eidolons) can inflate a team score beyond what the characters actually contribute. The tier system already accounts for universal investment by downshifting those characters' baselines, and the realization formula only squashes downward from there.

Supports do not rate DPS in the data model. Only DPS characters author teammate ratings. A support like Sparkle or Ruan Mei has no opinion on who should run them. All support-to-DPS synergy is computed at query time by reversing the DPS's authored ratings.

How does StarGuide's relic grading work?

Relic grading evaluates three axes multiplied together. Axis 1, set match: primary 4-piece gets weight 1.0, fallback 2-piece combos get 0.4 to 0.8, wrong set gets 0.15 (steep penalty, not zero). Axis 2, main stat fit: recommended main stat gets 1.0, first alternative gets 0.9, mismatched gets 0.3, applied per slot then averaged across all four stat slots. Axis 3, substat weighted effective rolls: each substat's accumulated value is divided by its minimum roll value, multiplied by the character's priority weight for that stat (four tiers: 1.0, 0.5, 0.25, 0.1), and summed across all 6 pieces.

Grade thresholds in weighted effective rolls: S at 20.0 or above, A+ at 18.0, A at 16.0, A- at 14.0, B+ at 12.0, B at 10.0, B- at 8.5, C+ at 7.0, C at 5.5, C- at 4.0, D below 4.0. A dynamic scaling factor normalizes scores across character archetypes so CRIT DPS characters with two tier-1 priority stats are not penalized against healers with just one. Per-character relic grading is free on StarGuide. Relic Insights (farming domain priority and team readiness, a premium feature) uses these grades as inputs.

How does StarGuide's Best Teams generator build teams?

The Best Teams builder does not try every possible combination of four characters. It uses a focused composition-guided pipeline. For each DPS character you own, the generator checks that character's authored compositions (team archetypes with defined structures, core characters, and path constraints). Compositions that require conditions your roster cannot meet are skipped entirely.

Authored prebuilt teams are scored first: if you own all four characters in a canonical prebuilt team, it goes straight to the top of the list and receives the S+ synergy boost. If not enough prebuilt teams are available, the generator fills remaining slots from ranked candidate pools. For each role (Amplifier, Sustain, Sub-DPS), candidates are gathered in priority order: characters the DPS explicitly rates (highest synergy first), characters who rate the DPS from their own data (reverse lookup, lower priority), and highest-tier characters of that role as a general fallback.

The generator then assembles teams within the composition's defined structure (hypercarry: 1 DPS + 2 Amplifiers + 1 Sustain; dual-carry: 2 DPS + 1 Amplifier + 1 Sustain; sustainless: 1 DPS + 3 Amplifiers). Every candidate combination is checked against hard filters: no duplicate Trailblazer variants (cannot run two versions of the same base character), and must meet any remaining composition path requirements. Candidate pools are capped at the top 5-6 per role to keep the search bounded. Every team that passes the filters gets scored using the team scoring formula, and the top teams across all compositions for every DPS are merged into the final ranked output.

What algorithms power StarGuide's personalized recommendations?

StarGuide uses seven independent personalization systems that all require your actual roster data to function. Best Teams uses composition-aware bidirectional synergy scoring with role-based fallback, not brute-force over all 4-subsets. The Light Cone Advisor uses the Hungarian algorithm for globally optimal LC assignment across your entire roster, with a retention bias that suppresses marginal swaps. The Endgame Planner uses balanced pair optimization for non-overlapping team selection per stage, trying every Team 1 times Team 2 pairing and picking the combined maximum. The Pull Advisor uses slot-gap coverage analysis weighted by DPS tier. Next Best Action ranks every possible upgrade by pull-cost divided by power-delta efficiency. Investment Paths track progress against curated multi-character build sequences with roster-aware step completion. Account Power computes a weighted non-overlapping team average across Memory of Chaos, Pure Fiction, and Apocalyptic Shadow.

What is the Strategy Coach and how does it work?

The Strategy Coach is an AI-powered conversational strategy advisor available on Trailblazer-tier passes. Coach operates in two layers. The first is context: it receives the user's full roster state (every owned character with eidolon levels, equipped light cones, and imported relic data), every unowned character (for pull analysis), current endgame rotation boosts, and the same authored character knowledge the rest of the site uses. Coach also carries memories from previous coaching sessions so it can pick up where it left off.

The second layer is tools. Coach has 20 callable tools that run in the user's browser against their actual roster data: the team generator, pull advisor, endgame planner, LC optimizer, relic grader, pity calculator, investment path tracker, account power analyzer, score breakdown, character profile lookup, composition details, banner schedule, farming priorities, and more. Coach calls multiple tools in parallel and chains multiple rounds of tool calls when a complex question requires it. For example, answering "who should I pull next" triggers a three-round pipeline: investment overview, active paths, and banner schedule together, then upgrade plans for top DPS characters, then detailed mechanical profiles for anything it plans to recommend.

Where Coach gets it right: team recommendations run the same roster-aware team generator documented on the methodology page. Pull advice simulates every possible upgrade and ranks by power gained per jade spent, cross-referencing investment paths and banner availability. Relic feedback references actual equipped gear through the relic grading engine. Pity and probability math uses a full statistical model. Character data is kept accurate with every patch. Coach also remembers goals, preferences, and account context across sessions.

Where Coach gets it wrong: in tool-grounded answers, the main failure mode is misquoted numbers when compressing large tool outputs into readable responses. In ungrounded answers (meta opinions, future characters, fuzzy questions), Coach can sound more confident than the data supports, and can underrate or overrate edge-case compositions that the authored data has not reviewed in depth. Coach cannot see character levels, trace levels, ascension status, or material inventory. Anomaly Arbitration data is preliminary. When a pattern of mistakes shows up, the fix lands in the tool responses or character data upstream, which fixes every future Coach response that touches it.

What are the limitations of StarGuide's scoring system?

Scoring is steady-state, not turn-by-turn. The system does not simulate energy regen cycles, rotation timing, or stage-specific mechanics. A team that scores high generally performs well, but edge cases (long boss phases, specific enemy patterns, strict SP economies) can flip the ranking.

Calibration reflects the current meta. When a patch shifts an archetype (new support, nerfed trace, buffed light cone), tier scores shift with it. Data is updated with every patch, but there is a lag between "the meta changed" and "every affected character has been recalibrated."

Relic grades do not measure team damage contribution. A perfectly-graded Pela still does negligible damage compared to an A-graded Acheron. The grade measures how well a character's specific build matches their ideal, not how much that build contributes to team output.

Edge-case compositions can get missed. With 85+ characters and a long tail of niche synergies, the authoring layer prioritizes common meta teams. The feedback button sits in the corner of every page on StarGuide for exactly this reason.

StarGuide is maintained by a solo developer. Calibration is a judgment call. Every rating, every tier placement, every composition pick has been reviewed at least once, but review is not proof. The methodology page exists because transparency is the best safeguard against human error.

How often is StarGuide updated?

StarGuide has been updated daily since December 2025 with character data updates, meta recalibrations, and feature additions. The methodology page shows recent updates from the changelog. Corrections driven by user feedback are the single biggest reason the numbers on the site get more accurate over time. The feedback button is visible in the corner of every page on StarGuide at all times.

Found something wrong with StarGuide's data?

Every rating, every number, every tier: if it looks off, report it. The feedback button sits in the corner of every page on StarGuide, visible on every view at any time. Every submission is read personally. Corrections driven by user feedback are the single biggest reason the numbers on this site get more accurate over time. The methodology page exists because transparency is the best safeguard against human error.