What verified means here
Verification standard 4.0 · record verified 3 Nov 2026Verified means four things. The hull is identified by its IMO number, not its name, so a rename never splits one vessel into two records. Every asking rate is read from a named page on a dated capture and kept in the currency it was published in; nothing is converted or interpolated. A figure enters a comparison only against another figure of the same hull, currency and season captured within thirty days. And a card is recorded as withdrawn only after two consecutive reads find the page gone; a failed fetch is never a withdrawal.
What is never published
Owner identities and the legal names of owning entities, at any tier. Achieved transaction prices, build costs and third-party valuations. Any grade, ranking or characterisation of a broker, agent or firm. Exact counts of what we measure on public surfaces: public counts are floors with a plus; each dossier states its own sample sizes.
How the record is built
- 300+ verified dossiers, each compiled by an analyst against a fixed verification standard, then re-verified on a 48-hour cadence keyed on the IMO number.
- 23,500+ dated observations. 92% are attributed to the page they were read from; the remainder to a record label naming exactly one house. Labels that describe a mandate are never a source.
- Readings more than 2.5 times away from the others on the same hull are held for review, not published as conflicts.
- The Stonys Charter Index is a repeat-sales index on same-hull, same-card, same-currency pairs with a hull-cluster bootstrap band. Periods without enough pairs are omitted.
Corrections register
A house that shows its own retractions can be trusted with the rest. Entries are permanent.
The first edition asserted that the hull was absent from several houses. The absences had been inferred from constructed page addresses that returned nothing, which is not evidence. A corrected edition was published after the pages were harvested. Five root causes were recorded and the rule changed: an absence is admissible only from a harvested page on a house whose address architecture is known, and a constructed-address miss on an id-based house is zero evidence.
Before anyone asked, the desk audited its own absence calls across the record. 23 of 38 did not meet the standard above, an error rate of 60.5%, one-way, in Stonys' favour. All were re-swept under the corrected rule. The figure is published because a measurement error that flatters the seller is the worst defect this desk can carry.
A review of the observation record found source labels contaminated with mandate descriptions and a small number of readings that could not be the same instrument (a sale asking recorded beside a weekly rate; a figure with a missing digit). Sources are now resolved from the fetched page; 5 readings are held for review and excluded from every comparison until re-read.
A second re-read the same day, of thirty published change events and the six public dossiers against their pages, found that the event check introduced on 23 October had tested that a new value was on the page but not that anything had changed. Rate changes that were the low- and high-season figures of one card, currency changes raised on cards that price winter in US dollars and summer in euros, flag changes that were fragments of a specification line, refit declarations that were a change of page layout, and arrivals of a different hull carrying the same name were all published. The feed was re-audited whole: of 234 candidates, 123 hold under rules that require the old figure to have left the card, both currencies not to sit on one card, a recognised flag state, the refit year to be printed beside the word refit and later than the build year, and the page's printed length to match the register. On the dossiers, movement, currency-fork and rate-change findings had compared two seasons of one card; one history on a reused address spanned two hulls; the cohort median was labelled as including the subject when it did not; and two cost tables applied VAT to the provisioning allowance. Public editions now state only findings computed from readings re-read on their pages, with seasons separated; the analyst's narrative findings remain on partner editions, labelled as written at dossier date.
The Command Deck's widest-spreads panel carried four rows that did not hold when the pages were re-read: a figure from a broker page for a different vessel of the same name (a 50 m hull against our 60 m record); a figure recorded in EUR from a page whose rate card prints it in USD; a winter rate paired with a summer rate; and a figure a publisher's structured data carried but its rendered card did not. Each row has been withdrawn. The rule is changed: a reading enters a published spread only after its page has been re-read and the figure with its currency, the vessel (by register length or builder) and the season label were found on it; readings that differ in season are never paired; a publisher's structured data is accepted only when its rendered card prints the same figure. The re-read count and the count of readings held are shown on the panel.
Measured accuracy
The desk measures its own error rates every morning and publishes them here. These are counts from the same files the figures are built from; nothing is estimated, and a figure that cannot be measured is not shown. Computed 3 Nov 2026.
301 of 332 monitored hull pages answered (90.7%). Basis: the scan's own run record.
2,114 readings: 70.1% from the page's own machine-readable offer, 17.8% from a labelled or repeated printed figure, 9.3% held under verification and never published. Basis: confidence class persisted by the extractor on every reading.
97 displayed figures re-fetched from the page they cite: 86 confirmed, 0 restored, 0 contradicted (100.0% stand). Basis: a random sample; value present on the page, vessel bound, second fetch agrees; a contradiction blocks the deploy.
216 controlling values tested against their live pages: 196 verified, 16 mismatched (90.7%); 103 pages could not be read and are counted, not inferred. Basis: every hull's controlling value against its page, rendered lane for script-walled houses.
91.6% of readings are bound to the page they were read from; 8.1% have no retained page and are held out of every comparison; 280 readings (1.4% of current captures) are a house's own currency conversion and are never read as rate movement. Basis: the observation record.
143 events, of which 22 are rate changes (12 in the last thirty days), 34 cards withdrawn and 11 restored; the rest are catalogue arrivals and departures, refit years, flags and currency changes. Basis: the published event record; a rate change needs a second read, a stable series and a figure inside the hull's own band.
120.8 on 109 repeat pairs in the latest period (95% band 106.3 to 131.2); 365 pairs in all; 8 published periods. Basis: the published index record.
Every figure, and how to check it
Each figure on the open surfaces links here. Every entry names the computation and opens the record it is computed from as a readable table; each record page also offers the file for repeating the arithmetic.
Count of hulls in the register with a published dossier, floor-rounded to fifty. Record: the register (vessels).
Count of rate readings in the observation record after exclusions (no positive rate, sale asking instruments), floor-rounded to a thousand. Record: the observation record, one row per hull with its latest page-backed reading.
Repeat-sales index on pairs of readings from the same hull, same card and same currency across seasons, base 2019 = 100, with a hull-cluster bootstrap band. Periods without enough pairs are omitted. Record: the index by period.
For each dossier with two or more live published figures, the gap between highest and lowest as a share of the lowest; the median across dossiers, n=290+; the share of gaps above EUR 10,000. Inputs are each dossier's cross-platform rate scan, every row of which links its page. Recomputed at every release by the claims audit; publication is withheld on disagreement.
Share of dossiers with a named mandate holder and two or more published figures on which the mandate holder's figure is the lowest. Same inputs as dispersion.
Earliest dated reading on any hull; earliest archive capture; count of readings whose page is an archive capture. Record: the per-hull observation files, type archive recovery, each linking its Wayback capture.
Same hull, currency and season, read within thirty days from two or more sources, difference of five per cent or more, every reading re-read on its page (figure and currency present, vessel bound by register length or builder, season label agreeing). Record: the cross-source comparison, every pair with its pages.
Every published event carries the address of the page it was read from. Rate changes need a second read, a stable series and a figure inside the hull's own band. Record: the change events, every one with its page.
Hulls placed at the city named in the dossier builder field or at the builder's principal yard, coordinates from GeoNames. Record: the build yards.
Computed from the partner instruments, each file carrying its inputs, method and stated assumptions; opened with partner sign-in.
Editions and access
Two editions exist. The open edition is published in full for a small number of dossiers chosen by depth of record, and as a preview for the rest: identifiers, the latest published asking rate and what the dossier holds, stated as counts. The partner edition is the full library and the partner layers of the Command Deck. Partner material is served only to an authenticated session; it is not present in public pages, page source or public records.
Both editions are built from the same record under the same rules. Nothing in the partner edition is withheld from the open edition on grounds of doctrine; the difference is depth, not content class. Owner identities, achieved prices and grades of firms appear in neither.
Coverage
The record is organised by hull, not by position. Build yards are located for the verified fleet from the builder named in each dossier, with coordinates from a public gazetteer; where only the builder is known, the hull is placed at that builder's principal yard and marked as such. Vessel positions, movement, marinas and anchorages are partner layers scheduled for the next release; no open page implies a live feed.
Freshness
Every figure carries the date it was observed. The newest observation in the record is 1 Nov 2026. Where a hull has not been re-verified within the scan cadence, the dossier says so beside the figure rather than replacing it.