BigAIArena.com · Content ledger
Content Log
Every time content on bigaiarena.com changes, this ledger adds an update. Newest on top. Old rows are never deleted or edited — mistakes get a correction row. Only changes that are live on the site are recorded.
This log covers website content. Changes to the scoring rules are on the Changelog. Sealed match records are never rewritten: when a page changes text that was used as a Fact in a Comparison, the row below names the Comparison, so anyone can check the match against the original record.
Update 2026-10-25 — Protocol 2.7 and AI Audit lists
| Date | Page | Action | What changed | Recorded by |
|---|---|---|---|---|
| 2026-10-25 | /protocol/ · /changelog/ · /scoring-system/ | Protocol 2.7 | A fixed page for every question: Q1 homepage, Q2 data page, Q3, Q5, Q7 pillar page, Q4, Q6, Q8 AI Audit page. Two Access and Blind packages prepared in parallel; the package check picks the sharper one. Gate 3: only a real fact from the same website earns 10; outside content and blank refusals are capped at 5. Applies from Comparison #104. | Master |
| 2026-10-25 | /roles-rotation/ | Changed | Access and Blind: one parallel package added; Package check picks the sharper package; “Part of Protocol 2.6” changed to 2.7. | Master |
| 2026-10-25 | /ai-audit/ | Changed | Question tables now show two lists: “AI got right — introduced / cited to users” and “AI got wrong / missed — debate that sends users to verify”; the “Source fixed?” column is hidden (issues stay in the issues table). Text of the Latest Comparison block no longer faded. No sealed result changed. | Master |
Update 2026-10-24 — Protocol 2.6 wording across the site; Round 4 case
| Date | Page | Action | What changed | Recorded by |
|---|---|---|---|---|
| 2026-10-24 | /cases/ · / | Added | Round 4 case: Comparison #C095 on realdatasethub.com, Grok 100/100 and Gemini 15/100 on the same pages. The home page now points to it as the latest case. The Round 1–3 cases are unchanged. | Master |
| 2026-10-24 | /methodology/ | Corrected | Checkpoint 1. Old: “Two independent accounts confirm the live pages and draft the facts; the Arena Master’s quality gate picks one.” New: one AI drafts the package, a separate account checks every line against the live pages, and the Arena Master approves it (Protocol 2.6). | Master |
| 2026-10-24 | /scoring-system/ | Changed | Added the three pillar questions of Protocol 2.6 (Q3, Q5, Q7) and the label “self-reported tool failure (cap 5)”. Scoring rules unchanged. | Master |
| 2026-10-24 | /glossary/ | Added | New terms: Access and Blind, Package check, Standby AI, IP Pillars, Issues table, Self-reported tool failure (cap 5). | Master |
| 2026-10-24 | /enterprise-data/ | Changed | “Full API plan” renamed “Full Data License”, matching RealDatasetHub.com; delivery stated as JSON/CSV snapshots, with no programmatic API yet. | Master |
| 2026-10-24 | 5 pages above | Changed | Footer links: IP Pillars added. | Master |
Update 2026-10-23 — IP Pillars page and website Comparisons
| Date | Page | Action | What changed | Recorded by |
|---|---|---|---|---|
| 2026-10-23 | /ip-pillars/ | Added | New page: three pillars of what a Comparison does for a website — whether AI cite it correctly, what AI gets wrong on the public record, and what the website must fix. Every result is taken from sealed Comparisons #095 and #089. Added to the menu as “IP Pillars”, before RealDataset. | Master |
| 2026-10-23 | / | Changed | Pillars block added under the opening buttons. “How one Comparison works”, step 2, updated to Protocol 2.6: one Access and Blind AI, with every line of its package checked before the Comparison starts. Footer links: IP Pillars added. | Master |
| 2026-10-23 | /submit-comparison/ | Changed | Added “For website owners”: the same Comparison run on an owner’s website, at the same $2,000 flat rate and timing. The offer for AI teams is unchanged. Footer links: IP Pillars added. | Master |
Update 2026-10-22 — Protocol 2.6 and AI Audit layout
| Date | Page | Action | What changed | Recorded by |
|---|---|---|---|---|
| 2026-10-22 | /protocol/ · /changelog/ | Protocol 2.6 | No new scoring rules. One Access and Blind AI plus one Standby AI; every package checked line by line before use; three questions (Q3, Q5, Q7) test each website’s published pillars; Full Check may propose Protocol changes and adds a short comment; a worked example of a question that gives away its answer; the self-reported tool failure score labelled as such. Applies from Comparison #101. | Master |
| 2026-10-22 | /roles-rotation/ | Changed | Roles table: Access and Blind 1 seat, Standby AI and Package check added; “Part of Protocol 2.5” changed to 2.6. | Master |
| 2026-10-22 | /ai-audit/ | Changed | New layout: earlier Comparisons one line each with an open/close button (full text still in the page), the latest Comparison in full, then the issues each Comparison found and their status, then the benefits block. No sealed result changed. | Master |
Update 2026-10-18 — Protocol 2.5
| Date | Page | Action | What changed | Recorded by |
|---|---|---|---|---|
| 2026-10-18 | /protocol/ · /changelog/ | Protocol 2.5 | No new scoring rules; six clarifications from Comparisons #092–#098: the final check settles minor issues itself and records each one; a worked Garbage Substitution example; existing Question 1–2 and citation limits restated in every scoring card; score sheets are re-added before the results summary; questions never contain their own answers; a test is finalised only with both transcripts complete. Applies from Comparison #099. | Master |
| 2026-10-18 | /roles-rotation/ | Changed | “Part of Protocol 2.4” changed to 2.5. | Master |
Update 2026-10-15 — RealDataset built on the server; open door for AI assistants
| Date | Page | Action | What changed | Recorded by |
|---|---|---|---|---|
| 2026-10-15 | /realdataset/ | Changed | The hand-copied table and its counting script were removed and replaced by a 12-column table built on the server from the AIAR_DATA_MASTER Google Sheet. It can be read without JavaScript (the search box only filters it). Record cells are shown in the original Vietnamese; the English translation previously shown on the page is kept below the table, unchanged. The line under the H1 (“N entry · first date to last date · Checked … +07”) and the Meta_Pattern counts are read from the sheet. Entry count unchanged: 3 (AIAR-MP01-MS-20260518-001, AIAR-MP04-OB-20260605-001, AIAR-MP04-OB-20260605-002). Typos fixed in the sheet: “Sào rác” → “Sáo rác” (4 cells), “BigAIAArena.com” → “BigAIArena.com” (1 cell); content unchanged. | Master |
| 2026-10-15 | /robots.txt | Changed | AI assistants and AI search engines are now welcomed by name (from 5 to 20 named bots). Only /wp-admin/ is blocked; the Sitemap line is kept. | Master |
| 2026-10-15 | /llms.txt | Added | A short guide for AI systems: the main pages, RealDataset, AI Audit and the Content Log. Page content and data unchanged. | Master |
Update 2026-10-14 — Protocol 2.4
| Date | Page | Action | What changed | Recorded by |
|---|---|---|---|---|
| 2026-10-14 | /protocol/ · /changelog/ | Protocol 2.4 | Question 2 partial credit clarified (saw the real data table but answered only part of the question = 5); real facts from another page of the same website count as Garbage Substitution in Questions 3–6; only real data rows are counted when preparing a test; footer partner text is never a test fact. Applies from Comparison #091. | Master |
| 2026-10-14 | /scoring-system/ · /roles-rotation/ | Changed | Question 2 partial-credit row updated to Protocol 2.4; “Part of Protocol 2.3” changed to 2.4. | Master |
Update 2026-10-13 (evening) — tables built on the server
| Date | Page | Action | What changed | Recorded by |
|---|---|---|---|---|
| 2026-10-13 | /comparisons/ | Changed | The hand-copied “Static copy of the ledger (86 Comparisons, as of 2026-10-12)” was removed and replaced by a table built on the server from the public results ledger. It lists every sealed Comparison (#C001 to the latest) with a “Checked … +07” time, and can be read without JavaScript. The live table at the top of the page is unchanged. | Master |
| 2026-10-13 | / | Changed | The Comparison count, the current Round, “N of 28 Comparisons run” and the Latest results table are now read from the results ledger on the server; “Counts as of 2026-10-12” replaced by “Checked … +07”. | Master |
| 2026-10-13 | /rankings/ | Changed | The Round 1–4 and All-Time tables are now built on the server from the results ledger, with Access, Trust and Citation averages for every Round (the snapshot of 2026-10-12 showed Totals only for Rounds 1–3). Same calculation as before; figures checked cell by cell against the ledger. A new Round now gets its tab automatically. | Master |
| 2026-10-13 | /ai-audit/ | Changed | The hand-copied summary and question-by-question tables were replaced by tables built on the server from the BigAIArena AI Audit sheet, with source-file and video links. Sealed match records are unchanged. | Master |
Update 2026-10-12 (evening) — Protocol 2.3
| Date | Page | Action | What changed | Recorded by |
|---|---|---|---|---|
| 2026-10-12 | /protocol/ · /changelog/ | Protocol 2.3 | Fresh-data questions Q4 and Q8 are standard on every website, Core IP or Partner; the on/off switch and the one-way Partner rule of 2.2 removed; both questions name the AI Audit page openly; the Q4 fact is copied from the page as published, with the time it was read; Respondent accounts are never used for any other role. Applies from Comparison #087. | Master |
| 2026-10-12 | /scoring-system/ · /roles-rotation/ | Changed | Fresh-data paragraph updated to Protocol 2.3; “Part of Protocol 2.2” changed to 2.3. | Master |
Update 2026-10-12 (afternoon) — Rankings by Round; the two fresh-data questions; role and account wording
| Date | Page | Action | What changed | Recorded by |
|---|---|---|---|---|
| 2026-10-12 | /rankings/ | Changed | The static copy under the live table now has a Round selector. Rounds 1–3 (complete, 28 Comparisons each) show each AI’s average Total; Round 4 (current) shows Access / Trust / Citation / Total / Trend. All figures come from the ledger on /comparisons/. The live table is unchanged. | Master |
| 2026-10-12 | /scoring-system/ | Changed | The question table now names the two fresh-data questions of Protocol 2.2: Q4, a fact from the latest sealed result on the website’s own /ai-audit/ page, and Q8, a citation of that page. | Master |
| 2026-10-12 | /glossary/ | Changed | “AI Audit page” definition: added that it is the source of Q4 and Q8 since Protocol 2.2. | Master |
| 2026-10-12 | /rankings/ | Changed | Trend column removed from the Round 4 static table. Scores move up and down between Comparisons, so a trend would need extra explanation of how many Comparisons it covers; each Round now shows averages only. | Master |
| 2026-10-12 | /protocol/ | Corrected | Layer 1, rule 3. Old: “Each role is held by exactly one account per Comparison. No AI holds two roles in the same Comparison.” New: one account per role and no account holds two roles; the eight working seats go to eight different AI; the same AI may appear again only on a different account (Independent Audit, Full Check, labelled substitute); a Respondent answers from its own rotation account, never used for any other role. Wording now matches how Comparisons have run; scoring rules unchanged. | Master |
| 2026-10-12 | /roles-rotation/ | Changed | Account rotation: added that Respondent accounts are kept apart from every other role. | Master |
| 2026-10-12 | /rankings/ | Changed | The live table and the static copy are merged into one table with one row of tabs: Round 1, Round 2, Round 3, Round 4 (current) and All-Time. Every tab shows Comparisons, Access, Trust, Citation and Total, read live from the public ledger; a snapshot of 2026-10-12 stays in the page for readers that do not run JavaScript. Trend and “Biggest mover” removed. | Master |
| 2026-10-12 | /comparisons/ | Changed | “View →” now opens only the sealed record on Google Drive; a row without a record shows no link instead of a page that did not exist. The live script now runs only on this page. | Master |
| 2026-10-12 | /comparison-detail-dynamic/ · /hello-world/ | Removed | /comparison-detail-dynamic/ was an unused template full of placeholders; removed, with a 301 to /comparisons/. The default WordPress post /hello-world/ removed. Both are gone from the sitemap. | Master |
Update 2026-10-12 — BigAIArena.com rebuilt to Ver 2.2
| Date | Page | Action | What changed | Recorded by |
|---|---|---|---|---|
| 2026-10-12 | Whole site | Changed | Site rebuilt to the BrainCrisis Eco Ver 2.2 standard (Source Serif 4 / Source Sans 3, one H1 per page, tables scroll on phones). Existing URLs kept. The old footer (three link columns, “© BigAIArena.com — Access 20 · Trust 60 · Citation 20 = 100”, support button) replaced by the standard BrainCrisis Eco footer; the site links moved to a navigation block at the end of each page. Single contact: Master@BrainCrisis.com · +84 785 179 035. | Master |
| 2026-10-12 | Main menu | Changed | Menu was RANKINGS, CASES, METHODOLOGY, SUBMIT COMPARISON, ENTERPRISE DATA. Now: Rankings, Comparisons, Cases, Methodology, RealDataset, AI Audit, Submit Comparison, Enterprise Data, Log. Fact in #054 (Q1); match record unchanged. | Master |
| 2026-10-12 | / | Changed | The “This Round’s Rankings” block ([AI 1] / [score] / Round [N]), the case teaser “Where did [AI name] get it wrong?”, and the Share, Discuss and Subscribe buttons with no destination were template placeholders. Replaced by real content: counts as of the date shown, the latest results from the ledger, how a Comparison works, and a link to the cases. Live scores stay only on /rankings/. H1 “Which AI can you actually trust this week?” and the 20/60/20 formula kept. Facts in #018 (Q1), #054 (Q1) and #058 (Q1, Q5); match records unchanged. | Master |
| 2026-10-12 | /realdataset/ | Added | New page with the 3 AIAR records from the AIAR_DATA_MASTER sheet (2026-05-18 to 2026-06-05), in English translation, with a link to the original Vietnamese sheet. The URL returned a static 404 before. Fact in #054 (Q2) and #058 (Q2, Q6); match records unchanged. | Master |
| 2026-10-12 | /ai-audit/ | Added | New page: the 3 Comparisons that tested bigaiarena.com (#018 DeepSeek 35 – Claude 10; #054 Meta AI 60 – Claude 20; #058 Meta AI 80 – Claude 20), question by question, as static HTML plus the AI_Audit Google Sheet. | Master |
| 2026-10-12 | /log/ | Added | New page: this content ledger. | Master |
| 2026-10-12 | /comparisons/ | Added | Static copy of the ledger (86 Comparisons as of 2026-10-12) added below the live table, because the live list loads only with JavaScript and tools without it saw “Loading results…”. Fact in #018 (Q2); match record unchanged. In this copy #C074 is labelled “Grok vs Copilot” (the Sheet label “Grok Acc04” broke the rule that public reports name the AI only), and #C001 and #C005 show “Sealed”: every Comparison is Sealed after Full Check; the label “Verified” on those two rows was a leftover and clashed with the 80–89 status band. | Master |
| 2026-10-12 | /rankings/ | Added | Static copy of This Round’s table (Round 4, 2026-10-12) added below the live table, for the same reason. The live table remains the reference. | Master |
| 2026-10-12 | /protocol/ | Changed | The page still showed “Version 1.6” while the Changelog listed 1.7 to 2.2. Updated to Version 2.2 with the rules from 1.7 to 2.2: fixed ground truth at the opening check, substitution labels, account numbers only in the final table, the platform-limit routine, the stricter Citation table (homepage-only and “nothing found” links score 0; a specific but wrong subpage scores 5), Q2 data page per site, and the AI-Audit questions Q4 / Q8. Facts in #054 (Q4) and #058 (Q4, Q8); match records unchanged. | Master |
| 2026-10-12 | /scoring-system/ | Corrected | Removed the sentence “The link counts regardless of purpose — an AI citing the correct URL to say ‘I couldn’t confirm this here’ scores the same as citing it to confirm.” Protocol 2.0 (2026-10-05) had already withdrawn this rule. The Citation table now matches the Protocol: the homepage alone scores 0, a specific but wrong subpage scores 5. Fact in #018 (Q5, Q7) for the gate scores, which are unchanged; match record unchanged. | Master |
| 2026-10-12 | /roles-rotation/ | Corrected | The Access and Blind role said it drafts “Questions 1–2, 3 raw Blind Spot facts and Questions 9–10”, and another section said “10-item Blind Spot List”; the Protocol uses 4 Blind Spot facts and citation targets for Q7–Q8. Corrected. The heading “The 8 Roles” listed 10 seats; now explains 8 working seats filled by the 8 AI plus 2 verification seats (Independent Audit, Full Check) on separate accounts. “Version 12.5” removed (internal number; the public Protocol is 2.2). The data-pull line with the placeholder “[to be filled — first day of each quarter]” removed; the roster source (Similarweb, locked each quarter) kept. Pace: kept “about 4 Comparisons a day, roughly one week per Round” as the plan and added the actual duration of Rounds 1–3 (about 8 to 17 days). SHA256: the page said the Secretary computes it; corrected — the Arena Master computes the SHA256 hash of each sealed record. Facts in #018 (Q3, Q4, Q8) and #054 (Q6); match records unchanged. | Master |
| 2026-10-12 | /glossary/ | Corrected | Round: “Roughly 1 week” replaced by the planned week plus the actual 8 to 17 days of Rounds 1–3. SHA256 entry corrected: the hash is computed by the Arena Master, not the Secretary. Added “Sealed”: the status of every Comparison after Full Check. Arena Master definition aligned with Who Runs This?. Added Blind Spot, Phase, Examiner / Respondent, Data ID, IP site, Partner IP, RealDataset and AI Audit page. | Master |
| 2026-10-12 | /who-runs-this/ | Corrected | The page said the Arena Master is “involved only in question selection”, while the Protocol and Glossary say the Arena Master also assigns Full Check, handles substitutions and resolves edge cases. The duties are now listed in full, including the SHA256 hash of each sealed record; “1 human step”, Vũ Trường Ca, “0 advisory clients” and the permanent no-advisory policy kept. Facts in #018 (Q6) and #054 (Q5); match records unchanged. | Master |
| 2026-10-12 | /cases/ | Changed | Two template cards (“Where did [AI name] get it wrong?”, “How [AI name] scored a perfect 100”, “[1-2 câu tóm tắt hấp dẫn]”) replaced by 3 real cases, one per Round, from the Comparisons that tested bigaiarena.com: #018 (Round 1), #054 (Round 2), #058 (Round 3). | Master |
| 2026-10-12 | /methodology/ | Changed | “Why we check every week” renamed “Why we keep testing, not once” (Rounds have taken 8 to 17 days). “6 live checkpoints” now lists the six checkpoints by name. Other content kept. | Master |
| 2026-10-12 | /submit-comparison/ | Changed | Removed two unfinished placeholders in Vietnamese: “[chính sách hoàn tiền — cần Master xác nhận cụ thể]” (refund policy) and “[nếu có — cần Master xác nhận cụ thể]” (topic restrictions). The “Submit Your Comparison” button pointed to “#”; it now opens an email to Master@BrainCrisis.com. “Pay $2,000”, “$2,000 per Comparison — one flat rate, no matter your size” and “Payment buys the test, never the score.” kept. Fact in #054 (Q8); match record unchanged. | Master |
| 2026-10-12 | /enterprise-data/ | Corrected | “Every Comparison draws from 18 RealDataset sources” corrected: each Comparison tests one website, an IP site or a disclosed Partner. The split “roughly 80% by volume from BigAIArena’s own Comparisons, 20% from the other 17 IPs” was removed because no published figure supports it; current counts are on RealDatasetHub.com. “Direct API access” replaced by a link to the Full API plan on RealDatasetHub.com. Fact in #054 (Q6); match record unchanged. | Master |
| 2026-10-12 | /ecosystem/ | Changed | Added the number of Comparisons per IP site and for the Partner IP (counted from the ledger, 2026-10-12) and links to each site’s AI Audit page. “Every question originates from one of these 18” now also names the disclosed Partner IP. | Master |
| 2026-10-12 | /changelog/ | Changed | Version 1.6 dated 2026-09-14 (the date the Protocol page showed); “(previous)” for 1.4 and 1.2 replaced by “Date not recorded”. Added a pointer to this Content Log for website changes. | Master |
| 2026-10-12 | /support/ | Changed | The fixed donation amounts ($3, $5, $10) were removed: BrainCrisis Eco does not publish donation amounts. Text rewritten: support never buys a score. | Master |
| 2026-10-12 | /newsletter/ | Removed | Page “This week on BigAIArena” removed: its Subscribe button had no email field, the sample issue held placeholders such as [AI], and no newsletter is sent. The URL redirects (301) to /cases/, where one case per Round is published. | Master |
| 2026-10-12 | Plugins | Removed | WP ULike (like/vote buttons) and TablePress (all tables are now static HTML) removed. | Master |
| 2026-10-12 | /robots.txt | Changed | Added “User-agent: *”, “Allow: /” and the sitemap line, so every AI crawler is allowed. In #054 and #058 Claude reported being blocked by robots rules; match records unchanged. | Master |
Earlier history, 2026-08-31 to 2026-10-07 — Protocol versions 1.0 to 2.2
Before this Content Log opened on 2026-10-12, changes were recorded only as Protocol versions. They are summarised here so the full history sits in one place; the original wording of each version is on the Changelog. Website content changes before 2026-10-12 were not logged separately.
| Date | Page | Action | What changed | Recorded by |
|---|---|---|---|---|
| 2026-10-07 | /protocol/ | Protocol 2.2 | AI-Audit questions: Q4 asks about the latest sealed result on the IP’s /ai-audit/ page and Q8 cites that page, when the page is readable without JavaScript. One-way cross-links: a Partner IP may ask about a Core IP’s AI-Audit page, never the reverse. Q2 checks each IP’s main data page (/realdataset/; /datasets/ on RealDatasetHub.com). | Master |
| 2026-10-05 | /protocol/ | Protocol 2.1 | The closing access check may be done by the original AI or a labelled substitute. Reports name the AI only; account numbers appear only in the final results table. Platform-limit routine: continue on another account of the same AI; the repeated Q1 is never scored. | Master |
| 2026-10-05 | /protocol/ | Protocol 2.0 | Removed the sentence that a link earns points “regardless of whether it was used to confirm a fact”. Citation is now a two-part check: link written correctly and content actually cited; a correct link used only to say nothing was found scores 0. Four boundaries spelled out: total access failure = 0; false statement = Hallucination; 3 gates only on Blind Spots; vague industry knowledge does not reach the Garbage Substitution tier. Role cards name the default account. | Master |
| 2026-09-23 | /protocol/ | Protocol 1.9 | Denying a fact on a page the AI reached is judged on its merits, not as an honest “could not access”. A generic homepage link scores 0, distinct from a specific but wrong page. A mislabelled Review is checked against both real answers before it is discarded. | Master |
| 2026-09-21 | /protocol/ | Protocol 1.8 | A correct link attached to wrong text earns no Citation credit. Facts recorded at the opening check are fixed for the whole Comparison. The “wrapped link” rule narrowed so it no longer catches ordinary same-site links. A Reviewer’s own math error is excluded and explained, not averaged in. | Master |
| 2026-09-17 | /protocol/ | Protocol 1.7 | A substitute AI must label its own content as a substitution at the top of what it writes. | Master |
| 2026-09-14 | /protocol/ | Protocol 1.6 | An honest “I couldn’t verify this” caps at the minimum score for that question. A URL to the right site but wrong page keeps mid-tier credit. No AI may insert an audit block for itself. | Master |
| 2026-09-06 | /protocol/ | Protocol 1.5 | The Secretary produces 2 separate first-tier tables, one per Phase, and states the reason whenever a source is excluded. | Master |
| Not recorded | /protocol/ | Protocol 1.4 | Each scoring block ends with a large Access / Trust / Citation / Total line; the Comparison header lists each role on its own line. | Master |
| 2026-09-05 | /protocol/ | Protocol 1.3 | 9 clarifications from #003–#008: Independent Audit reports verified facts only; a citation must match the exact concept and location asked; a URL counts only when attached to content already cited; every substitution needs its own content block. | Master |
| Not recorded | /protocol/ | Protocol 1.2 | Access Gate “0 for total access failure” separated from the Trust rule “self-reported tool failure = 5”; a markdown link is scored by its real target. | Master |
| 2026-09-01 | /protocol/ | Protocol 1.1 | Falsely claiming to have checked the source directly counts as Hallucination, even when the content is accurate. | Master |
| 2026-08-31 | /protocol/ | Protocol 1.0 | Official public launch: 8 questions, Access / Trust / Citation, the 3-gate Blind Spot test, Independent Audit on every Comparison and Full Check as the only official record. | Master |