Quantity reconciliation is the easy half. Two sources either agree on units or they do not, and the difference has a cause you can chase. Cost basis is harder, because the sources are not wrong when they disagree. They are answering the same question under different conventions, and there is no arithmetic that resolves a convention conflict.
On the book in the capture, the ACCOUNTS tile reads 8 connected sources against 4 unique positions, with a WALLETS tile reading 5 connected alongside it. That mix is the problem in miniature. The brokerage hands you a single average cost per instrument. The exchange hands you fills, each with a timestamp, quantity, price and fee. The wallets hand you balances and nothing else, because a chain records transfers rather than purchases and has no way of knowing whether an inbound 2 ETH was bought, earned, or moved between two addresses you both control.
Three sources, three answers to what did we pay
An average cost source has already destroyed information before it reaches you. Once fills are collapsed into a weighted average, the individual lots are gone, and with them the ability to identify which units you are disposing of. What you get is one number that is correct as an average and useless for lot selection.
A lot level source has preserved everything and pushed the decision onto you. Twenty fills across six months is twenty rows, each of which can be matched against a disposal under whatever selection method you adopt. That is more information and more work, and it produces a different unrealised figure per position than the average would.
A wallet source has nothing to give. Any basis on those rows is something you supplied, either reconstructed from the venue where the asset was originally bought or assigned by convention from a chosen date. Both are legitimate. Only one of them is your actual entry, and the difference between them has to be visible in your own records rather than buried.
The break shows up in P&L, not in units
This class of difference never trips a quantity reconciliation. Units agree at every venue, the market value agrees, and the only thing that moves is the number sitting between them. That makes it easy to miss for months and awkward to explain when it surfaces, usually in front of somebody who has just been told the same position is up 14 percent on one report and 9 percent on another.
When you see two unrealised figures for what looks like the same position, the useful reflex is to enumerate the reasons they can legitimately differ before assuming anything is broken. In practice there are five, and they are all mundane.
- Different basis conventions. An average across all lots and a lot selected disposal produce different remaining basis on a partially sold position, permanently, not just at the margin.
- Different populations. One figure covers a single account, the other covers the consolidated book. A view that includes or excludes a wallet, a paper account or a manually maintained account is a different set of rows and therefore a different number.
- Different measurement windows. A figure since inception and a figure over a selected period answer different questions about the same position, and on the performance view the period controls run from 1D through to 1Y.
- Different valuation inputs. Two marks from two order books at two timestamps, or two quote currencies converted at different FX rates.
- Different treatment of costs. Fees, funding and rewards capitalised into basis at one source and expensed at another shift the same position by whatever those flows amount to.
The way to settle which figure answers your question is to hold everything constant except one control and observe what moves. Fix the account selection and change the period. Fix the period and change the account selection. Compare the resulting figures against a single position you can calculate by hand from its own fill history. Three passes like that will tell you which population and which window each number on your screen is built from, and that knowledge is what you need before you cite either of them. Guessing at a formula and asserting it in a report is how a small ambiguity becomes a credibility problem.
Choosing a house convention you can defend
Pick one convention for the consolidated management book and write down why. Average cost across the consolidated position is the pragmatic choice for a book fed by sources of uneven quality, because it degrades gracefully. It survives a source that cannot supply lots, it is stable under re-derivation, and nobody can accuse you of selecting lots to flatter a number.
Lot level with a stated selection method is the better choice when the book is concentrated in venues that actually provide fills and when disposals are frequent enough that the choice of lot materially moves reported results. It costs you a mapping layer and a rule for what happens when a source stops providing lot detail.
What you must not do is let the convention vary by source, because then the consolidated figure is a sum of quantities computed on incompatible bases and it cannot be reproduced by anyone, including you. The test to apply is simple. Could a second analyst, given the same source extracts and the written policy, arrive at the same consolidated basis. If not, you have a preference rather than a convention.
Keep the tax question separate from this decision. The method used for tax reporting in your jurisdiction is set by rules that have nothing to do with what makes management reporting legible, and the two figures being different is normal. Documenting that they are different, and why, is far safer than forcing one to imitate the other.
Adopt from a date rather than restating history
The temptation once a convention is chosen is to push it backwards through the whole book. Resist it. Restatement on a book with wallet sourced positions means inventing history for exactly the rows where history does not exist, and every number you invent will be quoted back to you later as though it were observed.
The cleaner approach is an effective date. Choose a date, establish the basis at that date under the new convention, and run the new convention forward from there. Positions opened after the date are clean. Positions that straddle the date carry an opening basis that is stated as an opening basis, with the derivation attached. Anyone reading a return series that crosses the date sees a marked boundary rather than a silent change in methodology.
For the straddling positions, three derivations are defensible provided you say which you used. Reconstructed actual cost where the originating venue records survive. Weighted average of whatever lots you do have, with the unmatched quantity flagged. Or the mark on the effective date itself, which sets unrealised to zero on that row and starts the clock cleanly. The third one is the most honest option for wallet rows with no recoverable history, and it is only dangerous if the labelling gets lost.
Keeping the venue native record alive underneath
The house convention sits on top of the source records, it does not replace them. Retain the venue native extracts in their original form, retain the mapping from source symbol to internal instrument, and retain the transformation that produced the consolidated figure. Those three artefacts are what let you answer a question about a specific disposal eighteen months later without re-deriving the entire book from memory.
The practical test of whether the separation is working is the reconciliation you should be able to produce on demand for any instrument: venue reported basis, adjustment for convention, adjustment for capitalised costs, house basis, with each step evidenced. When that reconciliation exists, two different unrealised figures on the same position stop being an embarrassment and become a documented consequence of a policy you chose deliberately.