# Variational October 2026 — Data and methods

Editorial revision: **convergence-2026-11-03**. Original publication: 2026-11-02. This English edition and the Chinese edition use the same saved data, calculations and chart series. The statistical window is **2026-10-01 00:00 UTC ≤ timestamp < 2026-11-01 00:00 UTC**, unless a narrower window is stated. Amounts remain **USDC**: M means million and B means billion. They have not been relabelled as USD.

For a compact independent number check, see [shared metric proof](derived/final_metrics.json). This revision leaves all saved derived results unchanged; the public package contains the aggregate subset. [Readability changes](READABILITY_CHANGES.md) distinguish wording and source corrections from numerical changes.

## Coverage at a glance

|Measure|Evidence and usable window|Status and interpretation|
|---|---|---|
|Protocol TVL, two-sided OI and rolling-24H volume|Official API history; 736 of 744 fixed hourly slots in October|PARTIAL: eight hourly slots missing. Daily charts require all 24 slots, giving 29 complete days.|
|October volume estimate|Difference between cumulative counter readings near the two month boundaries|PARTIAL: no negative counter steps observed; sampling offsets prevent a claim of exact calendar-month volume.|
|Settlement pool creation|Arbitrum creation events and verified chain creation times; all 31 October dates|COMPLETE for the observed creation registry; pools are not users or registrations.|
|Daily pools with funding or return flows|Published day aggregates; October 1–30|PARTIAL: October 31 missing. Daily counts do not identify monthly distinct pools or addresses.|
|Verified ordinary USDC deposits|Covered depositUSDC channel; October 6, 18:40:12 through October 31 UTC|PARTIAL: 25 complete days and one partial day; not all-platform deposits or net inflows.|
|Market structure|Daily 23:00 UTC observations; 29 of 31 days|PARTIAL: October 13 and 16 missing; shares describe sampled rolling-24H volume, not exact monthly turnover.|
|Treasury and Core OLP balances|On-chain wallet balance snapshots near UTC boundaries|PARTIAL daily differences: 29 valid days; boundary net change is measured separately.|
|Swap quotation costs|RFQ history; qualified fresh quotes during standard opening hours|PARTIAL: market-specific coverage; quoted estimates, not executed trade costs.|
|Trading-active addresses, full net inflows, classified monthly revenue|No complete validated evidence set|MISSING; none is inferred from pool counts, ordinary deposits or balance changes.|

Unknown observations stay blank. “No event” is distinct from not collected, not indexed or failed processing. September partial windows are not full-month comparison bases.

<a id="scale"></a>
## Scale and volume

**Source:** Variational official API historical snapshots. **Evidence:** [hourly grid](derived/protocol_hourly_grid.csv), [daily observations](derived/protocol_daily.csv), [boundary evidence](derived/protocol_boundary_evidence.csv), [protocol summary](derived/protocol_summary.json).

TVL is the official API field, not cumulative user deposits. Core OLP wallet balances are not added to it. The precise equivalence between the documented OLP vault and the monitored Core OLP wallet has not been established. No price-based explanation for the change in TVL is asserted.

OI in the headline cards and protocol trends is **protocol-reported two-sided open interest**. Market OI is the sum of that market's reported long and short OI. The two series are kept separate; dividing official OI by two does not establish a compatible historical user-side series.

The volume trend averages the readings of “volume over the previous 24 hours” taken during each day; it is not that day's executed volume. For each UTC hour, the protocol grid selects the first stored observation within ten minutes after the hour. Within duplicate sampling minutes, the last stored version is selected. All valid hourly slots receive equal weight. Monthly observed means use 736 slots; daily charts use only dates with all 24 valid slots. October 16 and 17 are therefore blank in the daily charts. The seven-day line requires seven consecutive complete dates and never bridges an unknown gap. These are sampled hourly means, not continuously measured daily averages.

Cards show near-boundary **point values**; curves show **daily means of valid hourly samples**. Their final values need not be identical. Raw observed peaks use all stored samples and may be affected by varying sampling density.

The October volume estimate is `end cumulative volume − start cumulative volume`, **USDC 78,149,501,869.529172**. Both observations were collected near the respective month boundaries. There are no negative counter steps in the saved history, but sampling and publication timing still limit precision. Rolling-24H values are never added across hours or presented as exact monthly volume.

Lower underlying prices can reduce USDC-denominated OI even if contract quantities do not change. The [price diagnostic](derived/price_oi_diagnostic.csv) uses a fixed starting OI/price proxy; its price component and residual are not actual additions or closures of positions. The diagnostic uses October 1 and 31 observations near 23:00 UTC, rather than the protocol's month-boundary points.

<a id="participation"></a>
## Creation and ordinary-deposit participation

**Sources:** Arbitrum settlement pool creation events; verified ordinary USDC deposit events. **Evidence:** [daily creation and deposits](derived/pool_deposit_daily.csv), [deposit participation](derived/deposit_participation_daily.csv), [cohort contribution](derived/deposit_cohort_contribution.csv), [participation distribution](derived/deposit_activity_distribution.csv), [event summary](derived/events_summary.json).

A settlement pool holds settlement funds for the bilateral relationship between a user and the OLP liquidity provider. It is not a personal account or a registration. This description follows the saved PoolCreated participant structure and the project's settlement-pool explanation. New settlement pools are deduplicated by pool address and assigned by chain creation time, not backfill or insertion time. Creation transaction, block and log evidence are matched before being used. October has 10,505 new pools versus 15,606 in September; this is a full-month pool-creation comparison. A pool is neither a registration nor an independent person. User-side addresses can also represent more than one economic participant, or several addresses can belong to one participant.

Ordinary deposits require `VERIFIED_USER_SIDE` classification and `PASS` quality status. The chain/transaction/log identity deduplicates events; amount and user-side participant mapping are verified. The user-side depositor address, normalized consistently, is deduplicated across pools. `tx.from` is not used as a user identity.

The observed channel total is USDC 211,335,576.708232 across 46,006 events and 17,406 distinct addresses. Coverage starts October 6 at 18:40:12 UTC; later November evidence is excluded from October totals. The amount is a deposit flow: every qualifying deposit is counted, including repeat deposits, and withdrawals are not subtracted. It is neither new capital, net inflow nor growth in TVL. Oct 1–5 are not covered and remain blank; Oct 6 is partial, and later dates still represent only this channel. On Oct 7, USDC 10,302,563.632161 was deposited through the covered channel.

“Existing addresses” means user-side addresses first seen in the available chain creation history **before October 1**. Such addresses account for 82.8% of the covered ordinary-deposit amount: the denominator is the amount deposited, not the number of addresses or a retention cohort. Amount percentiles first sum each address across the observation window, then rank addresses; P90 uses linear interpolation and means that 90% of addresses had no more than that cumulative deposit amount over the covered window. Top-ten share uses the same covered-period denominator. Daily address counts are not summed to produce period distinct addresses.

Deposit-day distributions describe the number of UTC dates with a qualified event within this partial window. New addresses have shorter possible observation periods, so the distribution is not retention. The seven-day new-pool deposit fraction uses only the 6,887 newly created pools for which the full `[creation time, creation time + 7 days)` window was observed with complete channel coverage. Of these, 4,998 (72.6%) had a qualifying deposit. The denominator is not all 10,505 new pools. Pools with less than seven days of observation or incomplete coverage are excluded, not counted as failures, and this is not retention.

<a id="activity"></a>
## Settlement pools with fund movements on each day

**Source:** saved funding/return daily aggregates, checked against the prior backup. Direct evidence: [daily count CSV](derived/active_pool_daily.csv), [summary and all daily values](derived/active_pool_summary.json), [full activity investigation](ACTIVITY_INVESTIGATION.md).

The upstream website's “yesterday's active pools” field is derived from the union of:

1. Pool addresses in clearing events labelled `olp_to_pool_transfer` during that UTC date.
2. Sender addresses in ERC20 flows classified as `settlement_pool → core_olp` during that date.

The event label alone does **not** independently establish the inbound funding source. The report therefore uses “funding or return flows”, not “official Core OLP funding”. Transfers and returns can include automated settlement, so this is not trading DAU or daily active traders.

The previous investigation read local code and downloaded backups, then made a low-load read-only server check of daily snapshots, checkpoints and the running activity-count definition. Saved evidence is in the audit data package. Thirty October dates are supported and agree with the backup. Different versions for October 2 have the same count and are not double counted. Required event-stream checkpoints cover the snapshot block for each accepted day. October 31 has no published daily snapshot; a later caught-up checkpoint does not retroactively fill it.

The 30 observed dates average 15,716.53 pools, with a peak of 19,680 on October 29. The seven-day comparison is **October 1–7 versus October 22–28**, up 7.4%. It is different from the market-volume comparison ending October 31 and is not used to claim a synchronized divergence between trading and participation.

The aggregates retain counts, not complete daily pool sets. Although historical pool-to-user-side mappings exist, counts cannot be inverted into address sets. Monthly distinct pools or addresses are therefore unavailable. Current `last_activity` and sender `tx.from` are not used to reconstruct history.

<a id="markets"></a>
## Market contributions and structure

**Source:** Variational official API market history and saved market registry. **Evidence:** [market contribution table](derived/market_contributions.csv), [fixed-window comparison](derived/fixed_market_comparison.csv), [daily structure](derived/market_daily_structure.csv), [market names](derived/market_display_names.csv), [protocol reconciliation](derived/market_protocol_reconciliation.csv).

The main comparison is **October 1–7 versus October 25–31**. Both windows contain seven dates with the same weekday composition. It uses the 551 markets already in the September-end set that have all seven observations in each window. The −39.5% result is a change in mean rolling-24H observations, not “October monthly volume fell 39.5%”. Absolute contributions are changes in those daily mean rolling-24H observations.

Market structure uses 29 valid daily snapshots at 23:00 UTC. The identified Swap share is 57.1%, the fixed-window top-five share is 76.0%, and newly observed markets contribute 6.6%. Numerators and denominators use the same observations. Daily concentration charts rerank each date separately; they are distinct from fixed-window rankings. Newly observed markets have no assumed zero prehistory. Their first discovery date is not an official listing date.

Only the 18 Swaps identified by the saved source classification are labelled Swap. Unresolved product and asset classes remain unclassified. Display-name translations follow the saved registry name and do not change classification. No classification is guessed from ticker symbols.

A secondary cross-month check uses September 24–30 versus October 22–28 and a complete common set of 549 markets. It is not a full-month September–October comparison. Official two-sided OI and market OI are reconciled but not mechanically spliced.

The contribution chart uses one reading near 23:00 UTC each day, whereas the protocol trend averages readings throughout the day. For each instrument it compares the average previous-24-hour volume reading in Oct 1–7 with the average in Oct 25–31. The US100S and US500S decreases total USDC 804,807,431.792, or 68.6% of the fixed 551-market net decrease of USDC 1,173,124,882.5. They are major components of that statistical change, not its cause. The differences are not monthly cumulative turnover losses, capital outflows or user losses. Market OI is the size of contracts still open; a decrease is not a loss or withdrawal amount.

To read the three headline shares, add the previous-24-hour volume readings across the same 29 valid daily snapshots. The denominator is the total for all markets. The numerators are the identified Swap contracts (57.1%), the five markets with the largest totals across the whole observation period (76.0%), and markets first appearing in this dataset during October (6.6%). The daily concentration chart instead reranks markets on each date and divides that day's leading five by all markets on that date. These are different measures and neither is exact monthly turnover share.

<a id="treasury"></a>
## Treasury and Core OLP balances

**Source:** on-chain wallet balance snapshots. **Evidence:** [daily balances and changes](derived/wallet_daily.csv), [boundary evidence](derived/wallet_boundary_evidence.csv), [wallet summary](derived/wallet_summary.json), [TVL scope reconciliation](derived/tvl_scope_reconciliation.csv).

Each UTC boundary uses the closest available wallet snapshot within five minutes. The unavailable October 17 boundary leaves October 16 and 17 daily changes blank. End balance minus start balance is the **boundary net change**. The sum of positive daily changes uses only 29 valid daily differences. These have different coverage and need not match.

Treasury increased by USDC 2,599,028.012714 between its near-boundary samples. Its positive daily changes sum to USDC 2,453,652.600676 over the valid 29-day subset. Neither is verified revenue, profit or actual buybacks. Core OLP wallet growth likewise does not establish profit or user equity.

The experimental ratio of positive Treasury balance increments per USDC 1B in volume matches the same 29 valid dates; it is neither a fee rate nor an income rate. There are no usable historical per-pool balances for October, so current balances are not backfilled into historical concentration.

<a id="costs"></a>
## Single-side quotation cost estimates

**Source:** saved RFQ quotation history. **Evidence:** [cost distributions](derived/swap_quote_costs.csv), [quote availability](derived/swap_quote_availability.csv), [hourly states](derived/swap_quote_hourly.csv), [comparison windows](derived/quote_comparable_windows.csv).

For the same snapshot, reference mid is `(1K bid + 1K ask) / 2`. For USDC 100,000 notional, estimated Long cost is `(100K ask / mid − 1) × 100000`; estimated Short cost is `(1 − 100K bid / mid) × 100000`. USDC 100,000 is trade notional, not margin. The estimate measures the 100K quote's price deviation from the same snapshot's 1K bid/ask midpoint, converted to USDC at that notional. Separately charged fees and financing are not added. It is not total fees or the complete cost of an executed trade.

Only fresh, valid quotes in standard opening hours enter costs. Duplicate quotes are removed by ticker, quote timestamp and quoted prices. “Typical quoted cost” is the median of valid quote samples. P90 means 90% of those quote samples had costs no greater than that value; it is neither the maximum cost nor a guarantee about 90% of future executions. “Lower cost” is restricted to the observed Swap sample, not other venues or a September improvement. Existing and newly observed markets are separated, and market-specific dates and sample counts are retained.

Local collection gaps, source-unavailable or invalid-size quotes, stale quotes, standard closures and unknown schedules are kept separate. They cannot collectively be called platform trading failures. Standard opening schedules do not prove complete historical holiday or unscheduled-closure coverage. No stale observation was detected in this sample; that does not establish continuous historical availability.

<a id="funding"></a>
## Related fixed-window financing study

R3 remains a separate frozen study covering October 21, 16:48:26 through October 28, 18:48:26 UTC. It models financing cash flows for fixed quantities. Variational has five included charging periods per leg; TradeXYZ has 170 hourly events. The periods are not identical. Initial notional at a common start price is not verified. There is no equal-capital, equal-seven-day venue ranking, and no claim of October returns, realized account returns or complete holding costs. [Frozen research data](derived/r3-funding-snapshot.json) remain unchanged.

<a id="reproduction"></a>
<a id="reproduce"></a>
## Reproduction and downloads

The Chinese and English reports share all aggregate CSV files and chart-generation scripts. [Download aggregate data](packages/Variational_Oct2026_Data.zip) for the shared CSV and JSON inputs. The website’s report download also includes both reports, method pages and reproduction scripts. Raw events, per-address tables and internal engineering records are excluded.

See [reproduction instructions](PUBLIC_REPRODUCTION.md). `python3 scripts/rebuild_public.py` rebuilds charts and both report bodies offline from the supplied aggregate inputs. This verifies display and aggregate calculations; it does not reconstruct event classification or address mapping from raw events. No production access is required.

Concise display names come from `scripts/display_names.json`, which retains the original source name and product classification. Shorter names do not turn all Swaps into indices or fill unknown classes. Chinese deposit and wallet charts use units of ten thousand USDC; English retains M. Raw values are equivalent after unit conversion. See [readability change list](READABILITY_CHANGES.md).
