Rule 605: compare quality on matched order populations
A lower reported execution-quality average can reflect a different order mix rather than better results for comparable orders. Rule 605 research should begin with the reporting entity, applicable version and covered-order population, then calculate the statistic using its actual denominator.
This guide uses the SEC Trading and Markets staff’s Rule 605 FAQ page headed April 1, 2026, as read on October 25, 2026. Some entries carry later modification notes. The FAQs state that they are staff views, not a Commission rule or a source of new legal obligations. All numerical reports below are hypothetical.
Freeze the version and the information cutoff
The FAQ introduction and FAQ 39 distinguish the August 1, 2026 compliance date associated with the amended framework from November 1, 2026 for collecting specified best-available-displayed-price improvement information.
FAQ 39 states that November reports containing those statistics may be made public through the end of December 2026. At the October 25 cutoff, that required November collection period has not begun. Do not describe its mandatory monthly results as already observed.
Earlier voluntary provision is possible, but the FAQ requires the available best odd-lot information for that benchmark. Substituting the ordinary NBBO does not create an equivalent best-available-displayed-price statistic. An absent field before its compliance date is not automatically a zero-quality result.
Keep the rule’s legal effective date, applicable compliance date, report month, publication date and retrieval date separate. The August date in the FAQ is a compliance date, not the original adoption or legal effective date of every underlying provision.
Match entities and order groups
FAQs 1–5 and 9–12 address format, reporting roles, trading-month treatment and covered-order questions. One firm can have separate reports for different functions. A broker report and a market-center report do not necessarily describe identical populations.
Use a comparison key containing reporter and role, report month, security, order type, size category, execution population and benchmark convention. Preserve exclusions and empty fields rather than treating every row as the same type of order.
FAQ 3 uses trading days in the calendar month, not settlement days. An order executed on the last trading day does not move into next month’s execution report merely because it settles later.
Special handling can also differ by recipient role. The fact that the same economic order passes through two organizations does not authorize adding both reports as independent new executions without understanding their definitions.
Work the execution-share weighting
Assume two compatible strata in a fictional report have independently supported effective-spread inputs of 0.02 and 0.10 currency units per share. Executed shares are 100 and 900. These inputs illustrate aggregation; they are not downloaded observations.
| Stratum | Executed shares | Effective spread input | Shares times spread |
|---|---|---|---|
| A | 100 | 0.02 | 2 |
| B | 900 | 0.10 | 90 |
| Total | 1,000 | — | 92 |
Share-weighted effective spread = 92 / 1,000 = 0.092.
A simple mean of the two inputs is (0.02 + 0.10) / 2 = 0.06. It gives equal weight to the two rows rather than to executed shares. It is not the required aggregation for this statistic.
FAQ 37 specifies executed-share weights for the listed spread and midpoint statistics. A third covered order of 400 shares with no execution contributes no executed shares to this denominator. Dividing 92 by received shares of 1,400 would produce about 0.065714 and answer the wrong question.
For a partially executed order, use the quantity actually executed for these statistics, not its full original size. Yet an unexecuted order can still belong to covered-order reporting under FAQ 10; exclusion from this execution statistic is not exclusion from every report field.
Demonstrate a mix-driven ranking reversal
Assume a second fictional reporter has spread inputs of 0.04 in matched stratum A and 0.12 in matched B, each worse than the first reporter’s corresponding input. Its execution mix is 900 shares in A and 100 in B.
Its aggregate is (900 × 0.04 + 100 × 0.12) / 1,000 = 0.048, below the first reporter’s 0.092. A headline ranking would favor the second despite worse results within each stated stratum.
Using a disclosed 50/50 standard mix, the first reporter’s conditional average is 0.06 and the second’s is 0.08. This analyst standardization is not an official replacement report or a causal estimate. It shows why group compatibility and mix matter.
Other differences in security, timing and order characteristics can remain even after these two strata are matched. Do not promote the standardized figure into proof of universally better execution or a broker recommendation.
Calculate price improvement using the stated benchmark
Use a separate illustrative two-execution case. For 100 buy shares, assume receipt NBBO of 9.97 by 10.03, midpoint 10 and execution at 10.01. Improvement relative to the offer is 0.02 per share, or 2 in total.
For 900 buy shares, assume receipt NBBO of 19.96 by 20.04, midpoint 20 and execution at 20.05. Execution is outside the offer by 0.01, or 9 in total. Net improvement is 2 − 9 = −7.
FAQ 38 describes the summary percentage improvement as net monetary improvement divided by the sum of midpoint times executed shares. Here that denominator is 100 × 10 + 900 × 20 = 19,000.
The result is −7 / 19,000 × 100 ≈ −0.036842%. A negative result is possible because outside-quote amounts exceed improvement. Do not discard adverse executions or average individual percentage improvements using a different method.
This example uses the stated NBBO-based calculation. It does not substitute that benchmark for the distinct best-available-displayed-price statistic. Receipt, executability and later quote times must be identified according to the particular order and metric.
Execute a comparable-population register
| Check | Record | Error prevented |
|---|---|---|
| Version | FAQ/rule context and applicable dates | Old and amended populations merged |
| Availability | Month, publication and cutoff | Future required report treated as observed |
| Reporter | Identity and market-center/broker role | Same name treated as same population |
| Orders | Security, type, size and exclusions | Unmatched averages ranked |
| Weight | Executed shares for the relevant metric | Received or unexecuted shares substituted |
| Benchmark | Quote source and reference time | NBBO and odd-lot benchmark conflated |
| Aggregation | Components, numerator and denominator | Row means averaged without weights |
Public reports provide aggregates, not the complete order-by-order route history or every customer instruction. Some detailed reconstruction inputs remain unavailable. Keep any assumptions about those inputs separate from reported values.
If files disagree, check reporter role, schema, report month, corrections, missing fields and units before calling one source wrong. A blank field is not necessarily zero; FAQ 1 also discusses when rows and fields may be absent.
Update the comparison when a corrected report appears, coverage changes or a later compliance stage begins. Preserve the previous version and its information cutoff. The conclusion should state the matched population and metric actually compared, with the remaining uncertainty visible.
Preserve execution timing as part of the comparison
An order’s first system capture, the time a limit becomes executable and the time of its fill can be different events. The FAQ’s timestamping section, questions 33–35, explains the applicable receipt convention and quote-time alignment. A comparison that silently starts one reporter’s clock later can appear faster without describing a comparable interval.
For an analyst worksheet, store these events in separate columns even when the public aggregate cannot populate all of them. Mark the unavailable event rather than reconstructing it from a reported average. An average fill time cannot reveal each order’s latency distribution or how many orders waited unusually long.
Use an independent arithmetic check for each transformation. The effective-spread example has numerator 92 and executed-share denominator 1,000. The improvement example instead has net money of −7 and midpoint-weighted money of 19,000. These denominators serve different statistics; copying the first denominator into the second would create −0.7% rather than approximately −0.036842%.
Publish an interpretation only after checking the report’s original units. A decimal ratio, percentage and basis-point display can describe the same result at different scales. Preserve the raw value and state the conversion explicitly, so a presentation change does not become a false execution-quality improvement.
FAQ
Is the simple mean of 0.02 and 0.10 the executed-share average?
No. With 100 and 900 executed shares, it is 0.092.
Were mandatory November odd-lot-benchmark reports available by October 25?
No. The stated collection starts November 1, and November reports have a later publication window.