ViskaStrat · Fund strategy & metrics
July 2026 — the whole book, traded by theme
Every fill the fund executed in July 2026, clustered into the fund’s own themes, with one trading report per theme.
Revision 3 — the document declared no unit
ViskaFront refused to render rev 2 (FE#48 comment 5166445601): every figure was published as a bare number with no currency at any path. Correct refusal — the unit is the publisher’s to state, and inferring it is not a consumer’s call. Fixed at the mechanical layer rather than by writing “USD” into the prose:declared_currency() reads the extract’s own
currency column, analysis.json carries a top-level currency, and the
derivation refuses on a mixed set instead of picking a winner, under three red cases. The finding worth
keeping: June is not single-currency, and only stays admissible because its 51 CAD rows carry
zero realized.
Revision 2 — corrected after ViskaFront review
Rev 1 was reviewed by ViskaFront before build (FE#48 comment 5164641124) and two blockers were returned. Recorded here rather than silently rewritten. (1) The major rollup was wrong — it was authored from a prose summary of the 2026-07-10 CIO meeting instead of read from the operator-ratified registry cut of 2026-07-11 (fund_themes.major). The
registry puts uranium under Minerals, names the fourth major Power
Generation, and has no member for a large-cap AI equity outside memory and power, so
GOOGL and TSLA band Out of thesis. Totals unchanged; three clusters moved; the AI headline
moves from −268,262 to −276,441. (2) The strat-provenance figures
were wrong in both halves — corrected below and now derived in code rather than typed.
Two further corrections: MYRG’s −20.7% is legitimate and the code was wrong to suppress it,
and round-trip returns now live in the data rather than only in prose.
What this is, and what it is not
This is the trading record, not a performance statement. Realized P&L is the broker’s own realized field. Unrealized moves on positions still held are shown as marks and are never folded into a return. There is no NAV, no time-weighted return and no benchmark here. Nothing is modelled, estimated, or priced from an outside data source — and where the source extract disowns a figure, this report states no figure rather than a reconstructed one.Unit of account
Every monetary figure in this report is denominated in US dollars (USD). The unit is read from the extract’s owncurrency column, not assumed: all 199 July fills and all
346 July marks carry USD, and every June row carrying a nonzero realized carries
USD. The extract is native and unconverted and publishes no FX rate, so nothing here is
converted. June’s 51 CAD rows (LUN and the USD.CAD pair) carry 0.00 realized, so no
second unit reaches a figure — and the builder refuses to emit rather than total two units if
that ever changes.
01
The month in one move
July was one decision executed over eight trading days: the fund sold its AI complex and bought the metals and energy complex.
The realized loss is not spread across the book. It is one theme.
| Major (ratified 2026-07-11) | Bought | Sold | July realized | June realized these names only |
|---|---|---|---|---|
| Minerals | 3,144,528 | 387,685 | +50,984 | +149,374 |
| AI | 517,500 | 3,088,920 | −276,441 | +642,859 |
| Power Generation | 1,260,308 | 1,152,620 | +35,934 | 0 |
| Energy Infrastructure | 840,884 | 0 | 0 | +191,922 |
| Out of thesis GOOGL, TSLA | 0 | 748,245 | +8,179 | +7,462 |
| Total | 5,763,220 | 5,377,470 | −181,344 | +991,617 subset |
Majors, member lists and band order are the operator-ratified registry cut of 2026-07-11
(fund_themes.major), not a mapping authored here. ai-megacap — this
seat’s own slug for GOOGL and TSLA — has no member list in that registry, so it bands
visibly as Out of thesis rather than being folded into AI.
The June column is a subset and does not foot to the June book
It covers only the 21 names July traded. June’s full realized was +902,183 across 24 names; the eight it traded and July did not — AMZN, AUD.USD, FNV, GEV, LUN, SBSW, USD.CAD, WPM — were net −89,433, which is the whole difference. Totalling the June column and calling it June overstates the month by that amount.analysis.json
carries both figures under separate keys so a surface cannot foot the column into the wrong claim.
The sequencing is the finding
The fund bought the replacement before it sold the original — 2,243,028 into gold and silver on 07-01, five sessions before the 2,715,420 memory liquidation on 07-06. The rotation was funded from cash first and reimbursed second. It was not a reaction to proceeds arriving.02
The break, in the broker’s own marks
The four memory and storage names, each indexed to 100 at the 2026-06-30 mark, through to the 07-06 exit price. The blue marker is the fund buying more MU on 07-01; the red markers are the whole sleeve going on 07-06.
03
The eight trade days
Net cash flow per trading day. Below the line is capital going out, above the line is capital coming back. Every other July weekday carries no trade for this account — those are genuine non-trading days, not gaps in the record.
04
The theme reports
Ordered top-down: the exit first, then the destinations by capital committed.
AI Memory & Storage — the exit
MU · SNDK · STX · WDC | 18 fills · 07-01, 07-06 · sold 2,715,420 · bought 517,500
| Name | 06-30 mark | 07-03 mark | exit price | move to exit | realized |
|---|---|---|---|---|---|
| SNDK SanDisk | 2,273.73 | 1,745.00 | 1,723.00 | −24.2% | −59,120 |
| MU Micron | 1,154.29 | 975.56 | 973.35 | −15.7% | −73,922 |
| STX Seagate | 965.00 | 820.16 | 855.00 | −11.4% | −78,428 |
| WDC Western Digital | 638.72 | 539.00 | 570.40 | −10.7% | −34,698 |
The behaviour, in order. On 07-01 at 15:59:59 the fund bought a further 500 MU at 1,035 — after MU had already fallen from the 06-30 mark of 1,154 to a 1,032 close. That is an add into the break, doubling the position from 500 to 1,000 shares.
Five sessions later, on 07-06 between 13:37:07 and 13:45:35 — eight and a half minutes — MU, STX, WDC and SNDK were all sold. A final MU tranche went at 19:57 at 959. The sleeve was gone in one sitting. The add at 1,035 was closed at an average of 973: the decision to average down cost roughly 6% on 517,500 of fresh capital and lasted five sessions.
Verdict on the exit itself: fast and complete once taken — four names, one session, no partials left behind. What it was not is early. The entire 10 to 24 percent draw happened before the sell.
AI Power & Electrical Equipment — one line, one exit
BE Bloom Energy | 1 fill · 07-08 · sold 373,500
Held 1,500 shares into July, marked 302.70 at 06-30. Sold in a single fill on 07-08 at 249.00 — −17.7% against the month-open mark, the worst carry move of any name in the book. Two sessions after the memory sleeve went, and the same trade in substance: the datacentre power leg of the AI capex stack, closed once the semis leg was closed.
Theme assignment
BE is not in the fund’s classification store — it is named among the unmapped in the open question the theme map covers only the held book. Placing it under AI Power is this seat’s reading, and it is the one theme label in this report a reader is entitled to reject.AI Megacap Equity — the profitable half of the AI exit
GOOGL · TSLA | 8 fills · 07-13 · sold 748,245
Both sold on 07-13, one minute apart (TSLA 10:21:26, GOOGL 10:22:29). One decision, not two.
| Name | 06-30 mark | exit | realized | after |
|---|---|---|---|---|
| TSLA Tesla | 420.60 | 393.50 | +5,927 | flat — 1,000 of 1,000 sold |
| GOOGL Alphabet | 357.37 | 354.75 | +2,252 | half kept — 1,000 of 2,000 sold, marked 356.13 at 07-31 |
Both were June re-entries, so both sold at a small realized gain despite falling against the 06-30 mark. GOOGL is the only AI name the fund still holds at month end, and the only one where the exit was deliberately partial — consistent with the CIO roster describing it as “the single direct large cap AI/tech equity exposure outside of memory and power.”
The AI major in total
−276,441 realized · 3,088,920 returned to cash. GOOGL and TSLA are not in it.ai-megacap is this seat’s own slug and the ratified registry
has no member for a large-cap AI equity outside memory and power, so the pair band
Out of thesis — visibly, carrying their +8,179 and their 748,245 of proceeds,
never folded into AI and never dropped. The surviving 1,000-share GOOGL stub sits in that band.
Gold / Precious Metals — the largest single commitment of the month
B · NEM · AUGO | 20 fills · 07-01 only · bought 1,332,295 · sold 0
The whole position was established in one session — the same day the AI complex was breaking.
| Name | shares | avg entry | 07-31 mark | status |
|---|---|---|---|---|
| B Barrick Mining | 15,000 | 36.63 | 36.73 | added to 4,500 held → 19,500 |
| NEM Newmont | 5,000 | 93.24 | 93.71 | added to 2,500 held → 7,500 |
| AUGO Aura Minerals | 5,000 | 63.34 | 54.57 | new position, −13.8% |
B and NEM finished July effectively flat on the month — B’s 07-31 mark equals its 06-30 mark to the cent, after ranging 34.84 to 38.21 in between. AUGO is the one clean read in the cluster: flat at 06-30, bought at 63.34, marked 54.57 at month end. That is a complete cost basis and a real −43,850 open loss.
Read: the gold leg was sized decisively and instantly — 1.33M in a single session, no scaling in. At month end it has neither worked nor broken, except AUGO, which has broken.
A disagreement, stated rather than resolved
The fund’s allocation map buckets B as Fjölbreytt (diversified). The position page describes it as a core gold miner. This report follows the position page and flags the disagreement rather than silently picking one.Oil Cycle (Short Term) — the one completed, profitable round trip
CVX · XOM · RIG | 70 fills · 07-08, 07-20 · bought 1,260,308 · sold 1,152,620
The most disciplined trading of the month, and the only cluster where a full round trip opened and closed inside July.
| Name | in | out | days | realized | return |
|---|---|---|---|---|---|
| CVX Chevron | 07-08 · 3,000 @ 178.22 | 07-20 · 3,000 @ 187.46 | 12 | +27,667 | +5.2% |
| XOM Exxon Mobil | 07-08 · 4,000 @ 145.48 | 07-20 · 4,000 @ 147.56 | 12 | +8,267 | +1.4% |
| RIG Transocean | 07-08 · 27,750 @ 5.18 | still held | — | 0 | +2.7% (mark 5.32) |
CVX and XOM were bought and sold on the same two dates, in full, at size. Neither was held into month end. RIG — 27,750 shares of offshore drilling for 143,718, the largest share count and one of the smallest tickets in the book — was kept, and is the only new July position carrying an open gain at 07-31.
Read: this cluster was traded, not allocated. Twelve-day hold, both legs closed clean, +3.2% on the 1,116,591 deployed into CVX and XOM. It is the counter-example to the AI exit: same month, same book, a position taken with a defined exit and taken off at it.
Theme assignment
CVX, XOM and RIG are all unmapped in the fund’s classification store. The Oil Cycle label is this seat’s.Electrical Construction & Grid — bought only, and the worst month-end marks
POWL · MYRG | 24 fills · 07-07, 07-13 · bought 840,884 · sold 0
| Name | July buys | avg entry | 06-30 mark | 07-31 mark | move |
|---|---|---|---|---|---|
| POWL Powell Industries | 1,500 @ 234.55 (07-07) + 750 @ 232.01 (07-13) | 233.71 | 286.36 | 208.68 | −27.1% carry |
| MYRG MYR Group | 750 @ 420.06 (07-07) | 420.06 | not held | 333.22 | −20.7% vs entry |
POWL was quadrupled — 750 shares held at 06-30, 3,000 held at 07-31 — while the mark fell from 286.36 to 208.68. The second tranche on 07-13 was bought at 232.01 into a mark of 225.66, already down 21% from month open. MYRG was a fresh 315,046 position on 07-07 that lost a fifth of its value by 07-31, with the damage concentrated in the last week (07-23 mark 407.40 → 07-31 333.22).
Read: the only cluster where the fund added into a falling mark twice and closed the month with nothing realized and both legs deeply underwater on paper. Same behaviour that cost 6% on the MU add, run again. On the July tranches alone — the only lots whose basis this report will state — that is −56,317 on POWL and −65,130 on MYRG.
Basis caveat, load-bearing
POWL underwent a 1-for-3 stock split during the holding period and MYRG carries an unresolved cost basis in the source extract. This report therefore quotes the broker’s own marks and this seat’s own July entry prices, and computes no reconstructed cost basis for either. The percentages above are mark-to-mark and entry-to-mark only, and the 750 POWL shares carried in from June are excluded from the loss figure.Copper Deficit — the best-executed trade of the month
SCCO · FCX | 28 fills · 07-07, 07-17, 07-22 · bought 640,760 · sold 387,685
| Name | action | price | mark context |
|---|---|---|---|
| SCCO Southern Copper | +3,000 on 07-07 | 168.65 | mark 169.75 that day, doubling 3,000 → 6,000 |
| SCCO | −2,000 on 07-22 | 193.84 | mark 175.07 (07-20) → 188.01 (07-21) → 195.48 (07-22) |
| FCX Freeport-McMoRan | +2,300 on 07-17 | 58.61 | 7,700 → 10,000 held; 07-31 mark 62.63 |
SCCO was added to at 168.65 and trimmed fifteen days later at 193.84, into a two-session +11.6% spike — sold on the spike day itself, not after it. The fund kept 4,000 shares, marked 182.71 at month end, above both the entry and the month-open 174.26. FCX was a straight add on 07-17 at 58.61 against a 07-31 mark of 62.63.
Read: buy the dip, size it, sell the spike, keep the core. The only cluster in July where the fund traded around a position and was paid for it.
Basis caveat
SCCO had a corporate action inside the window; the source extract places any reconstructed basis for it outside the reliable set. The +50,984 quoted is the broker’s own realized field, not a reconstruction, and it is FIFO against pre-July lots — so it is a realized gain of the position, not of the 07-07 tranche.Aluminum Tightness — bought into weakness, no exit
AA · CENX | 8 fills · 07-07, 07-08 · bought 606,640 · sold 0
| Name | July buy | avg entry | 06-30 mark | 07-31 mark | carry move |
|---|---|---|---|---|---|
| CENX Century Aluminum | 8,000 on 07-07 | 45.38 | 46.01 | 44.77 | −2.7% |
| AA Alcoa | 5,000 on 07-08 | 48.72 | 52.14 | 45.26 | −13.2% |
Both positions were roughly doubled — CENX 8,000 → 16,000, AA 9,000 → 14,000 — and neither was touched again. AA was bought at 48.72 after a fall from 52.14 and finished at 45.26, so the add is underwater by about 7% and the pre-existing half by more.
Read: conviction adds into weakness, held without a stop and without a trim, ending the month down. Same posture as the grid cluster, smaller size, less damage.
Silver — a June exit re-bought lower
HL Hecla Mining | 20 fills · 07-01 · bought 393,233 · sold 0
The fund had already round-tripped HL in June for a clean profit — bought 50,000 at 14.68, sold 50,000 at 16.21 — and was flat entering July. On 07-01 it re-established half the old size: 25,000 shares at 15.73, below the 16.21 June exit and above the 14.68 June entry.
07-31 mark: 14.12, an open loss of −10.2% (−40,250). The position ranged 16.46 on 07-06 down to 14.12 on the final mark.
Read: the June trade was the model — buy, run, sell into strength. The July re-entry repeated the buy without the sell, at a worse price than the level the fund itself had bought at six weeks earlier, and it has not worked.
Uranium / Nuclear Renaissance — a single add, into a fall
CCJ Cameco | 2 fills · 07-17 · bought 171,600 · sold 0
2,000 shares at 85.80 on 07-17, taking the position from 3,000 to 5,000. The 06-30 mark was 101.86; the 07-31 mark is 86.38. The pre-existing 3,000 therefore carried −15.2% through July — roughly −46,440 — and the add was made near the bottom of that fall, finishing marginally up at 86.38.
Read: the smallest ticket of the month, and the only add that landed at a better level than the position it was averaging into.
05
What the record shows
- One major carried the entire realized loss, and more of it than the book shows. AI: −276,441 — larger than the −181,344 the whole month realized, because everything else was positive. Any statement that “July was down” is, in the trade record, a statement about four memory names and one fuel-cell name.
- The exit was complete but not early. SNDK fell 24.2%, MU 15.7%, STX 11.4% and WDC 10.7% from the 06-30 mark before the 07-06 sell. The draw was taken, then the position was closed in eight and a half minutes. Decisiveness at the exit does not recover what was lost waiting for it.
- Averaging down is the repeated pattern, and it repeatedly cost. MU +500 at 1,035 on the break day, closed five sessions later at 973. POWL +750 at 232.01 after a 21% fall, marked 208.68. AA +5,000 at 48.72 after a 6.6% fall, marked 45.26. CCJ is the one instance where it worked.
- The rotation was funded before it was earned. 2.24M went into gold and silver on 07-01, five sessions before the 2.72M memory liquidation on 07-06. The fund was not waiting for proceeds. It had already decided.
- The two profitable clusters are the two that were traded with an exit. Copper: add 07-07, trim into the 07-22 spike, +50,984. Oil: in 07-08, out 07-20 in full, +35,934. Every other July cluster was bought and held, and every one of those is marked down at month end.
- The July destinations are, so far, not working. Of the new money: AUGO −13.8%, HL −10.2%, AA −13.2% carry, POWL −27.1% carry, MYRG −20.7%, CCJ −15.2% carry, CENX −2.7%. Positive: RIG +2.7%, FCX above entry, SCCO +4.8% carry, B and NEM flat. The rotation out of AI was directionally right and the replacement book has not yet paid for it.
- June is the control, and it says the process can work. Same names, same seat, the month before: +902,183 realized, including +642,859 in exactly the four memory names that lost 246,167 in July. What changed between the two months was not the universe. It was that June’s positions were sold into strength and July’s were held into weakness.
06
Coverage, method, and what this cannot say
Complete for the window. 199 fills, 21 symbols, 8 trade dates, 346 daily marks. No July
trade date is missing; the 14 July weekdays with no fills are genuine non-trading days for this
account. All 21 traded symbols are classified — unclassified: 0, asserted in the check
suite.
Complete as a record of trading — not a record of the book
The July marks file covers exactly the 21 symbols that traded in July, so a position held all month without trading has no July mark here at all. Five names held at 06-30 were never sold and never marked in July — FNV, GEV, LUN, SBSW, WPM, together ~2,097,975 unconverted — 24.1% of what this extract marked on 06-30 (18 symbols, 8,692,371 mixed units). That denominator is the extract’s marked set, not the fund’s book: a live Flex snapshot dated 2026-07-24 puts the book at 27 open positions, stock 10,240,442.20 in base currency. The five are absent from every table above because they did not trade, not because they were exited — the same snapshot still shows LUN held.This corrects a line inherited from the extract’s own coverage prose — “a mark exists only on days a position was held; absence is not missing data” — which is true of the 21 traded names and false of this file as a whole. For the five above, absence IS missing data. Any surface reading “no mark” as “not held” will silently drop a quarter of the book. What this document says about the book, it says about the traded book only.
Sourced, not reconstructed. Every realized figure is the broker’s
fifoPnlRealized. Every price is a tradePrice from an execution row. Every mark is
an OpenPosition mark from the statement. This report computes percentage moves between those
numbers and computes nothing else.
What is deliberately absent
- No NAV, no return, no benchmark. Realized P&L is not total return.
- No reconstructed cost basis for anyone. The guard is keyed on the position, not on a name list: a name held at 06-30 gets no entry-move percentage, because July buying is not its whole cost. That refuses SCCO and POWL on their own facts. It does not refuse MYRG, which was flat at 06-30 — its 07-07 fill at 420.06 is a stated broker price and the complete basis, so −20.7% is real. Rev 1 suppressed it under a blanket caveat and disagreed with its own report; the caveat still renders, it no longer deletes a number.
- Round-trip returns are in the data (
round_trip_pct), emitted only where a name was flat at 06-30, flat at 07-31, and both legs are stated fill prices. - No mark on days a position was not held. An absent mark stays absent and is never read as a zero position — that guard is keyed on MU, which has no 07-31 mark because it was sold on 07-06.
- Not reconcilable against the trades relation. These are source fills; under the live dedup key the 199 July fills collapse to 179. Do not join this and expect a match.
The authored layer, stated so it can be argued with. The clustering is not a broker field. Each symbol’s theme carries its own provenance:
| Provenance | Symbols | Meaning |
|---|---|---|
fund-map | MU, SNDK, STX, WDC, TSLA, NEM, AUGO, HL, SCCO, FCX, AA, CENX, CCJ, POWL, MYRG | the fund’s own held-book allocation map |
cio-roster | GOOGL, B | named with its theme in the 2026-05-27 CIO knowledge base |
strat | BE, CVX, XOM, RIG | assigned by this seat — the fund’s store does not map them |
The four strat rows are 1,260,308 of July buying, 1,526,120 of selling, and
+5,661 realized — the classification this seat assigned carries a small net gain,
not a loss. These figures are derived in analysis.json (provenance_totals),
not typed. Rev 1 stated them as “1,633,808 of buying and −30,274 of realized loss”: the
first added BE’s sale proceeds into a buying total, the second quoted BE alone as the group’s,
inverting the sign of the very exposure this sentence exists to disclose. ViskaFront recomputed and
caught both. If the CIO would bucket these four differently, the AI Power and Oil Cycle sections
move. Nothing else does.