What changed, when it changed, and the source receipt behind it.
The report leads with one card: grade, yesterday's floor served by methodology 1.7.0, observed sales range and bid status. Range pending applies only to a bid-positive engine result held solely by the missing band. Named risks and data limitations keep their verdict and concern. The cert prints the same served number, with one methodology disclosure on the desk. No scoring or pricing rules change.
The Why page replaces ten placeholder design notes with eight figures drawn from the sample report and the canon constants. Illustrative figures are labeled as such, closed gates render closed, and the external review stage now shows its sign-off date. No scoring or pricing rules change.
The report separates the previous-day ask anchor from the legacy price and explains why bid clearance is held when no band is available. Long price explanations wrap on phones. The scoring and pricing rules are unchanged.
This candidate limits concurrent report enrichment, queued refreshes, lease cleanup and sanctions-freshness reads. Under saturation a report can remain incomplete and screening can stay unknown; refresh notices say requested rather than promising execution. These safeguards do not establish the cause of past memory incidents or claim that those incidents are resolved.
Trait pages now answer how many apes carry a trait and what its screened sales look like, with a combo filter for co-occurring traits, trait search suggestions in the top bar, a Traits tab on the collection desk, and a share card. Every figure is a rollup; a missing rollup shows a dash, never a fabricated number.
Earned account badges retain their achievement and show a historical earned date when the available evidence proves it. Badges without date proof stay earned without a date. Better history can move a date earlier, never later. Earning rules and sample accounts are unchanged. Deployment waits for the separately approved database migration.
Corrected displayed facts a three-pass audit had wrong on non-legal surfaces: the sample certificate (Ape #4521, score 100, LEGENDARY), BAYC/MAYC/BAKC supply counts, founder terms (20 credits — 1–5 mint free once minting opens, closed; 6–20 data-only; 365 days of active time, paused time extends), the account panel no longer offering the closed pass, and audit status (signed off). Prices and availability unchanged.
Connected wallets without a signed-in session no longer show sample account details. Sign-in is offered once more when you return to the tab, unless you rejected it. A stalled prompt times out with a retry message. Links on www redirect to apefax.com. Anonymous production visitors keep the connect gate.
Under review: daily evidence checks count the prior 30 days and the cutoff day exactly, with a timeout for each query. Older history keeps an estimated total for diagnostics. Loss inside the checked window remains a failure; loss further back that leaves the oldest day intact is outside this check. This entry does not claim the production incident is resolved.
Information pages and the updates feed now request five-minute shared-cache windows instead of long-lived cached copies. After deployment, a check rejects responses older than six minutes. Grades, prices and methodology are unchanged.
Between about 01:47 and 02:25 UTC on October 9 (9:47 to 10:25 PM ET on launch night) apefax.com stopped responding intermittently; pages hung instead of erroring. The database was healthy and nothing was deployed. Our best read: the web service ran out of headroom on a shared-CPU instance. At 02:52 UTC it moved to a dedicated 1 vCPU, 4 GB instance with nothing else changed; stable since. No payments were affected.
Entry 70 said the methodology 1.7.0 number would serve from October 8, 2026. It will not serve on October 8. The site opens October 8 with the number withheld on all surfaces; each surface prints the anchor definition only. The number is not served yet. Nothing about the method changed: lowest listing on the previous UTC day, from the independent ask series.
One line under the band sentence on the transparency page shows ask-floor days collected toward the 90 the band needs and since which day, days with both ask sources, trait premiums unlocked, the nearest key toward its 15-residual unlock, and the last data run. Missing values print a bracketed phrase. Live from the week of October 5; this notice publishes with the launch notes on October 8, 2026.
The number is scheduled to serve from October 8, 2026. Lowest listing on the previous UTC day, from the independent ask series. Not the highest offer. Not today's live floor. 43 of 241 holdout sales priced, 198 had no listing · MdAPE 4.39%, bias +0.33% on those 43. Our two replications: a Punks-style trait-floor replication and an unshrunk hedonic in the Mr Deedz style. We do not claim a beat over OpenSea's top bid.
Superseded by: Correction: the previous-day lowest-listing number does not serve on October 8
Forensic coverage rests on receipt evidence, not a transfer-count rule. A receipt-verified real sale counts as coverage even when outlier-priced (outliers stay excluded from fair value). Wash, self-trade, private-suspect, unverified-swap and operator prints are excluded. Never-sold is proven by receipts: every transfer a verified non-sale; an unverifiable transfer is a coverage gap naming the transaction and date.
An ape whose current owner — established from chain custody and read fresh — granted approve-for-all to the Limit Break Payment Processor V2 when buying through it between March 1, 2024 and September 7, 2024 now carries a high-severity advisory naming the contract, the window, and the action: revoke it at revoke.cash. It carries no score weight and never claims an ape was stolen, rescued, or at risk.
Shared report links now unfurl with the minted certificate — its public id, methodology edition and mint date — and a preview image served from stored data, never a grade computed on demand. Two honesty fixes ship with it: a report whose score cannot be read now says "could not verify" rather than a grade, and a graded ape is never shown as "Report pending" after a failed read. Scores and methodology are unchanged.
The image-host flag now judges where the collection itself hosts an ape's image, from the image address in the token's own metadata, never the copy ApeFax serves. Until now, ApeFax's own copy of an IPFS-hosted image could make a report say the image was centrally hosted: a false positive on 721 graded apes. Scores are unchanged, the flag carries no weight, and affected reports are re-graded after this update.
Our earlier note about the explore value badges quoted two internal reason codes as visitor copy. In words: when the read behind an explore value badge fails, explore marks its value scan degraded and withholds the badges, and the platform health check and launch monitor go red for it. A missing snapshot for one ape is still a normal blank, not an outage.
The collections strip count grades NFTs in the database, the sales count reads the real transaction column, and the sync cell counts down to the next fifteen-minute ingest. Account and billing counts, watchlist status and enrichment, lending labels, sanctions screens, trait-premium thumbnails, and admin counters now fail closed or disclose a sample.
The family lineage on each report, certificate and desk now names every slot on its own: a mutant, dog or Koda the index has proven is shown; one not yet proven reads NOT VERIFIED with the note that the index is not yet proven complete; one proven absent reads NOT MINTED. No internal codes appear in those slots, and the daily indexer restarts from the first block whenever its own record is incomplete.
Every Bored Ape picture on ApeFax is now served from a copy ApeFax stores itself. Each copy was kept only after its bytes matched the token's content identifier, and today every stored image address was pointed at those copies. Frozen reports and certificates are untouched; they keep the addresses they were issued with.
Our earlier note today said a verified copy of each Bored Ape's picture was already kept and served. That ran ahead of the work. The copies are being made now, each stored only after its bytes match the token's content identifier; stored image addresses are pointed at them in a later step. Until then every picture stays on the working IPFS gateway. Frozen reports and certificates are never rewritten.
HOLDERS and LISTED on the collections desk now lead with how much they changed over the selected window. When the two ends of the window are not comparable, the headline shows a dash, with the current count and the reason beneath it. LISTED counts OpenSea listings; a change is only shown between two complete counts.
We keep a verified copy of each Bored Ape's original picture and point the stored image address at that copy, so the page loads the same bytes the token's content identifier names. Frozen reports and certificates are not rewritten; a new report picks up the new address. Pictures that cannot be verified stay on the working IPFS gateway.
The desk's LISTED figure came from a live OpenSea scan that stopped after a fixed budget, so it was a lower bound. A background job now walks every listing page every fifteen minutes and records a run only when it reached the end; the desk shows that exact count and when it was captured. An unfinished run stays a labeled lower bound. Until the database step is applied, the desk keeps today's partial count.
Most cached ape images still pointed at the retired ipfs.io gateway, which now refuses the BAYC art, so those pictures did not load. The resolver serves images through Pinata with Filebase as backup, and a one-shot backfill rewrites the stored addresses to the new gateway while keeping each image's content identifier unchanged. This makes images load; it does not remove the third-party dependency or fix clarity.
Until now a family report was computed the first time someone opened it, which took twenty seconds or more for almost every token. A background job now builds the reports for the whole collection a batch at a time and keeps them refreshed, so the first visit opens as fast as a repeat visit. While a token is still waiting its turn, the page says so and shows when the report is ready.
A background rollup that stops responding is now reported as stalled after a fixed deadline, so the health page and the monitor show a real failure instead of waiting for it. The fax data feed refreshes its edge cache faster while a report is being rebuilt, the owner time stamp is resolved one way everywhere, and an unused desk data endpoint was removed.
The lineage tool used to recompute a family report from the data providers whenever its cached copy was older than fifteen minutes, which took over twenty seconds. It now shows the cached report at once, stamped with the time it was computed and a refreshing state, and recomputes it right after the page is served. Certificates and minting never use the lineage tool cache.
A fax page not viewed for an hour used to rebuild its sales and transfer history from the data providers before showing anything, often twenty seconds or more. It now shows the last verified report at once, stamps each market figure with its capture time and a refreshing state, and rebuilds the history after the page is served. The owner is re-checked on every view; minting always uses a fresh rebuild.
The sales-ledger health check now decides readiness from the ledger cron’s own recent runs. If the separate launch monitor goes quiet, the health page notes that as a monitoring gap instead of marking the ledger not ready. A database-side heartbeat is added so the monitor signal no longer depends on a single scheduler.
Every wallet check and every certificate runs through a sanctions list we sync daily. If that sync failed, certificates were refused while our health page still read green. The sync's freshness and integrity are now a first-class readiness signal: a stale or unverified list turns the health page red, the launch monitor stops, and a public probe at /api/health/ofac shows the age and the cause, never the data.
Switching the collection desk between 30, 60 and 90 days now moves the holders, floor and market-cap cards too. Holders show the change from the daily holder scan against its own count. Floor and market cap show the change in the sales-based floor proxy, labelled as such, until a daily snapshot of the live floor has enough history. Listed count starts a daily snapshot and shows its change once a baseline exists.
While the sales-ledger cron is being installed, the health check now says the ingest leg is not installed yet instead of reporting a failure, and the hourly ledger writer keeps writing the way it did before the cron existed. Every other ingest problem still reads as a real failure. Nothing changes for readers once the cron is installed.
Sales used to arrive only when GitHub ran our hourly job, and GitHub skipped hours at a time this week, which is why several reports and the daily rollups went stale. The ledger now refreshes on a schedule inside our own database every 15 minutes, with one writer at a time and a record of every run that the health page shows. The hourly job still runs as a second trigger of the same code.
When the read behind the explore value badges fails, explore used to fall back silently to the price proxy. Now the explore response marks its value scan degraded (cause the judgment read failed) and withholds the badges, and the platform health check and launch monitor go red for it (reason the explore judgment read failed). A missing snapshot for one ape is still a normal blank, not an outage.
If GitHub is down at the hour our daily rollup runs, a scheduled job on our own hosting now runs it instead, once, and records what it did. The gate that requires every user-visible change to appear here now checks against the newest published entry, so a change built on an older list is refused rather than passed. Two remaining database reads that used a JSON path selector now fetch the record whole.
The desk's 30/60/90 selector now drives every metric computed per window: volume, median, buyers, the sales tape, wallet flow, top sales, resale pairs, the leaderboard and the windowed forensic checks. Floor, market cap, listings and holders keep their current value and show a move only when matching history exists; otherwise the card says so. Anything not computable for a window is marked unavailable, never zero.
Since 12 September the daily rollup's two loan-screening steps had failed on a database read our data API rejected, and the run still reported success, so sales that were really loan liquidations could slip into the comparable set behind trait premiums and fair value. The read is rewritten and kept small, and a run that loses a required step now reports itself as degraded on the health page instead of green.
Yesterday's picture guard hid real BAYC pictures: OpenSea's image host files a picture under whichever collection uploaded it first, so a genuine ape can sit under a copycat's address. That rule is gone. When the marketplace copy sits under a foreign address we fetch the original from IPFS, and a picture whose identity we have verified is shown, not hidden. The home page sample certificate has its picture again.
In review. Scoring moves to methodology 1.1.8, and each version is now frozen as a self-contained package, so a certificate can be re-checked years from now without ApeFax. One grade change: a never-sold ape can establish its history at any rarity. An ape we could not screen enough still shows its grade but cannot mint until coverage exists. Every report gains a review request: image, sale, flag, grade, or coverage.
In review. Pudgy Penguins and Lil Pudgys are paused from the plan for now; anyone who chose them at sign-up keeps that choice. The collection surfaces now read one registry, so adding the next collection is a single entry. The collections top bar no longer adds up OpenSea totals for collections we do not serve yet: it shows what is live and planned, apes graded, sales our ledger has indexed, and when we last synced.
In review. Some cached images pointed at artwork from other collections. A picture fetched from a marketplace is now confirmed to describe the ape requested, by token number and collection, falling back to the on-chain artwork if not. And where a stored image address clearly names a different collection, we show no picture rather than the wrong one. Repairing the stored addresses follows separately.
In review. The lineage tracer used to require an email address before it would trace anything, and it was the only tool on the site that did. Every lookup now runs immediately. After three completed traces we ask once whether you want the fax when BAYC goes public; "Not now" is remembered, and anyone who already signed up never sees it. What we send when you do sign up, and the cookie we set, are unchanged.
In review. A locked report was listing each past sale with its price, date, marketplace and transaction. That is part of the paid report and was never meant to be public. A locked report now shows how many sales an ape has had and nothing more about them. Paid reports are unchanged, and the free sample reports still show everything.
In review. Every ETH and USD figure on the site now goes through a single formatter, so the symbol, the decimal places, the currency and the sign are decided in one place instead of drifting between surfaces. Two malformed negative amounts are corrected. Every other displayed number renders exactly as before, and that is asserted rather than assumed.
In review. A script error fired as the site loaded and stopped the wallet panel from finishing its start-up, so reconnecting a previously connected wallet could silently fail. The cause was a browser rule about how timers and network calls must be invoked. Reports, scores, flags and fair-value numbers were never affected and are unchanged.
In review. The ape head in the ApeFax mark was not #4521. It is now, and every copy of it was replaced from the collection’s own metadata rather than a lookalike matched by eye. Nothing else changed: no scoring, no pricing, no report content.
In review. The collection directory and fax reports now show the UTC time the collection floor was captured, so a cached figure cannot look newer than it is. The directory regenerates every 30 minutes, so its floor can differ slightly from a live report's — the timestamps make that visible. With no capture time we show a dash, never a substituted one. Floor values and their calculation are unchanged.
In review. Fax reports load faster and ownership data stays fresher on the paid path; entitled holders keep access when a later refunded or expired renewal row exists (no false paywall); scores, flags, and fair-value numbers are unchanged.
The collector binder now says when a scan is scoped rather than a lifetime record. Unknown custody is shown as unknown, not a fabricated zero. Tax exports refuse incomplete scans instead of inventing a lifetime IRS file.
In review. Comparable sales now exclude operator-wallet and lending-protocol transfers, and the sales history labels those prints. When a screening source is unavailable, the report withholds the fair-value point, range, and deltas and says Insufficient data — screening source unavailable instead of guessing a floor number.
Wallet sessions now stay signed in through brief network drops, wallet reconnects are faster, and gated purchase buttons stay readable instead of dimming.
In review. If a fax report cannot load and there is no request id to quote, the error page now shows the Sentry event id when one exists, or says no reference is available — see Sentry event. It no longer prints Error reference: unknown with no guidance. Display only; no scoring or pricing change.
While the product is still closed, gated methodology and glossary citation surfaces show a closed-launch note, methodology links stay reachable as notes instead of dead ends, and the launch state matches across pages. Payments stay closed until public launch.
In review. Fax report pages are private and not stored in public caches. A temporary load failure says the report is unavailable, offers Try again, and shows an error reference you can quote. Footer nav and legal links stay reachable. The sitemap is built from the live token catalog and asks crawlers to retry when that source is down. Display and reliability only; no scoring or pricing change.
In review. Where same-token sale dispersion exceeds the 15% point-accuracy bar, the report leads with a defensible fair-value range and demotes the single point to an indicative secondary figure; with insufficient coverage it says Insufficient data instead of a number. The homepage sample certificate stays withheld until real coverage supports a figure. Display and copy only; the pricing model is unchanged.
In review. The lineage tool restores the 9,602 indexed BAKC companions from the chain-derived pairing index. Koda and Otherdeed slots now pass through token art and use visibly labelled representative art when token art is unavailable. Otherdeed lookup failures render as degraded, not absent, and lineage index reads have a two-second query deadline. Display and data-truth fix only; no scoring or pricing change.
In review. A /dmca page presents ApeFax's counsel-delivery DMCA draft from the checked-in source: scope, notice requirements, contacts, counter-notices, and repeat-infringer policy. Contacts use the same constants as Terms and Privacy. Tests pin the source hash, section order, and contact sourcing. The draft is awaiting counsel's one pass. Legal copy only; no product behaviour changed.
In review. Every rarity rank on ApeFax now names its source: it is the rank OpenSea publishes for the collection, not one ApeFax computes. Rank is a separate axis and never changes the grade; if OpenSea publishes none, the cert shows it as unavailable and the integrity score records a low-severity flag with a fixed deduction. No scoring, pricing or rank computation changed.
Locked report previews keep the exact integrity score private. Reports with insufficient verified history now show an unverified state without a numerical score across report, tool, account, certificate, and verdict surfaces. No scoring methodology or tier threshold changed.
Fair Value and Calculators can again resolve the public BAYC teaser and locked viewer data while launch remains closed. Refresh, history, payments, minting, and paid report data remain unavailable until their existing launch gates are approved.
Applied 2026-07-22. The View-refund ledger and revocation machinery are live — every future refund is recorded immutably, bound to its exact source payment, verified on-chain, and refunded access is revoked everywhere. No refund has yet been issued; refunds remain founder-executed under our operating procedures; the Terms govern when a credit or refund may be requested. This is the machinery, not a policy change.
Deployed 2026-07-21. Handled degraded states on the freshness health surface now return a readable JSON body — previously they could surface as opaque gateway errors. Import-time failures remain a disclosed residual. Monitoring and the data-pipeline gates now classify from payload content. This is monitoring and operations honesty; no scoring or pricing change.
Payment-support links now carry report context, refund and renewal wording matches current behavior, and privacy copy accurately states SIWE, IP, telemetry, retention, and data-rights handling. Deployment, mailbox proof, operator rehearsals, counsel review, and an atomic refund writer remain pending.
ApeFax now records only normalized aggregate product events and LCP, INP, and CLS, never raw search text or dynamic identifiers. Critical routes, writer and evidence freshness, deployment identity, and fair-value quarantine now feed one daily health digest; production delivery remains pending.
Signed backup, independent-copy verification, preview restore, edge checks, monitoring delivery, and incident-response controls now pass candidate code gates. Release remains pending until the production backup, isolated restore, edge, Sentry-to-Discord, custody, and founder approval receipts are complete.
Collections now use 30-minute static regeneration, public API inputs and rate budgets fail closed, OG errors cannot be cached publicly, and entitled trending responses are cache-separated. The exact implementation passed a bounded 286-request local smoke; live edge, proxy, deployment, and monitoring proof remains pending.
Signup receipts, the founder-controlled launch announcement, preference handling, bounce and complaint suppression, and replay-safe batch delivery now pass the candidate gates. Delivery remains disabled until the founder-owned production proof is complete; no launch email has been sent.
Public launch, pricing, account, roadmap, methodology, email, and fair-value surfaces now share canonical state and commercial terms. Sample fair values and unverified operator data are withheld, while minting and the numeric model remain closed.
The Founder section no longer promises an expired July 4 opening or a broad discount on every transaction. Price, capacity, credit terms, early access, and availability now follow the canonical product state; the pass remains closed pending review, deployment, and founder approval.
Trait floors now distinguish warming, unavailable, and truly unlisted states. Collection modules disclose live, partial, or planned availability; Trade Value names its selected basis; Explore rejects incomplete value scans; and non-serving fair-value views show receipt-verified evidence only.
Release identity now uses the full source commit, launch checks verify one declared gate state without mutating production, and the operator packet defines exact cutover, rollback, monitoring, and incident evidence steps.
The root route now changes from the intake gate to the full homepage only when the public launch gates open together. View Unlock copy follows that same state, minting remains closed through external review, and expired launch-date and unsupported sample-band claims are removed.
The methodology page now exposes dated grade weights, the four-trait sale waterfall, comparable-evidence coverage tiers, custody digests, and the prospective activation gate. It publishes no median-error claim until contract v3 earns one.
ApeFax now publishes product changes in an append-only ledger with plain-language summaries, rollout status, and an immutable pull request or commit receipt. An RSS feed carries the same verified record.
When model-backed fair value has not passed its evidence gates, ApeFax withholds the number and shows receipt-verified comparable ranges instead. Numeric fair value remains off and certificate minting remains closed.
Terms and product copy now match implemented calibrated ranges, Treasury SDN screening, degraded screening behavior, and the refund path. Counsel review remains scheduled after launch under the recorded waiver.
Same-source holder counts and movement are implemented behind migration 076 and the daily includeHolders rollout. This remains in review until both production steps are applied and verified.
Collection volume, sale windows, tape dates, trait last-sale rows, and premium-floor inputs now read from screened sources and expose honest missing states instead of substituting contaminated data.
Public trending responses no longer include score fields without entitlement. Market discovery remains available while paid report intelligence stays private.
The collections desk now serves a fast public shell while paid forensic analysis stays isolated behind its entitlement boundary. Behavioral leak tests cover the split.
