What a finalised snapshot actually keeps

The close begins with two checks

The engine refuses finalisation while any row in the selected period still needs review. It also refuses an empty eligible period and requires a nonblank run name. Passing those checks means the supported workflow can create a record from the current inputs. It does not independently verify that the source report is complete or that the supplied arrangement is appropriate.

For the fictional ten-row case, four review rows become resolved through the documented decisions. Eight rows contribute to the result, while two are excluded. The final commission is $255.00 and the eligible basis $660.00. This is the state we used for the snapshot investigation.

The record contains inputs as well as answers

The legacy example snapshot stores the inclusive period, basis policy and flat rules used for the calculation. It also stores reviewed rows, staff totals, net basis, commission, eligible fingerprints and the draft audit context. The run has an identity, label and creation time. A dated close adds a staff directory snapshot and the agreements selected by its sale dates. These are actual fields in the product’s saved-run structure, not a proposed feature list.

Excluded rows remain part of the explanation

The second A102 and the voucher A107 remain in the saved row list with their treatment and zero contributing commission. They are not part of the eligible basis or eligible fingerprint list. Keeping them explains why the raw file’s $1,056.00 gross did not become the close’s $726.00 included gross.

This has an important consequence for anyone rebuilding totals from detailed exports: filter by treatment and the documented calculation semantics. Summing every row’s stored basis without considering exclusions would include $300.00 of basis that does not contribute to the $660.00 result. Preserving evidence is useful only if its treatment remains visible.

Changing the draft does not change the copied record

We finalised the ten-row state, then changed the original draft object in three places: Mia’s service rate from 40% to 50%, the first row’s note and the basis policy from excluding to including tax. The saved snapshot remained byte-for-byte equal when serialized for the comparison. It still held Mia’s 40% rule, the original note, the excluding-tax policy and $255.00 commission.

The experiment checks that finalisation makes independent copies of the relevant data instead of retaining references to mutable draft objects. It does not claim that a saved file can never be edited by someone with direct filesystem access. That is a different property from avoiding accidental changes through ordinary draft editing.

The account server adds a separate protection

For account workspaces, the server recomputes a finalisation from its stored draft and assigns authoritative run identity and time. Ordinary draft saves must preserve the existing history. Closed runs are retained separately from editable draft state, and server activity records are independent of the client-supplied audit array. Those controls prevent a normal save from quietly rewriting a completed amount.

An explicit restore is a different operation with its own provenance and preserved pre-restore records. Do not describe restoration as ordinary editing of a past close. The sample demo is stored in the browser and does not provide the same server-enforced history protection as an authenticated account workspace.

A snapshot has boundaries worth naming

The row’s source snapshot preserves parsed CSV columns and values, not the original file’s exact bytes. Legacy duplicate matching includes a normalised staff name, so a changed source name can affect a comparison with an older run. Dated closes add a linked staff-ID key for exact repeats across aliases, but neither model detects every altered source line. Older workspaces keep flat rates until explicitly upgraded; dated mode selects an agreement by each sale’s date and snapshots the chosen agreement.

The snapshot also does not prove that anyone was paid. It is a preserved commission calculation and review record. These limits do not make the record unhelpful; they tell the reviewer which questions it can answer and where to look for separate source, identity, payment or arrangement evidence.

Inspect one preserved decision, then one preserved rule

Use the snapshot inventory download to follow the fields in a completed example. Start with the excluded A102 copy and confirm its zero contribution and retained reason. Then inspect Mia’s stored rate and compare it with the rate used by a later draft. A later change should not retroactively alter the earlier result.

We executed the controlled copy experiment on 27 September 2026 and checked the expected ten rows, eight eligible fingerprints, $660.00 basis and $255.00 commission. The useful outcome is not simply a locked-looking screen. It is a record whose source, assumptions and decisions remain available when someone asks how the final number was reached.