Verification log

September 2026

1. 2026-09-07

   Pass 8 — UI/experience confirmation round (presentation, navigation, and gift paths)

   • Scope: presentation, navigation, and copy only. No prices, renewals, limits, or verified_at values changed — every verified_at remains 2026-09-06. The Z.ai GLM-5.3-Flash row's caveat wording now states both blends explicitly ($0.119 at the live promo, $0.2375 at list); the row's underlying promo/list figures are unchanged.
   • Consistency fix: the break-even calculator's flat-plan price is now an input defaulting to the actual Kimi Allegretto tier ($39/mo, crossover ≈49), matching the homepage and crossover chart; the $40/50-task figure survives only where it is captioned as the illustrative reference basket.
   • Mobile AA-score attribution: leaderboard and frontier mobile cards now carry source + access date with every quoted Intelligence Index figure, per the citation policy.
   • Header compacted to a 56px sticky bar (wordmark, nav, CTA) with a <dialog> hamburger menu on mobile; filter chips and small controls raised to touch targets; sticky table headers now clear the site header.
   • Gift path: new /guides/buying-ai-access-as-a-gift/ guide, gift-readiness blocks on the tracked-offer pages and the offers hub, a buying-as-a-gift block on the providers hub, and a plain-English entry card at the top of the homepage.
   • New /cost-calculator/ page gathers the calculators: break-even (with ?tasks/&tokens/&seats/&sub= deep links), cost of one task per route, monthly-vs-annual prepay math, and a per-12-months comparison for tracked offers.
2. 2026-09-07

   Pass 7 — coverage-gap merge, citation drill-down, team-size math

   • Scope: same-day evidence merge from the provider-universe gap pass. No offer data, prices, renewals, limits, or verified_at values changed — every verified_at remains 2026-09-06. Universe snapshot stays 2026-09-07.
   • Universe: 121 → 129 access routes (39 subscriptions, 48 API price rows, 10 credit/prepaid systems, 18 coding-tool plans, 14 free tiers/promos; 88 DIRECT, 35 EXCERPT, 6 UNCERTAIN). Closed this pass: ChatGPT consumer tiers (Plus $19.99, Go $8.00, Pro 5x $100 / 20x $200, credit packs), Claude Max $100/$200 split, SuperGrok Lite $10 / Heavy $300, Cursor Pro+ $60 / Ultra $200 / Teams Premium $120, Copilot Business $19 / Enterprise $39 per seat, DeepSeek peak/off-peak hours, Kimi K2.7 Code / K2.6 API prices, Qwen Model Studio list prices, Groq console list prices, Devin post-merge ladder (via official-page snapshot), Jules task tiers, and GPT-6 Astra plan availability.
   • Still open — shown as “not fetched”, never estimated: GLM Coding Pro/Max prices, Llama 4 pricing, Fireworks Fire Pass price/duration, ChatGPT Business per-seat dollar, and third-party-resale sightings. ChatGPT Free carries no standalone row (unverified this pass).
   • Citation drill-down: every benchmark citation now links its exact AA source row (the gpt-5-mini stale-leaderboard link now points at its verified AA model page); each provider page gains a cited-scores section plus a Sources & methods table listing every datum's source and access date; the methodology page gains a benchmark-provenance section (v4.3 suite, effort variants, cite-vs-link, consent gate). No AA table is reproduced.
   • Break-even calculator: new team-size input (1–50 seats, shareable via ?seats=) multiplies verified per-seat plan math and recommends the cheapest multi-seat-compliant setup; solo behavior is unchanged.
3. 2026-09-07

   Pass 6 — cost-intelligence expansion (universe, leaderboard, frontier chart, providers)

   • Scope: new data layer added on top of the three tracked offers. No offer data, prices, renewals, limits, or verified_at values changed — every verified_at remains 2026-09-06.
   • Universe snapshot published: 121 access routes across 35 subscriptions, 46 API price rows, 10 credit/prepaid systems, 16 coding-tool plans, and 14 free tiers/promos — every row carrying its source URL, access date, and evidence label (70 DIRECT, 35 EXCERPT, 16 UNCERTAIN). Unverified routes are shown as unverified, never priced from memory.
   • Homepage rebuilt as a cost leaderboard: an interactive cost-vs-intelligence frontier chart, an API cost table ranked on blended $/M = (3 x input + 1 x output) / 4 at list price, and a full-universe filter table including UNCERTAIN rows.
   • Artificial Analysis Intelligence Index v4.3 scores added as quoted reference values: one citation per datum, each linking its exact source row, accessed 2026-09-07. No AA table is reproduced; machine feeds deliberately exclude AA scores entirely. Our own weighted ranking (TPVS) is implemented and documented but not rendered pending written consent from AA.
   • New sections: /providers/ index and 24 provider pages cross-linked with the leaderboard; /api/leaderboard.json machine feed (cost side only); llms.txt and llms-full.txt extended with the leaderboard and universe; sitemap extended with per-provider URLs.
   • Known gaps recorded on the methodology page: ChatGPT subscription prices, Claude Max 20x split, GLM Coding Pro/Max prices, Qwen first-party API prices, Llama 4 pricing, Groq per-model prices, Devin/Windsurf post-merge pricing, and Kimi K2.7/K2.6 API prices were not verifiable this pass.
4. 2026-09-07

   Pass 5 — data-platform reorientation (tone and layout only)

   • Scope: presentation-only. No offer data, prices, renewals, limits, or verified_at values were changed in this pass — every verified_at remains 2026-09-06.
   • Homepage rebuilt around the comparison table (offer / price / annual effective / est. $ per 100 tasks / key limit / renewal behavior / verified date); the break-even calculator moved below it as a compact card; the publication log demoted to a single verification line linking here.
   • Copy pass: headlines, verdict labels, badge text, and call-to-action labels rewritten in neutral documentation style; decorative full-bleed bands, tinted route color-coding, and on-page anchor bars removed.
   • Motion: scroll-linked bar and tally animations, hover-lift cards, and stamp animation removed; verification stamps are static.
   • Offer pages restructured to lead with a spec table (tiers, price, annual effective, estimated cost per task, limits, renewal, verified date), followed by the caveats block, summary, and details; page titles made descriptive.
5. 2026-09-07

   Pass 4 — presentation polish

   • Scope: presentation-only. No offer data, prices, renewals, limits, or verdicts were changed in this pass — every verified_at remains 2026-09-06.
   • Offer cards: raster thumbnails replaced with inline ledger mini-charts (same evidence-table language); offers with a Changed entry in the publication log carried a visible accent tick from that pass on.
   • Reader-trust copy: footer disclosure reordered (no affiliate links at that pass, commissions only possibly future); a no-gifting line added on the offers hub; the free-route boundary hedge moved above the fold on the homepage.
   • Navigation and motion: sticky anchor bars gained active-section highlighting; verification stamps given a short reduced-motion-guarded fade; the hero calculator deep-links its computed answer (with your URL inputs) to the matching verdict section.
   • Copy repairs: two unexplained references removed or given context (a promo-window aside on the monthly-vs-annual guide; a fleet-allowance aside on the cost-per-task guide). Plain non-link styling applied to in-product access routes on the free-route pages.
6. 2026-09-06

   Pass 3 — link integrity, canonical scheme, and identity rebuild

   • Kimi tier ladder ($19 / $39 / $99 / $199 monthly; ~$15 / $31 / $79 / $159 annual effective) re-checked against the staged Sep 6 snapshot data — values unchanged. The live pricing page responds but renders figures client-side, so no new external confirmation of the numbers is claimed beyond the original snapshot.
   • Terms link corrected: api.kimi.com/coding/ was verified serving a bare JSON greeting (stale as a terms page) and the offer points at the official membership pricing page, reached from the kimi.ai help center — both URLs fetched and status-checked this pass.
   • Canonical scheme recorded: this site uses trailing-slash URLs everywhere — next.config.ts sets trailingSlash, canonical() appends the slash, and the sitemap, feeds, and breadcrumbs all emit slashed URLs. Non-slash requests redirect; treat the slashed form as canonical when citing.
   • Visual identity rebuilt to the round-3 ledger spec (self-hosted type system, token ramp); no offer data, prices, or verdicts changed in this pass.
7. 2026-09-06

   Pass 2 — fix-round re-verification and hardening

   • All three offer data files re-read and re-validated against the schema checks; verified_at values retained at 2026-09-06 — no external re-check claimed beyond the same-day snapshot.
   • Offer pages rebuilt verdict-first; verification dates removed from title tags and moved to badges, lastmod, and snapshot lines (evergreen titles).
   • Feed switched to active offers; robots.txt, sitemap, and ItemList structured data regenerated and spot-checked by serving the build.
   • FAQ expanded from reader questions on three pages; mock community-data band and mock signup removed — nothing mocked remains.
   • Verification log created (this page) so every future pass is dated and auditable.
8. 2026-09-06

   Pass 1 — initial full snapshot

   • Kimi K3 membership: tier prices $19–$199/mo, annual effective ≈$15/$31/$79/$159, shared credit pool with 5-hour and weekly controls — captured from the official coding-endpoint page.
   • Zen Muse Spark 1.3 Contributor Free: $0 input/cache/output and the exact-string access path (/connect → Zen → /models) — captured from the in-product listing; promo end date not published, stated as such.
   • NVIDIA Build Kimi K3: free development/prototyping terms with reasoning and tool calls preserved, limits account-variable — captured from the official catalog listing.
   • Methodology v0.1 published; machine feeds (llms.txt, llms-full.txt, offers.json, feed.xml) brought live; cost-per-task reference model ($40 basket, $0.80/task, 50-task break-even) documented.