Numerical acceptance benchmark: Tarnished.dev public v1.17 UI only. Matching two independent same-input reference-site observations is sufficient for a site-parity PASS; current game-client 1.17.3 evidence is not an additional numerical gate. Unobservable source-site scenarios remain UNOBSERVABLE, not MATCH. Data rights and publishability are separately reviewed.

Final four-engine accuracy decision · 2 November 2026

Full-engine parity not certified: 0 of 4. All four calculator areas have working limited-scope functions, but none can be advertised as generally equivalent to Tarnished.dev v1.17 or to the current game client. This is the final combined evidence disposition, not another repeated baseline audit.

Armor Optimizer — actual site rank observed: two public v1.17 Chromium sessions reproduced the same Poise-only Top 10. All first eight reached Poise 70, but their order is not a simple lightest-first tie; the local exact independent 13-metric Top 5 uses a different stated objective and is not claimed to match full site candidate/rank rules.

Damage Calculator — four real actions independently reproduced: Wretch STR15 DEX20, Standard Uchigatana +25, real selected 1H R1 1/2, R2 1 and charged R2 1 yielded 294 / 298 / 389 / 526 both browsers, and the pinned motion-stage model matches all four in this same-input scene. Changing the site's visible Physical Defense 100 → 150 did not change those displayed numbers, so nondefault target defense and true source DPS 464/547 remain unverified.

Spell Calculator — seven sorceries, 62 actual site matches: 37 original exact-input rows, 17 new staff/reinforcement combinations, and eight fresh Magic Glintblade and Night Comet normal/charged rows. All use independently sourced attack coefficients and pinned catalyst math. Other spells, incantations, boosts, AoE, real nonzero enemy defense and arbitrary effects remain unverified.

Build Planner — eight actual activation/mixed scenes: two browsers independently read four Wretch10 Godrick Rune/INT-knot Physick ON/OFF states and four additional Starscourge Heirloom + Rune + Physick states (32 independently summed attributes). OFF/OFF does not grant +5/+10; the conflicting historical selected-only preview was retired. Other character origins, equipment combinations, AR and effect duration remain unverified.

Rechecking closed 487 weapon weights and 568 legacy armor records is discontinued. An unobservable reference field cannot be changed to a verified value by repeating the same test. Reopen only with new, legally usable independently versioned source parameters or genuinely new game-client evidence. No full game-client numerical verification is claimed.

Read machine-readable four-engine acceptance and explicit blockers · Read final consolidated decision and evidence.

Latest v1.17 evidence closure · 2 November 2026

Completed reference-data work does not need repeating: all 487/487 weapon weights; 460/487 directly visible weapon requirements; 549/568 distinct legacy armor weight/Poise pairs; 515 armor pieces × 12 context-safe fields (514 legacy + Gravebird Bracelets); 10 original classes, 360 naked base-vitals readings, and 18/18 exact-input independent Glintstone Pebble calculations now match actual Tarnished.dev v1.17 readings. The spell result applies only to the stated staff, reinforcement, target and effects context.

Closed as unobservable, not as numerical errors: 27 weapon Req fields which the site never renders under accepted inputs, and 19 legacy armor rows that are unavailable or duplicate/miscategorized in the reference selector. These are intentionally not represented as zeroes or invented reference values.

New, independently verified second catalyst (2 November): Academy Glintstone Staff +25 with Glintstone Pebble, Wretch other stats 10, no effects and target defense/negation 0, now has four additional exact site-matched magic results: INT40 = 319, INT50 = 368, INT60 = 416, INT80 = 472. All four were observed in two independent reference-site Chromium sessions and matched the independent scaling formula 4/4. These are distinct from the original 18 Astrologer's Staff samples, and do not certify arbitrary spell/catalyst builds. Inspect real-browser comparison or read exact-scene coverage.

New +20 Academy catalyst (2 Nov 2026): Wretch INT40/60, Academy Glintstone Staff +20 and Glintstone Pebble, no other effects, source zero enemy defense/negation. Two fresh independent real browser sessions agree with our independent formula on 294/294 and 379/379 (2/2 exact matches). The reference Spell Calculator explicitly assumes zero enemy defenses and resistances and has no nonzero-target inputs, so those target scenarios remain UI UNOBSERVABLE / NOT CERTIFIED, not numeric passes. Actual browser evidence · Exact acceptance scope.

Four features are still not equivalent to the reference implementation

Armor: complete DLC catalogue, all 13-score settings and the reference's exact ordered/tied top-five ranking. Damage: real enemy/NG/PvP data and move-specific timing, motion values and DPS. Spell: other spells, catalysts, effect stacks and nonzero target defense. Build: unrestricted armor, talisman, Physick, Great Rune and conditional-effect composition. Existing local calculators work for their explicitly supported inputs, but these general same-input parity claims are NOT CERTIFIED.

Release decision: verified core data is closed; missing algorithms and unobservable reference UI data are separate from unfinished item-number checking. Do not repeat large weapon/armor audits to hide functional gaps. Read the machine-readable, regression-checked scope.

New local implementation (2 Nov 2026): The opt-in Armor Optimizer now calculates and displays up to five exact independently weighted armor combinations under the existing 13 attributes, Poise, equipment-load and slot constraints. Separate exhaustive test-oracle coverage checks its mathematical ordering. This does not establish Tarnished.dev's unpublished ranking/tie semantics or missing DLC pool, so the Armor Optimizer remains NOT PARITY CERTIFIED. Read the precise algorithm and acceptance limits.

Remaining-only Tarnished.dev v1.17 parity disposition

Completed prior direct-site evidence is not repeated. Among the original 568 armor source records, 549 unique items have site-verified Weight and Poise; 514 of them also have 12 stat-neutral, directly site-derived negation/resistance fields. The remaining 19 are classified: 6 Brave's and 6 Ragged items not selectable by exact name, 5 other names not selectable, and 2 duplicate/incorrect-slot records (Champion Gauntlets / Redmane Knight Helm). These are UNOBSERVABLE, not zero-stat equipment and not numeric mismatches.

The original exact Icerind +0 1H 60/60/30/30/30 browser observation was 184 while source IEEE-754 math was 183. The site-matching total now uses 184 only for that exact frozen double-read input; other AR inputs and raw source arithmetic are unchanged. Historical 347/348 audit counts below describe the original pre-fix run.

The opt-in armor optimizer now defaults to the site-verified-only item subset: 549 candidates for Poise or 514 stat-neutral candidates for each additional metric. Selecting source-only research is explicit. This makes the input fields more defensible but does not certify optimal four-piece combinations against Tarnished.dev's independently different minima/maxima/multiplier rules.

Still not equivalent, and therefore not numerically signed off: Tarnished.dev's selected enemy/NG/moveset Damage Calculator versus our manual-defense damage estimate; automatic spell/talisman/Crystal Tear effect stacking; and 29 stat-modifying armor pieces in arbitrary build contexts. This is missing feature equivalence, not more weapon or armor item records to recheck. Exact remaining IDs and feature-level decisions.

Direct Tarnished.dev v1.17 armor audit — 568 attempted, 549 confirmed

All 568 existing armor records were tested as real individual selections in the public Build Calculator, across 15 independent batches with two fresh Chromium reads per candidate. 543/568 were directly observed with matching Weight and Poise on both reads. Six additional unique source items were recovered by independent direct-site alias/boosting-helm comparison, bringing the verified total to 549/568; 19 source IDs remain without approved distinct-site goldens (including two duplicate legacy rows).

Of the 543 successfully observed pieces, our historical source disagreed on 23 weights and 531 Poise values. These exact fields have now been corrected in the user-initiated armor loader to match Tarnished.dev. Together with six more independently recovered items, that is 1,098 verified numeric fields (549 weights + 549 Poise values), not 568 complete armor stat blocks. Six recovered aliases and two duplicate source records.

Additional 12-field live verification: for these same 543 selected armor items, two independent browsers reproduced all 6,516 damage-negation and resistance display values. Of these, 514 pieces left final character attributes unchanged, allowing 6,168 individually baseline-derived site-reference armor fields to replace old source numbers. The other 29 stat-changing items retain contextual-only observations, not universal armor-property overrides. Direct 12-field numeric evidence.

Not covered by this result: general stat-changing item interactions, special effects, optimal armor combinations, spells, talismans or enemy damage. Those features require separate exact-input public-site comparisons. Initial batch evidence and 25 original exceptions.

Direct Tarnished.dev v1.17 numerical audit · 1 November 2026

This is a comparison to the actual public Tarnished.dev v1.17 calculator, not certification against a current game client. The original community catalog and separately obtained pinned v1.17 parameters have different evidence levels. Every direct-site number below was read twice with identical effective browser controls.

All 86 previously approved same-input AR configurations remain separate and passed. The new 348 +0 AR observations are not a complete test of every upgrade, affinity, or grip. A proposed global floating-point rounding change was withdrawn because it broke six of 39,552 independent source-math checks. Base attack components, Armor, active Build effects, enemy Damage, and Spell numerical parity are not certified by these weapon tests.

Evidence and field-by-field limitations: 86-case AR audit · 348-case +0 AR comparison · 487 directly observed weights · 460 public Req checks + 27 pinned-only source records.

Important unresolved requirement source conflict: Frenzied Flame Seal was not directly measurable via the public site's conditional Req field. An older community file says Faith/Arcane 10/10, while the pinned v1.17 source says both 0. This page labels the latter as pinned-source-only and preserves the disagreement rather than presenting it as a verified UI value.

Release gating: Complete comparator-only verification does not require game installation. Current-game-client certification, data redistribution rights, performance/SEO checks and indexing release are distinct claims. Public pages remain research previews until their own relevant gates pass.

Reference benchmark first — no game installation required

Historic 50-case reference benchmark (an earlier subset of 86 approved cases). The detailed table below shows the original 15 cases; all 35 additional exact inputs, source AR outputs and evidence are documented in the 50-case report. Browser input evidence: first full 15-case public UI run and independent recheck. Our v1.17 numerical results: AR computation run. For new cases see first full 50-site run and independent 50-site recheck. The final raw-sum aggregation follows 50-input rounding evidence. This does not certify every weapon, all mechanics, actual PvE damage or the current Regulation 1.17.3 game client.

Tarnished.dev publicly labels its Elden Ring calculator data v1.17 (site update 28 August 2026). We use that site as our primary functional and calculation-output benchmark for a separately implemented v1.17 reference calculator, not as a claim that every value is independently certified against the newer Regulation 1.17.3 game client.

Non-blocking website release rule: real-game screenshots are optional for a clearly marked third-party reference based v1.17 feature, once a reproducible set of same-input weapon/affinity/upgrade/attribute/one-hand/two-hand cases agrees with Tarnished.dev and all known discrepancies are labelled. Do not imply that simply linking to Tarnished.dev is equivalent to a successful numerical comparison. The pinned v1.17 reference AR engine now meets the documented representative 50-case external parity matrix; this is limited to these cases and not a blanket patch-accuracy claim.

Strictly separate claim: saying verified current Regulation 1.17.3 game-client AR still requires current-version evidence. If a numeric case disagrees, display or block that specific result; do not silently use the competitor as an absolute ground truth, copy its proprietary software or redistribute its full data. Source permits an independent implementation, not automatic relicensing.

Working reference-AR path: local v1.17 weapon data workbench (free pinned public JSON, opt-in download; no account or paid API). 487-weapon reference catalog shows named published +0 candidates. The former screenshot queue remains an optional research track, not a prerequisite for user-facing v1.17 benchmarked reference tools.

Live catalog evidence audit

Counts below are generated during the same build as this site from the exact bundled weapon JSON, not manually transcribed. A present field is not proof of current-patch accuracy. Weapon requirements and weights are community-reported; this preview does not contain fully validated per-weapon AR profiles.

Unresolved source conflicts (7 catalog entries)

Source snapshot version: not established; do not assume 1.17.3 compatibility. Upstream revision: 1d2626ceda5eb8c08997e345e303344a64edc106. This is a catalog consistency audit, not game-client QA. Open build-generated machine-readable evidence.

Seven unresolved weapon-source disagreements

These are two contradictory published data sources, not two sets of game-client screenshots. The first value is from our attributed community +0 reference; the second comes from the pinned Regulation 1.17 parameter source after applying the +0 reinforcement factor. Data order for attacks: Physical / Magic / Fire / Lightning / Holy; for requirements: STR / DEX / INT / FAI / ARC. A source disagreement is not proof that either side is correct for App 1.17.2 / Regulation 1.17.3.

Tracked version continuity (not weapon-menu certification): 1.17 → 1.17.1 had 184 hash-identical parameter tables; 1.17.1 → 1.17.3 had no tracked headline numeric changes, while some projectile and animation tables changed. The complete 1.17.3 regulation binary is not asserted identical, nor are the six exact weapon +0 vectors independently verified by that patch comparison.

Unversioned third-source +0 checks: 6 of 6 published base-attack disputes match the pinned Regulation 1.17 vector. This favors the pinned source as a provisional published-data candidate for these six weapons. However, the third-party weapon pages lack an explicit current App/Regulation version and do not document independent data extraction lineage. Agreement is NOT game-client verification. All six original source disagreements stay open; trusted AR and upgrade calculations continue to block the disputed +0 figures.

Current-patch corroboration: 1 of 7 historical disagreements. Fire Knight's Shortsword also has a contradiction inside the same wiki page: its basic Stats section says PHY 75 / FIRE 32, while the Standard +0 row says PHY 72 / FIRE 22. The pinned Regulation 1.17 source says PHY 75 / FIRE 22. These three different published readings are shown for provenance, not spliced into a manufactured answer. See the original article. Madding Hand FAI 20 in an independently published Regulation 1.17.3-labelled parameter reference agrees with pinned Regulation 1.17; unversioned wiki reports FAI 13. This supports 20 as a version-labelled source reading, not an original game-menu screenshot or definitive verified wield verdict. Six +0 attack disputes remain without independent comparable current-patch +0 game-menu readings.

How to resolve one entry

In the actual App 1.17.2 / Regulation 1.17.3, capture the title-screen version and the exact weapon at +0, default/uninfused state. For six base-attack cases, compare the weapon information panel’s five base damage components, not character AR. For Madding Hand, capture the weapon’s five attribute requirements, not effective two-hand Strength. Keep screenshots locally and record platform, date, file SHA-256, weapon, reinforcement and affinity. A separate reviewer must inspect authentic game images and both version labels before changing a source flag. Numerical agreement between websites alone cannot close an item.

Current accepted game screenshot count: 0 · all seven items UNREVIEWED. Download source comparison and manual verification checklist JSON · Record the seven +0 game readings locally · Open existing local AR/observation workbench. No game files, screenshots or user identifiers are uploaded.

Compare all seven disputed weapons against a current Regulation JSON

A versioned parameter-difference tracker reports all 184 regulation parameter tables hash-identical from 1.17 → 1.17.1, and no changes to its tracked weapon/requirement headline numbers from 1.17.1 → 1.17.3. There are other changes to projectiles and animation data in 1.17.3. This is valuable evidence for version continuity, not a per-weapon original-client verification. Inspect the 1.17.3 diff.

Provide your own legitimately obtained compact JSON with weapons and reinforceTypes arrays/tables (the same research data format as the local AR workbench). This is not a regulation.bin or arbitrary CSV importer. The browser compares six neutral-affinity +0 base attacks and Madding Hand's five requirements, without uploading or changing any live weapon record.

Loading the site's conflict reference data…

App/Regulation labels you enter are not authenticated. A matching local source file, its SHA-256 fingerprint, or identical parameter numbers do not establish an original game client, valid screenshot, distinct extraction lineage or true character AR. Any export stays UNREVIEWED; all seven published conflicts remain unchanged until separately reviewed.

Expanded game-client evidence collection

The previous 7-case capture remains available below. A separate 20-field original-game evidence queue now includes the seven legacy cases plus Stone-Sheathed Sword, Ritual Thrusting Shield, and the eight newly disputed recovered weapons. It records local screenshot fingerprints and UNREVIEWED values only; no screenshots have been independently inspected.

Record real game-menu evidence for the seven disputed weapons

Use this form only while viewing the actual game. It records all five +0 weapon-detail attack components or the five requirement values, not your character's total Attack Rating. Select two local screenshots: the version/title screen and the weapon details. The browser computes SHA-256 file fingerprints; screenshot pixels are not uploaded, inspected or included in JSON.

All exports are UNREVIEWED evidence drafts. Matching one published source, matching a screenshot's hash, or manually typing version numbers never means independent gameplay verification.

Each picture must be a real PNG/JPEG/WebP file of at most 12 MB. This form only verifies file signatures and SHA-256 bytes, not that the image actually depicts the stated game, patch or weapon.

Loading the local seven-case audit…

Collection progress (not verification progress)

0 / 7 player observations captured. 0 independently reviewed.

No data is automatically saved or uploaded. The downloadable JSON contains screenshot filenames, SHA-256 digests, and your typed numeric values, but no screenshot contents. Keep both original screenshots privately to allow separate human review. Reloading this page discards unsaved session entries.

Reopen your seven-weapon evidence checklist

Import a JSON checklist exported above. The page checks all seven cases against the current source catalog, re-derives the numeric source comparisons and rejects altered approval flags. This checks record consistency, not screenshot authenticity.

Choose an exported checklist to recheck. Nothing is uploaded.

Source & licence policy

Data in this preview is limited to small published numerical facts and independently implemented rules. The ERDB data parser is MIT-licensed code, but that license does not automatically authorize redistribution of all data/assets extracted from the game. A different comprehensive public calculator repository offers no license declaration; no code, images, or tables have been copied from it.

For reference-labelled v1.17 AR, require traceable data source and rights assessment, a pinned version, repeatable same-input benchmarks against Tarnished.dev for base/DLC weapons, upgrades, affinities and 1H/2H, plus explicit disclosure of discrepancies. Game-client screenshots are needed only for claims of independently game-verified current-patch accuracy; reference calculations may be published without them after these benchmark tests pass. Enemy damage/PvP claims need their own independent benchmarks.

Additional cross-checks: AR, armor, spells and enemy source integrity

The following counts reflect separate evidence tiers. They do not certify real PvE damage, full game effects or the current Regulation 1.17.3 client.

A separate ten-spell Regulation 1.17.3 published-source comparison identified two outdated legacy FP entries: Black Flame 15 → 18 and Ranni's Dark Moon 62 → 57. Terra Magicus is the older name for Terra Magica. No current-version spell database is automatically inferred from this small sample; older source FP values are not certified. See 10 current-reference spell checks. Missing game combat parameters remain UNKNOWN. Do not substitute FP cost for spell scaling, text effect descriptions for numeric talisman modifiers or boss HP text for enemy defense. This source-only inspection is not permission to redistribute copyrighted assets or current-game certification.

30 direct AR comparisons · 568 armor source records · Spell, talisman and enemy source audit