Decentralised News · 2027 planning edition · Evidence reviewed 22 November 2026
What Matters
A long historical-trade window does not guarantee access to every failed order. Record status, product, request interval, pagination and download expiry each matter. Build the collection schedule around the shortest relevant documented window, then verify that saved files cover the account activity you intend to analyse.
Evidence and method
This is a 2027 planning edition based on official Binance, Bybit, OKX and Kraken documentation checked on 22 November 2026. The comparison concerns specific retrieval routes, not each company’s complete internal retention policy. API documentation takes priority for endpoint behavior; website export guides describe a separate access path. Where those documents give different canceled-order windows, the article preserves the difference. No private account was queried, no real-money trading test was performed, and no missing record was recovered. Calculator defaults, capture schedules and ledger examples are original hypothetical scenarios. Recheck the relevant product, account type and current documentation before relying on any retrieval window.
DN Alpha Thesis
An exchange connection is incomplete if it can place orders but cannot preserve enough evidence to explain them later. The critical design variable is the age of the oldest uncaptured record at the next verified successful collection, compared with the applicable access window. A retrieval route that works for filled trades can still fail for rejected or canceled orders. DN therefore treats record coverage, status exclusions and export completion as explicit operating requirements. The goal is a reproducible history with declared gaps, rather than a dashboard that appears complete because its most recent request succeeded.
DN Capture Window Planner 2027
All defaults are hypothetical. Enter the shortest documented access window for your exact record class, then estimate how old uncaptured data could be when collection finishes. No exchange connection or credentials are used.
Scenario method and limits
The initial inputs and results are an original, synthetic DN example. They are provided to make the method reproducible, not as live prices, measured performance or a prediction for 2027. Change the inputs and download your own scenario above.
- User-entered retention assumptions are not independently verified.
- No API calls, account access or recovery are performed.
- Model assumes prior capture was complete and next collection succeeds; late or migrated records require separate checks.
- Cannot determine statutory recordkeeping periods or tax completeness.
Your exchange history has more than one expiry date
A trader notices a rejected order on Friday, opens a support ticket on Monday, and discovers that the order-history endpoint no longer returns the original record. The trades are still visible. The balance is intact. Yet the missing item is the one needed to establish whether a request reached the venue, why it failed, and what the trading system did next.
This is an evidence problem before it becomes a tax or performance problem. An exchange can retain completed trades while exposing unsuccessful orders for a much shorter period. A download page can offer a longer archive than an operational API. A query can cover a week even when the underlying history extends much further. Treating those statements as interchangeable creates avoidable gaps.
This 2027 planning edition compares documented retrieval rules checked on 22 November 2026. It does not claim to have queried a funded account or recovered anyone’s missing records. The practical objective is to identify which evidence needs frequent capture, which belongs in periodic exports, and which cannot be assumed recoverable at all.
The five clocks an archive policy must separate
| Clock | Question | Consequence of confusing it |
|---|---|---|
| Historical coverage | How old may an eligible record be? | You assume a record remains queryable after the endpoint stops serving it. |
| Request interval | How much time may one request cover? | You mistake a seven-day query limit for seven-day total retention. |
| Status-specific coverage | Are unfilled, rejected and canceled orders treated differently? | You preserve trades but lose the evidence behind failed execution. |
| Publication delay | How soon does a completed event appear in history? | You mark a late-arriving record permanently absent. |
| Download-link expiry | How long is a generated export link usable? | You save the link and lose the file. |
There is also a business deadline: when you need the records. A monthly accounting close, an incident investigation and a strategy execution review need different evidence at different times. The archive should satisfy the shortest relevant access window, rather than waiting for the longest reporting deadline.
For each record family, write down the venue, legal account, subaccount, product, endpoint or export route, time basis, status exclusions, maximum request interval and pagination method. A single spreadsheet cell saying “two years of history” is insufficient. It leaves out the conditions that determine whether the particular record you need belongs to those two years.
What the current primary documents actually say
The table is deliberately narrow. These are endpoint and export observations, not a ranking of exchanges by overall data retention. Account type, product migration, regional service and documentation changes can alter the applicable route.
| Documented route | Relevant limit or distinction | Planning implication |
|---|---|---|
| Binance USDⓈ-M futures all-orders API | Unfilled canceled or expired orders older than three days are excluded; orders created more than 90 days ago are also excluded. | Capture unsuccessful orders promptly; investigate archive exports separately. |
| Bybit V5 order history | The checked page limits fully canceled, rejected and deactivated records to the last 24 hours; beyond seven days it describes orders with fills. Requests span at most seven days. | Status matters more than a broad historical-history label. |
| OKX order history and archive | The ordinary history route covers seven days and documents a two-hour window for the specified canceled incomplete orders. Its three-month archive excludes canceled orders without fills. | An annual or monthly export routine cannot be the sole evidence capture for those order states. |
| Kraken trade history API | Pagination and filtering are explicit; the checked reference allows 1–100 trades per page and supports cursor-based pagination. | A successful first page is not a complete history. |
Sources: Binance futures trade reference, Bybit order-history rules, OKX trading API reference, and Kraken trade-history reference. The observations describe the pages checked, not a guarantee that every account can retrieve every otherwise eligible item.
Notice what this comparison does not establish. It does not prove that an exchange has deleted a record from all internal systems when a customer endpoint stops returning it. It also does not prove that support can retrieve it, that a different export contains it, or that its absence means the underlying event never happened. Those are separate questions requiring evidence.
Why an order archive is not a trade ledger
An order records an instruction and its lifecycle. A fill records an execution. A ledger entry records a balance effect. One order can generate several fills. A canceled order can have earlier fills. A rejected request may have no fill at all. A funding payment can change the balance without being a purchase or sale.
Consider an illustrative order to buy 2 units. It receives fills of 0.8 units and 0.7 units, then the remaining 0.5 is canceled. An order-only summary may show the final state and aggregate executed quantity. A fill export reveals the two prices and two fee records. A ledger explains how the acquired asset, quote currency and fee asset changed. Each answers a different question.
Bybit’s trade-history reference exposes execution identifiers, quantities, prices, timestamps and fee information, and defines a separate pagination and request-interval contract. The distinction supports a practical design: capture order states and executions independently, then join them by stable identifiers rather than assuming identical row counts.
Kraken’s export guide similarly distinguishes trade details, ledger movements, balance snapshots and account statements. A balance snapshot is useful for reconciliation, but cannot recreate every intermediate action. A statement is useful evidence of a period, but its existence should not be mistaken for a full machine-readable order lifecycle.
A worked capture-window calculation
The DN Capture Window Planner models the age of the oldest uncaptured record when the next successful collection finishes. It does not connect to an exchange. All defaults are hypothetical planning inputs, not venue guarantees.
Assume a record class remains queryable for 24 hours, your last complete capture was six hours ago, the next scheduled collection could be an hour away, and a disruption could add 12 hours. Allow 15 minutes to finish the batch and a two-hour safety buffer.
The projected oldest-record age is 6 + 1 + 12 + 0.25 = 19.25 hours. Adding the buffer uses 21.25 hours of the assumed 24-hour window. The remaining headroom is 2.75 hours. This is arithmetic under the entered assumptions, not proof that a particular record will be returned.
Now change only the disruption allowance from 12 to 16 hours. Projected age becomes 23.25 hours; with the same buffer, the plan exceeds the assumed window by 1.25 hours. Increasing the archive’s storage capacity does nothing to fix that gap. You must collect sooner, reduce recoverable downtime, use a documented longer-access route, or accept that this record class may be missing.
A zero-headroom result is not a robust schedule. It leaves no room for clock differences, late publication, slow pagination or a retry. The model intentionally flags the exact boundary rather than awarding it a green light. An unknown retention rule remains unknown; entering a generous guess disguises the issue.
Pagination can lose records without producing an error
Suppose a hypothetical endpoint returns 100 records per page and your chosen interval contains 1,250 records. One successful response retrieves only the first 100. A process that records HTTP success as “export complete” silently misses 1,150. Correct processing needs every page, a clear stop condition, and a record of what interval was actually covered.
The problem becomes harder when new records arrive while the archive job is running. With offset pagination, insertion at the front can shift rows between pages. Cursor behavior varies by API. A production collector should follow the documented pagination contract, preserve cursors where applicable, and reconcile a small overlap between adjacent completed intervals.
Overlap creates duplicates by design. That is acceptable when deduplication uses a stable execution, order or ledger identifier with the appropriate account and product context. Deduplicating only on timestamp, price and size risks collapsing two legitimate identical fills. Deduplicating order updates too aggressively can erase the transition you were trying to preserve.
A useful completion receipt contains the requested interval, returned minimum and maximum event timestamps, page count, row count, terminal cursor condition, retrieval time and any excluded record classes. It should say “complete for the declared scope,” not simply “complete.” A week of spot trades does not include futures funding, and a main account does not automatically include every subaccount.
Historical export is a second route, with its own limits
Binance publishes a separate asynchronous futures order-export workflow. Its checked documentation specifies a maximum one-year request interval, a request allowance shared with the website, and expiring download links. That is evidence of a different retrieval route, not unlimited recovery. See the Binance futures account/export reference.
Bybit’s self-export guide describes its website order-history export separately from the API, including six-month slices within the previous two years, row limits, canceled-order exclusions and seven-day download links. The checked website guidance and V5 API page do not describe identical canceled-order windows. Preserve that difference rather than choosing the more convenient number.
The operational sequence is request, wait for completion, download, open, validate and save. A job identifier is not the file. A completed request with a broken download is not a preserved archive. A downloaded ZIP that has never been opened is not yet evidence that the expected records are readable.
Keep the raw export intact and create a separate normalised copy for analysis. If later calculations change, the original remains available for a fresh import. Name the file using the venue, account label, product, covered period and retrieval date. Avoid putting secret credentials or unnecessary personal data in filenames.
Three layers make the archive useful
Layer one is the raw record. Keep the original response or export, its schema version where available, retrieval timestamp and source route. Preserve the supplied asset identifiers, decimal strings and event timestamps. Early conversion into floating-point numbers or local-time strings can remove precision or ambiguity you may need later.
Layer two is the normalised ledger. Translate records into consistent internal fields while retaining a link to their originals. Make the unit explicit: base quantity, quote value, settlement currency and fee currency are different columns. Store both the venue event time and the capture time so you can examine late arrivals.
Layer three is the reconciliation report. Begin with a known balance, add documented inflows and credits, subtract outflows and debits, and compare the result with the ending balance in each asset. Explain residuals instead of forcing them to zero. A discrepancy can reflect a missing fee, an internal transfer, a conversion, rounding or an unrecorded product activity.
For example, an illustrative USDC ledger begins at 1,000. It records a 500 deposit, a 200 purchase debit, a 1 fee and a 2 funding credit. The expected ending amount is 1,000 + 500 − 200 − 1 + 2 = 1,301 USDC. If the snapshot is 1,300, the one-unit gap is a research task. It is not evidence by itself that the venue lost money.
Capture frequency should follow evidence value
A frequent trader may need a rapid order-event feed plus periodic history reconciliation. An occasional spot buyer may mainly need complete trades, cash movements and a regular document export. Neither should copy the other’s schedule without examining the records that could disappear first.
A practical policy separates short-lived unsuccessful-order evidence from slower accounting archives. It also sets a maximum unobserved period. When collection fails, the issue becomes more urgent as the gap approaches the shortest relevant retrieval window. Alert on age of the oldest uncaptured interval, rather than only on whether the latest scheduled job failed.
Record both the last attempted capture and the last verified complete capture. A process that repeatedly downloads the first page can look healthy in a scheduler while its complete-data watermark remains weeks old. Operational status must reflect coverage, not activity.
Decide in advance which gaps are tolerable. Losing an unfilled experimental order may not alter tax calculations, but it can invalidate a claim about rejection rates. Missing executions can affect performance calculations. Missing fees can change net returns. The cost of a gap depends on the decision the data is supposed to support.
What to do when history already appears missing
- Define the missing record. Separate an order, execution, transfer, funding payment and statement. Identify the account, product and approximate UTC interval.
- Check filters and units. Confirm product category, symbol, subaccount, timestamp units, request interval and status eligibility.
- Finish pagination. Verify that the collector followed every page or cursor and did not stop on a default limit.
- Check an official alternative route. Review document exports, archived endpoints and account-migration guidance without assuming they contain the same fields.
- Preserve the gap honestly. Record what is missing, what was attempted and which calculations are now uncertain. Request help through the venue’s official channel if needed.
Do not manufacture a fill from a balance change, treat a screenshot as a complete ledger, or convert an absent response into “no trade occurred.” Reconstructing a plausible history and establishing an evidenced history are different tasks. Where the original cannot be recovered, calculations should expose the limitation.
Designing the 2027 record
The strongest archive is boring to inspect: clear ownership, explicit scope, readable files, completed pagination, repeatable reconciliation and dated retrieval rules. It needs no proprietary safety score. Its value comes from allowing another person to reproduce what happened without relying on a dashboard that may have changed.
Before enabling a new exchange integration, answer four questions: which records it needs; how quickly the shortest-lived ones must be captured; how an interrupted collection resumes; and how completeness will be checked. Then repeat that review when the venue changes its API, account model or export interface.
This article’s comparison is documentary and intentionally limited. It does not establish a jurisdiction’s legal recordkeeping period, promise a successful recovery, or infer future 2027 platform behavior. The useful decision today is whether your evidence capture finishes comfortably inside the applicable documented window, and whether your saved files actually support the conclusions you plan to draw.
Frequently asked questions
Does a seven-day API query limit mean older history is gone?
No. A maximum request interval and a total historical coverage window are different constraints. An endpoint can require many seven-day requests to retrieve a much longer eligible period. Check the record status and pagination rules as well.
Why can I find a trade but not the canceled order around it?
Orders and executions are different records, and venues may expose them for different periods. Some documented order-history routes restrict older unfilled canceled, rejected or deactivated records. A surviving fill does not imply that the full original order lifecycle is still accessible.
Does an empty API response prove that no trade happened?
No. Filters, product category, subaccount, pagination, timing, migration and status exclusions can produce an empty result. Reconcile executions and ledger movements and check documented export alternatives before concluding that an event never occurred.
Can the DN calculator recover my missing history?
No. It is a local planning tool using your entered retention window, collection delay, outage allowance and safety buffer. It performs no network request, accepts no account credentials and cannot establish whether any particular record is recoverable.
Is an export download link enough for my archive?
No. Download and open the actual file, check the covered interval and record families, and keep the original with a completion receipt. A link may expire and a generated export may still omit the records needed for your analysis.
Sources and verification
Primary documentation checked on 22 November 2026. Exchange rules, regional availability, fees and API behavior can change; recheck the exact product and account before using this 2027 planning guide.
- Binance USDⓈ-M futures trade API: All Orders — Three-day unfilled canceled/expired exclusions and 90-day order creation cutoff.
- Bybit V5 Get Order History — Current status-specific 24-hour/seven-day rules, asynchronous updates and maximum request interval.
- OKX V5 API reference — Seven-day order endpoint, two-hour specified canceled-order window, three-month archive and no-fill exclusions.
- Kraken Get Trades History — Trade history pagination, page limit and cursor support; no unsupported universal retention claim.
- Bybit V5 Get Trade History — Execution records, identifiers, fee fields and separate request/pagination rules.
- Kraken account history exports — Distinct trade, ledger, balance and statement export families.
- Binance USDⓈ-M futures account API: historical downloads — Asynchronous export request interval, shared allowances and download expiration.
- Bybit self-export account data — Website order-export windows, limits and distinctions from operational API.
Related DN research
- 8 Exchange API Changes a Trading Agent Must Review Before Deployment
- Your Crypto Funding Fee Looks Wrong: Rebuild the Bill Before You Blame the Exchange
- Your Crypto Exchange Balance Doesn’t Match: 9 Checks Before You Blame Missing Funds
- Your Crypto Order Timed Out: Did It Trade? What to Check Before Retrying