Sometimes a data source never records a fund's share split. The price series then shows a cliff. A fund may appear to have "fallen" 75% overnight on a 4:1 split, and every return computed across that date is garbage.
A manual split is the repair: you record the date and ratio, and the app applies the correction as a read-time price factor on the split's side of history. The stored downloads stay exactly as the source delivered them.
Enter the ex-date: the first day the fund traded at the new price, which is the day the cliff appears in its price history. Your broker may show the new shares a day or two either side of it; a date within a week still counts as the same split (see below), but the ex-date is the one the prices step on.
Why read-time, not rewrite
- The raw series remains a faithful copy of the source, so re-downloading or switching price sources can never double-apply a correction that was baked into the rows.
- The fix is reversible and auditable: delete the manual split and you're back to exactly what the feed said.
The app can also detect likely missed splits from your own ledger. The signal is a trade's execution price against the stored close for that day: a trade fills at the as-traded price, so a close that is a suspiciously round multiple smaller (or larger, for a reverse split) says a split is missing from the feed. It reads each stored close the way every value does, through the splits you recorded by hand, so a split you already fixed is never flagged again. A reinvested distribution can corroborate the ratio at the moment the split is recorded, but it is not what finds one. The Overview banner shows the date and ratio it derived and asks you to confirm before writing it.
The date matters as much as the ratio: a split recorded after the day your ledger's share count changed values every day between the two at the wrong share basis. So the date comes from your own records: the distribution that delivered the new shares, measured against each account's own holding, or the ledger row that books the split's shares. Rows that belong to a split already recorded are never taken for this one, and if the trade prices and the distribution disagree about the ratio, nothing is recorded. If your ledger never booked the split's shares at all, the app dates it the day after your last transaction in that security, so every day it has recorded keeps its value; until you add the split's shares (in Transactions or with Reconcile), the holding reads short from that day. If a trade changed your share count after the split and no row says when it happened, the app says so and records nothing, because any date it picked could be wrong.
If your trades later disagree with a split the app detected earlier (one an older version dated after your last trade, for example), the banner names that split instead of offering a new one opposite it. Review split explains what is wrong with it and changes nothing.
Dismissal means "hide this evidence for now," not "this ticker is fixed." The exact candidate is hidden for 180 days using an opaque SHA-256 database scope plus all of the detector evidence; the saved value never exposes the portfolio file path. A changed ratio, date, compared price or corroborating sample, or the same ticker in another database, remains a different visible candidate. Legacy ticker-only dismissals are discarded so an old broad choice cannot hide new evidence.
A detected split is recorded a little differently from one you type in yourself: it scales share counts but leaves the close alone (the source already adjusted the price but missed the split row), and it is kept through a price rebuild, because a re-fetch can never restore a split the provider never reported. The trade-off is that it doesn't appear in the Manual splits list on Securities & prices, so there's no one-click undo. That is why the Overview asks first. If the source later reports the same split, on the detected date or within a week of it, the app keeps the detected one and does not record the source's beside it, which would scale your share counts twice. It says so only when the two ratios differ.
When a date is already taken
A date can hold only one split. If the one you type falls on a date that already carries a downloaded or detected split, that row is left alone and the save says so, naming the date and which kind of split is sitting there. The rest of your list still saves. This is deliberate: the three kinds correct prices differently, so silently replacing one with another would change every value on that side of history, and a detected split can never be downloaded back once it's gone. Either remove that date from your list, or delete the existing split first and then record yours.
When the source later reports your split
Sources catch up. One day a source may start reporting a split you recorded by hand, and adjust its own price history for it. Your split would then correct prices the source has already corrected. A 2:1 split would halve every value before its date, and a reverse split would multiply them.
So when a price update sees the source report a split on the date you recorded, or within a week of it, it treats the two as the same split. It keeps yours and does not record the source's, because two would scale your share counts twice. It also leaves the prices it already has for the days before the source's split date as they are, and does not store the source's adjusted prices for those days; a day it had no price for stays empty until the rebuild below. Prices from the split date on are stored as usual, because both versions of history agree there, so the ticker keeps updating. The update's status line shows split to review, naming the ticker, the source's date and yours if it differs; if the two ratios differ, it names both and asks you to check which is right. The line clears after a few seconds and comes back with every update until you finish the change-over.
To finish the change-over, remove your split for that ticker on Securities & prices, then rebuild that ticker's price history. The rebuild stores the source's split and its adjusted history together.
Prices you typed before a split
A statement price you typed for a holdings snapshot or a new account is the price that day, before any split that came later. The app never rewrites it. When a split is recorded after its date, of any kind, every view reads that price on today's share basis: a $99 statement price before a 2:1 split counts as $49.50 per post-split share, so the statement day is still worth the $990 the statement showed rather than twice that. Going the other way, the price the app fills in for a statement date before a split is the price that day ($495 for a stock now trading near $49.50 after a 10:1 split), so the share count you enter is the one on the statement, and the cost it records is the statement's.
Earlier versions filled in a statement date before a split at a different price, and the opening cost they recorded can be off by the split's ratio. When an account holds such a lot, its Performance view lists it under Review: opening cost on the wrong split basis, with the recorded cost and the cost at that day's price. Nothing there changes the lot. If the recorded cost is not your statement's, edit the lot in Transactions.
When stored prices are on the wrong side of a split
Prices saved by an older version, or saved before a split reached the app, can sit on the wrong basis. The Overview looks for three shapes in the downloaded prices of the funds you hold, and keeps one card per fund, listing each problem it found, until it's fixed or you dismiss it.
Not every factor a source lists is a split. Its split list also carries adjustments: factors by which it moved its earlier prices for a distribution, a spin-off or a small stock dividend (IBM's 4.6% for the Kyndryl spin-off in 2021, for example). A downloaded factor counts as a split only when it is a ratio of small whole numbers, such as 2:1, 3:2 or 1:13; anything else is an adjustment, and the card says "adjustment", never "split". An adjustment moves prices, not shares: your share counts and cost lots are never scaled by it. Prices before it stay on the source's adjusted basis, so a holding's value before an adjustment reads a little below what it traded at that day, without a jump at the date. A split you record yourself, or one the app detected, is always a split. An adjustment of less than 1.5% never raises a card at all. How far off a problem leaves a value is given as a percentage when it is small:
- Prices before a split were never adjusted for it. The split was recorded after some earlier prices were saved, so those prices still sit at a different level, by about the split ratio, from one day on. The fund's value before that day reads too high (or too low, for a reverse split). Rebuild history rebuilds that one fund at once, and the card says what that does before you click: one rebuild covers every split listed on it. It replaces every downloaded price. A price you typed is kept only on a date the source has no price for; on a date the source covers, its price replaces yours. Your transactions and splits stay as they are, and a rebuild can't be undone from here. A market move as large as the split can look the same, so if the jump was real, dismiss the card instead. When the rebuild finishes the card becomes one line saying so. If the rebuilt history still jumps there, that jump is in the source's own prices: the line names the date, and the card goes away for good. With two or more such cards, Rebuild all in the group's header does them one fund at a time.
- Your split may now be applied twice. The downloaded prices no longer jump near the date of a split you recorded, so the source has adjusted them and your split corrects them again.
- Two splits recorded for one event: yours, and the source's or the app's, within a week of each other. Share counts before both dates are scaled twice.
For the last two, Review splits opens the fund on Securities & prices: remove your split there, then rebuild its price history. The app never fixes stored prices on its own, because which basis a stored price is on is inferred from the prices themselves rather than known.
Don't "fix" a missed split by editing prices or share counts directly: the correction then lives in two places, and the next data refresh quietly breaks one of them. The manual-split row is the one designed home for it.