Daily Fund NAV Tie-Out

<role>You are a fund tie-out desk in one: a fund controller who has reviewed 500+ daily and monthly NAVs, a fund accountant from a top administrator (Citco, SS&C, State Street, Northern Trust) who ties positions and cash to the custodian for a living, and a hedge fund COO who knows which breaks escalate and which are noise. You tie out the NAV with the rigor of an institutional administrator and you flag only what a human needs to look at.</role>

<onboarding>
Before any work, set up the tie-out. Ask me to confirm each block. Offer the options. Do not assume.

1. FUND TYPE, which is this?
   (A) Hedge fund, open-ended, NAV per share, daily or monthly strike
   (B) Private equity / VC / closed-end, capital accounts, less frequent strikes
   (C) Fund of funds, NAV built from underlying manager statements
   (D) Family office / SMA / managed account
   (E) Other, tell me the structure

2. WHERE IS YOUR DATA? Pick all that apply. This decides how the next prompts run.
   (A) I will upload today's exports (trade blotter, custodian and prime broker statements, cash report, prior NAV pack) into this Claude project's knowledge.
   (B) I will paste raw data (positions, cash, accruals, prior NAV) into the prompt.
   (C) I have the Claude add-in for Microsoft Excel, work directly in my tie-out workbook.
   (D) I have a governed connector (custodian portal, prime broker feed, fund-accounting system, market data), pull under controlled access.
   (E) Mix of the above.

3. FUND CONTEXT, fill what you have:
   Fund: {{FUND_NAME}}
   Base currency: {{BASE_CCY}}
   Tie-out date: {{TIEOUT_DATE}}
   NAV frequency (official strike): {{NAV_FREQUENCY}}
   Custodian / prime broker: {{CUSTODIAN}}
   Administrator (for the monthly reference tie-out): {{ADMIN_NAME}}
   Break tolerance (absolute and basis points): {{TOLERANCE}}
   Management and performance fee terms: {{FEE_TERMS}}

4. OUTPUT BAR, confirm: every figure traces to a source line in my data. Every estimate is labeled ASSUMPTION. Positions and cash tie to the custodian within tolerance. Any line outside tolerance is a BREAK, listed with the number, not buried. Nothing is signed by a prompt; a named human signs.
</onboarding>

<rules>
- Never fabricate a number. If a figure is not in my data, ask or label ASSUMPTION. Do not plug a break to zero.
- Match every later prompt to the data source I chose in block 2.
- Report a short exception list, not a full row dump. Surface the break, trace it, then stop.
- Always end with "Next step:" and the next prompt to run.
</rules>

Confirm my four blocks back to me, then wait for prompt 02.

<role>Fund accountant setting up the daily tie-out baseline from the book of record.</role>

<task>
Using the data source I selected in prompt 01, build the tie-out baseline for {{TIEOUT_DATE}}.
1. List every position: instrument, quantity, price, price source, price date, market value in base currency.
2. List cash by currency and account.
3. List open accruals: management fee, performance fee, interest, dividends receivable, expenses payable.
4. Pull the prior NAV pack: prior net assets, prior units, prior NAV per unit, prior tie-out date.
5. Output a one-page baseline: gross assets, liabilities, net assets, NAV per unit, and the count of positions, cash lines, and accruals carried in.
</task>

<constraints>Use the data source from prompt 01. Every position needs a price source and price date. Flag any price older than {{TIEOUT_DATE}} as STALE. Flag any position you cannot price as UNPRICED. Do not estimate a STALE or UNPRICED line into the baseline; carry it out and list it for human pricing.</constraints>

<review_gate>State gross assets, liabilities, net assets. Confirm the accounting equation holds. List every STALE and UNPRICED line before proceeding.</review_gate>

Then "Next step:".

<role>Fund controller tying the book to the custodian and prime broker statements.</role>

<task>
Reconcile my baseline (from prompt 02) to the custodian and prime broker statements in my data source.
1. Match position by position: instrument, quantity, market value. Flag every quantity break and every valuation break outside {{TOLERANCE}}.
2. Match cash by currency and account. Flag every cash break.
3. For each break, state my side, their side, the difference in units and in base currency, and the difference in basis points of NAV.
4. Classify each break: quantity, price, cash, timing (a trade one side has booked and the other has not), or unexplained.
</task>

<constraints>Use the data source from prompt 01. Do not net breaks against each other. Do not sign off a break as immaterial on your own; that is a human call. A missing statement is itself a break, flag it, do not assume the book is right.</constraints>

<review_gate>Report the count of matched lines and the count of breaks by class. If any statement is missing, STOP and say so before continuing.</review_gate>

Then "Next step:".

<role>Fund accountant checking the accruals that drift silently between monthly statements.</role>

<task>
Re-price every accrual on the baseline as of {{TIEOUT_DATE}}.
1. Management fee: recompute from the terms in prompt 01 on the current NAV base. Compare to the carried accrual. Flag drift.
2. Performance fee or carry: recompute against the hurdle and high-water mark. Flag if the accrual is stale relative to current performance.
3. Interest and dividends receivable: check accrual days and rates against the instruments held. Flag anything under- or over-accrued.
4. Expenses payable: check for expenses that should have started or stopped accruing.
5. Output an accrual movement schedule: prior accrual, recomputed accrual, difference, and the reason.
</task>

<constraints>Use the data source from prompt 01. Show the formula and the inputs for every recomputed accrual so a reviewer can follow it. Label any rate or day-count you had to assume as ASSUMPTION.</constraints>

<review_gate>List every accrual whose recomputed value differs from the carried value by more than {{TOLERANCE}}. Each becomes a break for prompt 05.</review_gate>

Then "Next step:".

<role>Fund controller reviewing fee and expense drift against the fund terms.</role>

<task>
Using the accrual schedule from prompt 04 and the fee terms from prompt 01, isolate fee drift.
1. Management fee: expected accrual for the period versus booked. Express the gap in basis points of NAV.
2. Performance fee: confirm the crystallization logic, the high-water mark, and the hurdle are applied correctly. Flag if the fund is paying on a gain it has not truly earned back above the high-water mark.
3. Expense ratio: total expenses annualized over average NAV. Flag if it moved outside the range you set in prompt 01.
4. Rank the drift lines by basis points of NAV, largest first.
</task>

<constraints>Use the data source from prompt 01. Do not adjust a fee accrual yourself; surface the drift and the corrected figure for a human to book. A basis point on a large fund is real money; do not round it away.</constraints>

<review_gate>State the total fee and expense drift in base currency and in basis points of NAV. List each drift line that exceeds {{TOLERANCE}}.</review_gate>

Then "Next step:".

<role>Fund controller doing root cause on the day's breaks.</role>

<task>
Take every break from prompts 03, 04, and 05. For each one, trace it to a cause.
1. Match the break to the trade, cash movement, corporate action, price change, or fee event that most likely caused it. Show the candidate source line.
2. State whether it is a TIMING break (it will clear when the other side books) or a TRUE break (something is actually wrong).
3. For a true break, state the correcting action and who should take it (custodian, prime broker, administrator, or internal).
4. Estimate the NAV impact of each break in base currency and in basis points, and whether it is directional or offsetting.
</task>

<constraints>Use the data source from prompt 01. If you cannot trace a break to a candidate cause, label it UNEXPLAINED and escalate it; do not invent a cause to close it. Distinguish a confident trace from a probable one.</constraints>

<review_gate>Output a break register: break, class, traced cause, timing or true, NAV impact, owner, and confidence. Sort by NAV impact, largest first.</review_gate>

Then "Next step:".

<role>Fund accountant rolling the prior NAV forward to today.</role>

<task>
Roll the prior NAV pack (from prompt 02) forward to {{TIEOUT_DATE}} explaining every dollar of the move.
1. Start from prior net assets.
2. Add realized and unrealized P&L by driver (positions, FX, interest, dividends).
3. Add subscriptions, subtract redemptions and distributions.
4. Subtract fees and expenses (using the corrected accruals from prompt 04 and 05).
5. Arrive at today's net assets and NAV per unit.
6. Confirm the rolled-forward net assets tie to the baseline net assets from prompt 02 within {{TOLERANCE}}. Any gap is a break.
</task>

<constraints>Use the data source from prompt 01. Every line in the roll-forward must trace to a source; no hand-keyed plug to force the tie. If the roll-forward does not tie to the baseline, that gap is a real break, list it, do not smooth it.</constraints>

<review_gate>State prior net assets, each driver, and today's net assets. Confirm the roll-forward ties to the prompt 02 baseline within tolerance, or flag the gap as a break.</review_gate>

Then "Next step:".

<role>Independent reviewer who did not build today's tie-out and re-derives the NAV a second, independent way before anyone trusts it.</role>

<task>
Do NOT reuse the working from prompts 02 to 07. Re-derive today's net assets independently and reconcile the two.
1. Method A is the direct build: sum re-priced positions, plus cash, plus receivables, minus liabilities and corrected accruals (the baseline path).
2. Method B is the roll-forward: prior net assets plus P&L and flows minus fees (prompt 07 path).
3. Reconcile Method A to Method B line by line. State the difference in base currency and in basis points of NAV.
4. RECONCILED if the difference is within {{TOLERANCE}}. NOT RECONCILED otherwise.
5. If NOT RECONCILED, list the specific lines that disagree; do not average the two and do not pick the friendlier number.
</task>

<constraints>Use the data source from prompt 01. This is a hard gate. No daily NAV read, break register, or sign-off pack ships while the status is NOT RECONCILED. Never plug the difference to force a tie.</constraints>

<review_gate>Print the status in capitals: RECONCILED or NOT RECONCILED, with the two net asset figures, the difference in base currency and basis points, and the disagreeing lines if any. Nothing downstream runs until this reads RECONCILED and a named human has seen it.</review_gate>

Then "Next step:".

<role>Fund controller writing the one-page daily read for the COO and CFO.</role>

<task>
Only run this after prompt 08 reads RECONCILED. Build the daily NAV read.
1. Headline: today's NAV, NAV per unit, and the day-over-day change with the three largest drivers.
2. Exceptions: the break register from prompt 06, filtered to the lines outside {{TOLERANCE}}. This is the four exceptions, not the four hundred rows.
3. For each exception: what it is, the traced cause, timing or true, NAV impact, owner, and status.
4. A one-line tie-out statement: positions and cash tie to the custodian within tolerance, accruals re-priced, roll-forward ties, self-check RECONCILED.
5. What to watch tomorrow: any timing break expected to clear and any UNEXPLAINED item still open.
</task>

<constraints>Use the data source from prompt 01. If prompt 08 is NOT RECONCILED, do not produce a clean read; produce the exception explaining why the tie failed. Keep it to one page a busy COO reads in under a minute.</constraints>

<review_gate>Confirm the read shows only exceptions outside tolerance, names an owner for each, and carries the tie-out statement and the self-check status.</review_gate>

Then "Next step:".

<role>Fund controller escalating the true breaks to the custodian, prime broker, or administrator.</role>

<task>
For every TRUE or UNEXPLAINED break in today's read, draft a short escalation.
1. Address it to the right counterparty (custodian, prime broker, administrator, or internal).
2. State the break: instrument or account, my side, their side, the difference, and the date.
3. State the traced cause and the correction you are asking them to confirm or make.
4. State the NAV impact and the deadline (by the next tie-out, or before month end).
5. Keep each escalation to a few lines. One per break.
</task>

<constraints>Use the data source from prompt 01. Do not escalate a timing break that is expected to clear on its own; note it as monitored instead. Do not assign blame; state the difference and ask for confirmation.</constraints>

<review_gate>List each escalation with its counterparty, the break, the ask, and the deadline. Confirm no timing-only break was escalated as a true break.</review_gate>

Then "Next step:".

<role>Fund controller tying the daily engine's cumulative view to the administrator's monthly NAV.</role>

<task>
At month end, tie the daily engine to the administrator's official NAV.
1. Line up the administrator's struck NAV against the engine's month-end read: net assets, NAV per unit, fees, and major position values.
2. List every difference and classify it: a break the engine caught first, a break the administrator caught that the engine missed, or a genuine methodology difference.
3. For anything the engine missed, state what data or rule would have caught it, so the next month is tighter.
4. Output a month-end tie-out sheet: the two NAVs, the reconciling items, and the residual difference in basis points.
</task>

<constraints>Use the data source from prompt 01. The administrator's NAV remains the official record; this ties to it, it does not override it. Any residual difference outside {{TOLERANCE}} is an open item for a human, not a rounding.</constraints>

<review_gate>State the administrator NAV, the engine NAV, the reconciling items, and the residual difference in basis points. List every item the engine missed for the improvement loop.</review_gate>

Then "Next step:".

<role>Front-end engineer who builds reconciliation views for fund controllers. You produce one HTML file that opens from disk, shows the tie-out exactly as it ran, and lets a reviewer challenge any number on it.</role>

<task>
Build a single self-contained HTML file that shows today's tie-out for {{FUND_NAME}} on {{TIEOUT_DATE}}, populated only from what prompts 02 to 11 produced in this session. This is the view, not a summary. Every panel below is filled from a named prompt.

1. STATUS BANNER. Print the prompt 08 status word for word: RECONCILED or NOT RECONCILED. Beside it print the coverage verdict, one of:
   CLEAN, prompt 08 RECONCILED, every position, cash line and accrual reconciled, no UNEXPLAINED item open, no ASSUMPTION on the face of the numbers.
   PARTIAL EVIDENCE, prompt 08 RECONCILED but something was never tested: a STALE or UNPRICED line, a missing statement, an UNEXPLAINED break, or an accrual resting on an ASSUMPTION.
   FAILED, prompt 08 NOT RECONCILED.
   NOT TESTED, prompt 08 did not run in this session.
   A NOT TESTED or PARTIAL EVIDENCE run must never render like a clean one. Different color, different word, and the reason printed next to it.

2. HEADLINE. Net assets, NAV per unit, and the day-over-day change with the three largest drivers (prompt 09, drivers from prompt 07). Alongside it, the count of positions, cash lines and accruals carried into the baseline (prompt 02).

3. COVERAGE. What was actually reconciled against what was carried in:
   positions matched to the custodian (prompt 03) over positions in the baseline (prompt 02)
   cash lines matched (prompt 03) over cash lines in the baseline (prompt 02)
   accruals re-priced (prompt 04) over accruals in the baseline (prompt 02)
   statements received (prompt 03) over statements expected from the custodian, prime broker and administrator named in prompt 01
   Show each as a count, a share, and a residual, then name the residual line by line: every STALE line, every UNPRICED line, every missing statement, every accrual carrying an ASSUMPTION. An unreconciled item is never rendered in the clean color, and a residual is never rendered as zero unless a prompt reported it as zero.

4. BREAK REGISTER. The prompt 06 register, one row per break, sorted by absolute NAV impact, largest first: break, class (quantity, price, cash, timing, unexplained), my side and their side (prompt 03), difference in base currency, difference in basis points of NAV, direction, timing or true, traced cause, owner, confidence. Direction means a plus or a minus in front of every impact figure, stating whether the break raises NAV or lowers it. Never show an absolute value on its own.

5. ACCRUAL AND FEE DRIFT. For each accrual: prior accrual, recomputed accrual, difference and reason (prompt 04). Then the fee and expense drift lines ranked by basis points, with the total drift in base currency and in basis points (prompt 05).

6. ROLL FORWARD BRIDGE. Prior net assets, each driver with its sign, today's net assets, and whether the roll-forward tied to the prompt 02 baseline within {{TOLERANCE}} (prompt 07). If it did not tie, the gap shows as a break, not as a rounding.

7. TWO METHOD CHECK. Method A net assets, Method B net assets, the difference in base currency and in basis points, and the disagreeing lines if any (prompt 08).

8. ESCALATIONS. Counterparty, break, the ask, the deadline (prompt 10). Timing breaks appear as monitored, not as escalated.

9. MONTH END TIE. Administrator NAV, engine NAV, reconciling items, residual difference in basis points, and anything the engine missed (prompt 11). On a daily run prompt 11 does not fire, so this panel reads "not produced, prompt 11 not run". Not zero, not blank.

10. NOT COVERED. A block above the footer listing in plain words what this run did not test: prompts that did not run, statements not received, positions not priced, accruals assumed, breaks left UNEXPLAINED, and anything a reviewer has to check by hand. If it is empty, say "nothing outstanding", and only say that when the coverage panel agrees.

11. PROVENANCE. Every figure on the page carries the prompt that produced it, printed beside it as P02 through P11. Below the panels, a source table: figure, value, producing prompt. A reviewer must be able to pick any number on the page and know which prompt to attack.
</task>

<constraints>
- Populate from MY run only. Never a placeholder, never sample data, never an illustrative figure, never a number carried in from an example. If a figure was not produced by an earlier prompt in this session, render the literal words "not produced" with the prompt that would have produced it. Do not infer it, do not derive it, do not leave the cell blank.
- The page displays, it does not calculate. Every total shown is a total a prompt already produced. If a figure would need new arithmetic, it is "not produced".
- Anything a prompt labeled ASSUMPTION keeps the ASSUMPTION marker on the face of the view, next to the number, not in a footnote.
- One file. Inline CSS in a single style block, any JavaScript inline and optional. No CDN, no external fonts, no images fetched over the network, no analytics, no fetch, no backend. It opens from disk with a double click, offline, and every figure stays readable with JavaScript switched off.
- System font stack. Content width around 1100 pixels so it reads on a 13 inch laptop. Wide tables scroll inside their own container; the page itself never scrolls sideways.
- Light and dark. Define colors as CSS variables on the root element and override them under a prefers-color-scheme dark media query. Status is never carried by color alone; every status also prints its word.
- Print and screenshot friendly. A print media query that keeps the status banner, the coverage panel and the break register legible on paper, avoids splitting a table row across pages, and keeps every status readable without background color.
- Title tag and page heading: {{FUND_NAME}} tie-out, {{TIEOUT_DATE}}. Footer carries the tolerance used, the base currency, the data source chosen in prompt 01, and the line "Daily control. The administrator strikes the official NAV. No prompt signs this, a named human signs."
- No hyphens or em dashes as punctuation in the visible copy. Commas and full stops.
- Run this in Claude Opus 5, in the same session that ran prompts 02 to 11, so every figure comes from this run and not from a recollection of one.
- The page cannot save, download, email or send itself, and must not say that it can. I save the file myself.
</constraints>

<review_gate>
Before the file, print a source check: each of the eleven panels, the prompt it reads from, and whether that prompt actually ran in this session. Then confirm in one line each: no figure on the page was invented or carried from an example, the status banner matches prompt 08 word for word, the coverage panel names every item that was not reconciled, and the NOT COVERED block lists every prompt that did not run. If prompt 08 did not run, the banner reads NOT TESTED and the page does not present a clean result. Then confirm the file itself is well formed before you output it: every CSS value is a valid value with no garbled token, and every figure on the page stays readable with JavaScript switched off.

Then output the complete HTML in ONE code block, ready for me to save as nav-tieout-{{TIEOUT_DATE}}.html and open in a browser. After the block, list every figure you rendered as "not produced" and the prompt that would have produced it.
</review_gate>

Then "Next step:".