Snapshots are periods, not points

A snapshot is in force from its date until the next one. Editing the 1 July snapshot changes what the model is considered to have held for all of July.

Correct existing against New rebalance

Correcting replaces a snapshot. New rebalance adds one from a date you choose. The line under the switch states which is about to happen and what it supersedes.

Reprice shares

Recalculates share counts at the as-of date prices, in the grid only. Nothing is written until you save.

Daily returns are never changed here

Composition and performance are separate records. Use the Returns editor for figures.

Values are decimal fractions

0.0042 is up 0.42 percent. Anything beyond 50 percent is refused as a typo, because a percentage typed where a fraction belongs would corrupt the index permanently.

Saving rebuilds the index

The index is a running product, so changing a return in 2019 changes every figure after it. The rebuild happens automatically from the earliest date you touched.

Missing sessions

Trading days absent from the loaded window are listed above the grid. A gap is invisible in a list of dates, so it is computed against the market calendar.

Add a missing date

Refused on weekends, market closures and dates that already exist. Only for a session that was genuinely never written.

Returns

File format

Kwanti return exports. One model per file, or one wide sheet with a column per model. A date column plus either a return or an index value; where both are present the return is used and the index checked against it.

Or the Master Returns Sheet, which is where the hedged models come from — they are not in Kwanti and never will be. Its Hedged and Enhanced tabs carry fourteen models each, as daily decimal returns rather than index levels. Any other tab in that workbook is ignored.

The two are told apart from the file itself, by whether it has a Hedged or Enhanced tab. There is nothing to select. Declaring the format would be one more thing to get wrong on a monthly job, and getting it wrong would not fail — it would read returns as index levels and produce plausible nonsense.

Hedged model columns

The Master Sheet's headers are not always model codes: two versions of the same workbook are in circulation, one using HTC-100-0 and one using HETF 100. Both are mapped in dbo.MasterReturnsColumnMap, so either works. A column that is not mapped is reported and skipped, never guessed at from a resemblance.

A blank cell is not a zero. These series start at different dates, and writing zero for a period a model did not exist would put a flat line in its history and drag every trailing return computed across it. Blanks produce no row.

Dates

Read from the file. Weekends and NYSE closures are filtered out automatically, so an export that includes them will not write rows on days the market was shut.

What happens on commit

Rows are appended. Existing dates are reported as conflicts in the reconciliation and are not overwritten unless you choose to.

Holdings

File format

Two columns: ticker and weight. Nothing else is read. CSV or XLSX. Headings are recognised from a list that includes Ticker, Symbol, Weight, Target and Allocation; if none are recognised the columns are inferred from the data and the preview says so, so check it.

Sheet names

Ignored. The first sheet is read. If a workbook has more than one, the preview warns you which was used and how many were skipped, because a Kwanti export with a summary tab first would otherwise import the summary.

Which model

From the filename, matched against the model register: SAS.csv, Core60_40.xlsx. Case does not matter and extra text is tolerated, so SAS (1).csv and core100_0 export.xlsx both work. A name matching no model, or two models, is refused rather than guessed.

Which date

From the form, once, for the whole batch. Deliberately not from the file: a wrong ticker fails at lookup and a wrong weight fails the balance check, but a wrong date succeeds silently and rewrites what a model held for a whole period. If models rebalanced on different dates, run one batch per date.

Percentages or fractions

Either. The scale is detected from the column total: near 100 is percentages, near 1 is fractions. Anything else is rejected rather than guessed, because being wrong here is wrong by a factor of a hundred.

Weights must balance

To 100% within 0.05%. Sixty equal positions at 1.6667% come to 100.0020% and are fine. A file at 99.2% is refused with its total shown, because a snapshot that does not balance corrupts every return computed from it.

Shares

Derived, never imported: floor(notional x (1 - cash) x weight / price) at that date's close. The same formula the Holdings editor uses, so an import and a manual rebalance of the same composition produce identical rows.

Duplicate tickers

Combined, and the preview says which. Two rows for the same symbol usually means two lots of one position, and keeping only one would understate it.

What happens on commit

The composition at that date is replaced entirely, not merged. A snapshot is in force until the next one, so this changes what the model is considered to have held for the whole period after it. Daily returns are not changed.

All or none

If any file fails, nothing is written. A rebalance where eleven of twenty models updated is harder to unpick than one that never ran.

Both modes

Nothing commits without review

Files are parsed and reconciled against what is stored, and you approve before anything is written. The commit sends back what you reviewed rather than re-reading the files, so what was on screen is what lands.

Limits

40 files per batch, 2 MB each.

Audit

Every write is logged with your name, the file it came from and the row count. Returns go to the returns edit log, holdings to the holdings edit log.

What this page is for

Every data quality check in one place. A section only appears when it has something to report, so an empty page is the goal rather than a sign the checks did not run.

The tiles across the top

Integrity issues and Models with gaps are blocking. Stale models counts series that have stopped updating. Awaiting lookup counts symbols queued for provider enrichment, which clears itself. Checked is when this snapshot was taken, not when the pipeline last ran.

Blocking: fix before publishing

Holdings Integrity covers two unrelated conditions in one table, which is why the Why listed column exists. Weights not totalling 100 percent is an error. Flagged rows are migration review items, not errors. Return Series Gaps and Missing Sessions are holes in the middle of a series, invisible everywhere else and quietly wrong in every trailing calculation. Index Does Not Match Returns means the stored index has drifted from what the daily returns imply, so one of the two was written without the other. Duplicate Holdings is the same symbol twice in one snapshot. Active Models With No Composition is a model advisors can reach that has nothing to show.

A decision, not an error

Returns on Market Closures lists rows dated to days the market was shut. There is no defensible default, so nothing is filtered automatically. Only the non-zero rows change any calculation. Review opens the Returns editor at that model with closure dates flagged, so the decision stays visible in the data afterwards rather than disappearing into it.

Pipeline state

Nothing written for N sessions means the nightly append has not run or has been failing silently, and is the first thing to check when several other sections light up at once. Models Behind and Stale Return Series are individual series that stopped while others kept going, which usually points at the model rather than the pipeline.

Securities

Incomplete Security Details and Holdings Not In Security Master are the securities editor, documented in its own help block in that section. Neither is cosmetic: a holding with no name renders as a bare ticker, and one with no asset class drops out of the allocation chart entirely.

Worth a look, not necessarily wrong

Large Daily Returns flags outliers, which are usually real and occasionally a decimal point. Incomplete Model Setup lists register fields left empty, which affects display rather than arithmetic.

Every row links to its fix

The button opens the tool that corrects that problem with the model and date already filled in.

What this page is for

The two lookup tables the model form reads. Both used to be hardcoded in the console, which meant a new hedge level or a changed risk band needed a code change and a deploy.

Hedge levels

A hedged model's fee is joined from this table through its level, never stored on the model row. So a repricing is one edit here instead of a figure typed on sixty rows, where the failure mode is the handful you miss quietly charging the old rate.

The fee is charged on the equity portion and is separate from the model fee. HUSQE pays 27 basis points of manager fee and 20 of hedge, and the calculator shows both, so this is not an either/or.

The code prefix builds the offering code the advisors see: H plus SAS is HSAS. Two levels cannot share a prefix, because that would generate two offerings claiming one code.

A level cannot be deleted while any model carries it. Adding one makes it selectable when creating a model straight away — no deploy.

Risk bands

The StratiFi score range each classification covers. Before this table the boundary between Moderate and Moderate Aggressive existed only in the head of whoever was assigning it, and a model whose StratiFi score disagreed with its stored classification was not something anything could detect.

Bands are contiguous at 0.1, which is the granularity StratiFi quotes. 3.3 followed by 3.4 is not a gap: no score falls between them. Overlaps are refused, because a score has to map to exactly one classification.

Renaming a band carries every model on the old name with it. Nothing in the database enforces that link, so the rename is done here or not at all.

Deactivating hides a band from the selector but keeps the name valid for models already carrying it. Deleting is refused while any model is classified that way.