Fair value 1.7.0

{point} Ξ · yesterday's lowest listing (independent ask series, previous UTC day) · no trait premium has cleared its uplift test, so none is added · band not yet available (90 days of ask history needed) · 43 of 241 holdout sales priced, 198 had no listing · MdAPE 4.39%, bias +0.33% on those 43

This ask series has 0 of 640 pre-freeze days; the series starts 2026-09-16, after the freeze.

Methodology 1.7.0 sets a non-grail ape's day-one fair value at the independent previous-day ask, collection_listed_daily.floor_eth from the OpenSea stats cache, previous UTC day only, with no carry-forward, and withholds the point when that day is absent. predicted = ask_floor(previous UTC day) × (1 + Σ β_k). Every β_k is 0. A key needs 40 distinct non-grail tokens before the freeze, a positive median of price/ask − 1, a one-sided sign test under 0.01 with zeros dropped, and shrinkage median × n/(n+80) of at least 0.005. This series has 0 of 640 pre-freeze sale-days (history starts 2026-09-16; freeze 2026-07-11T02:26:08.000Z), so the test has no rows. Capture after the freeze adds no pre-freeze days. The floor is the honest day-one worth of a common ape. Long-term trait worth stays withheld on this anchor. On 43 priced canonical non-grail first sales the point equals the ask: MAPE 6.945647%, MdAPE 4.388715% inside 15%, median signed bias +0.325449% inside 5%, within-15% rate 0.906977. Non-inferiority passes with a 0 point gap inside the pre-declared 2.0 point margin: a tie to the ask. The full 241 fails because 198 sales have no anchor. The 80% band is withheld (0 band-window ask days; the rule needs 30). Seventeen canonical grail first sales stay withheld. Leakage failures are 0 in every cell. The sales low-decile is a circularity check and is not this anchor. Methodology 1.6.6 stays a diagnostic failure line and is not a price. D2 is unsigned, so the candidate prints no number.

Fair value today is the lowest listing on the previous UTC day. No trait premium has cleared its uplift test, so this number is that listing.

Your trait is unpriced. This ask series has 0 pre-freeze days, and the premium test reads only sales before the freeze, so no key can pass on this anchor.

4.388715% is the median absolute percent gap between the previous-day ask and the sale, on 43 priced sales. The point equals the ask, so the gap is the ask's own miss. 198 of 241 sales have no number.

8,616 of 10,000, the page's 86 percent, and 280 of 302, the page's 93 percent, are population shares. Priced sales are 43 of 241.

Band not yet available: 0 of the 90 pre-freeze days have an ask, and the rule needs 30.

Tomorrow's number is the new previous UTC day in collection_listed_daily, or withheld if that day is missing. Trait premiums stay unpriced. Capture is still a rehearsal.

The ship anchor is the OpenSea stats-cache ask, previous UTC day only. The sales low-decile is the circularity check. The 43 rows were not paired to that column, and that pairing stays uncomputed.

Priced means 43 sales. 198 are withheld. A cohort sample bar means the cohort has at least 30 sales, which is a separate bar from the priced subset.

43 of 241 canonical non-grail first sales had a previous-day ask (founder 15 of 74, five non-background traits 19 of 132, six or more 7 of 29, and 2 of the 6 index-length-3 sales inside the 241) and 198 were withheld because that ask history starts 2026-09-16.

Shrinkage is median × n/(n+80), at least 0.005, after 40 tokens and a one-sided sign test below 0.01 with zeros dropped. This fit had 0 pre-freeze ask sales, so every β_k is 0.

The ship gate is a tie to the previous-day ask inside a 2.0 point margin. Gap 0. The 5% bias bar and the 15% MdAPE bar are reported checks. The locked file still says scoringRan: false; measuredAt 2026-10-02T12:58:57.145Z is the score. The hash 767c759d2da0d553f390e86075a582cedfd42fa45f6024d75f48f43e2242644d seals version, model name, and config. The measured sentence sits outside the hash.

Leakage failures are 0 in every scored cell. A prediction at or before the last fitted input is refused. Every constant stays on the page; the validation-gain cap and the grail experiment constants are marked off the ship path. Published accuracy is the previous-day ask's accuracy. Trait prices stay withheld, which is the difference from a calculator that prints one.

ApeFax fair value here is the lowest listing on the previous UTC day, from the OpenSea stats-cache ask series, labelled as that listing, with trait worth withheld.

The floor is the value for a common ape until a premium is earned, and no premium has been earned: the fit saw 0 pre-freeze ask sales, and the band is unavailable.

Methodology 1.7.0 is a candidate. It is not served. Trait premiums stay withheld. Grail apes stay withheld. Founder decision D2 is unsigned. No report, homepage, explore rail, or certificate shows this number yet.

Methodology 1.6.6 stays diagnostic-only. It is a transparency failure line and not a price on the report, the homepage sample, the explore rail, or the certificate.

Lowest listing on the previous UTC day, from the independent ask series. Not the highest offer. Not today's live floor.

Unsigned. Prints no number.

D2. I authorize methodology 1.7.0 as a floor-anchored candidate. For a non-grail ape, fair value is the independent previous-day ask floor times one plus a trait premium only where the published uplift test has already passed. A common ape with no passing key is the floor. Grail apes stay withheld. Methodology 1.6.6 stays diagnostic-only and is not printed as a price. Serving a 1.7.0 number still requires serve governance together with this candidate. This signature is the only authorization that can turn the candidate on. Until this decision is signed, the candidate prints no number.

Benchmarks we tested

We benchmarked the served number against two of our own replications: a Punks-style trait-floor replication and an unshrunk hedonic in the Mr Deedz style.

The Punks-style trait-floor replication has no per-trait ask book yet, so it stays withheld until its ask-floor history accrues.

We did not measure our number against OpenSea's top bid, so we do not claim a beat over it.

A floor would be the number if we did nothing. We tested trait premiums, a Punks-style trait-floor replication, and a hedonic in the Mr Deedz style against real sales; the lowest listing beat them all on common apes, and we say exactly where it fails.

Trait-premium evidence

No key has per-trait evidence on this anchor yet: the ask series has 0 pre-freeze days, so the uplift test has no rows.

Constants

• methodologyVersion: 1.7.0
• floorSource: independent_ask_floor
• floorSourceFilter: collection_listed_daily.floor_eth
• floorDay: utc_calendar_day_before_prediction
• anchorRule: previous_utc_day_only_no_carry_forward
• nonInferiorityMarginPct: 2
• premiumDefault: withheld
• grailShipPath: withheld
• salesLowDecileRole: circularity_comparison_only
• maximumUsableFloorProxyEth: 1000
• nonGrailHistoryDays: 365
• grailHistoryDays: 730
• innerValidationDays: 90
• evidenceUnit: latest_sale_per_token
• scoreUnit: first_sale_per_token_in_window
• minimumNonGrailTokens: 40
• kappaNonGrail: 80
• alphaNonGrail: 0.01
• minimumPremium: 0.005
• minimumMdapeImprovementPct: 0.05 · off the ship path
• maxAcceptedNonGrailPremiums: 8 · off the ship path
• minimumGrailTokens: 8
• kappaGrail: 12 · off the ship path
• alphaGrail: 0.1 · off the ship path
• minimumGrailPoolTokens: 20
• kappaGrailPool: 20 · off the ship path
• bandLowQuantile: 0.1
• bandHighQuantile: 0.9
• minimumBandCalibrationTokens: 30
• maxAbsoluteMedianBiasPct: 5
• maxMdapePct: 15
• grailWideMarginMapePoints: 10
• freezeIso: 2026-07-11T02:26:08.000Z
• historyStartIso: 2024-07-11T02:26:08.000Z

Methodology 1.7.0 record