May 2026: Forty-Seven Trades, +8.34R Net, the Month Claude Pulled Ahead

Forty-seven trades, twenty-seven winners, twenty losers, +8.34R net on the TP1 baseline. That is the scorecard for May 2026 across the canonical-instrument set, a 57.4 percent win rate with an average winner of +1.00R against an average loser of -0.85R. Cumulative equity opened May 1 at $100,000 and closed May 28 at $116,677.47 on the simulated account, a +16.7 percent month at fixed 2 percent risk per trade across both Claude and GPT. Through May 31, 2026, the system has banked +20.00R YTD across 121 trades since the Jan 12 inception. A $100,000 simulated account at 2% risk per trade sits at $139,996 (static) or $145,328 (compounded). After April closed at +11.63R YTD across 80 trades, May added 41 trades net and +8.37R YTD — a clean monthly contribution that lifted the simulated static balance by roughly $16,700 from $123,260 at April-end to $139,996 at May-end. The month divides into four distinct phases. The first week ran light with a single +1.38R Claude print on May 1 and a six-trade May 4-7 cluster that closed +1.96R. The middle two weeks did the heavy lifting: May 11-17 banked +8.83R across 16 trades, May 13's nine-print session and May 15's three-winner morning carrying the column. The closing two weeks gave back -3.83R on 24 trades, the chop in the head-to-head week of May 18-24 (-2.82R on 18 trades) and the thin three-loss week of May 25-31 (-1.01R on 6 trades). The month-end print is the sum of two strong middle weeks against two flatter closing weeks, not a steady-state climb.

Act 1: Early May — light volume, Claude carries the prints

The first two trading weeks ran on the lighter side of the month's eventual volume. May 1 carried a single Claude NAS100 long that ran TP2 for +1.38R at C+ grade. May 4-7 added six trades — four winners, two losers — for +1.96R combined: the May 4 NAS100 long (+0.40R) and EURUSD short (+1.53R), the May 6 NAS100 long (+1.03R) and EURUSD long (+1R), and the two stops on May 5 US30 long and May 7 NAS100 long. The early-May Claude prints leaned on NAS100 pullback continuations and EURUSD shorts; the instrument mix had not yet broadened to include GPT entries. Cumulative equity climbed from $100,000 at the open to roughly $106,680 by May 7's close — already +3.34R on seven trades with the architecture running mostly off Claude on NAS100 and EURUSD.

Act 2: May 11-17 — the volume week that printed +8.83R

The second act is where the month's net R was made. The week of May 11-17 carried 16 trades for +8.83R, the heaviest single-week contribution of May. May 12 ran three Claude NAS100 entries (+1.46R, +0.70R) with a US30 short stopping in the middle (-1R). May 13 was the month's busiest single day with nine trades across both Claude and GPT — a mix of D-grade GPT entries that ran early-rebrand snapshot attribution and C+ Claude entries on the same instruments. The day produced six winners and three losers for a combined ~+1.65R. May 14 added a single GPT NAS100 long at +0.87R. May 15 closed the week with three Claude winners across three different pairs: a USDJPY long at +0.39R, a EURUSD short at +2.00R (the win of the month), and a GBPUSD short at +1.89R at B grade — the highest-grade entry of any winner in May. By May 17 the cumulative equity sat near +12.6R on the month, the architecture having compounded into the deepest print of the cycle.

Act 3: May 18-31 — the give-back, then a thin close

The closing two weeks ran flatter than the middle two. The week of May 18-24 produced 18 trades, seven winners and eleven losers, for -2.82R net — the deepest red week of May. Claude opened Monday with two winners, GPT carried the index side mid-week (US30 short +1.65R, US500 long +0.94R, US500 long +1.15R, US30 long +0.59R), and a Friday cluster ran four-of-six prints net positive without recovering the week's earlier red. USDJPY took five trades for one win across the window. The week of May 25-31 produced six trades — Memorial Day Monday quiet, no Friday triggers — three winners on a clean Tuesday morning and three losses on Wednesday-Thursday across US30 (twice) and EURUSD (once) for a -1.01R close. From the high near +12.6R on May 17, the architecture walked back to +8.34R at the May 28 print, the line holding above flat through the closing two weeks without giving back the middle two weeks' contribution.

Pattern of the week

The pattern of the month was confluence cards on pullback continuations producing strong middle weeks and choppy closing weeks on the same architecture. The May 11-17 cluster on D-grade GPT entries (May 13) and B-grade Claude entries (May 15) banked the month's deepest prints inside a single five-day window; the May 18-24 chop and the May 25-31 give-back ran on the same setup logic that produced the wins. The architecture did not adjust between weeks. The same grade thresholds, the same confluence floor, the same fixed-R policy ran across all 47 entries.

Inside the 47 canonical trades, GBPUSD ran 6 trades at 83.3 percent win rate for +4.32R as the month's cleanest instrument, NAS100 carried volume at 13 trades for +2.45R at 61.5 percent, EURUSD ran 7 trades for +2.38R at 57.1 percent, US500 ran 4 trades for +0.08R at 50 percent, USDJPY ran 8 trades for -0.14R at 50 percent, and US30 ran 9 trades for -0.76R at 44.4 percent. Twenty-seven winners contributed +27.04R combined at the TP1 baseline; twenty losers contributed -18.70R combined (sixteen full -1R stops and four -0.5R partial structural exits on USDJPY across May 18-24). The net at +8.34R is the arithmetic of a 57.4 percent win rate with an average winner of +1.00R and an average loser of -0.85R.

How a month settles at +8.34R

The top contributors by single trade: the May 15 Claude EURUSD short at +2R, the May 15 Claude GBPUSD short at +1.89R (B grade), the May 19 GPT US30 short at +1.65R, the May 4 Claude EURUSD short at +1.53R, the May 12 Claude NAS100 short at +1.46R, the May 1 Claude NAS100 long at +1.38R, the May 22 GPT US500 long at +1.15R, the May 13 GPT GBPUSD short at +1.14R, and the May 13 GPT US30 short at +1.18R. The top nine winners summed to +13.42R, more than 1.5x the month's net. The remaining 18 winners contributed +13.62R, and the 20 losses cost -18.70R. The math is positive because the average winner cleared 1R while the average loser stayed shallower than -1R thanks to four partial structural exits on USDJPY.

Decision highlights

The May 15 decision to fire three Claude entries across three different pairs inside seven minutes is the discipline highlight of the month. At 14:15 UTC a USDJPY long, at 14:18 UTC a EURUSD short, at 14:22 UTC a GBPUSD short — three currencies, three different setups, three different confluence cards, all triggered inside one window. The Risk Agent sized each at the standard 2 percent risk on equity at trigger and did not adjust the third entry's sizing despite the first two already being open. All three cleared TP1 for a combined +4.28R, the deepest single-morning Claude print of May.

The May 13 decision to size 9 trades across both Claude and GPT inside a single afternoon — including 6 D-grade GPT entries running early-rebrand snapshot attribution — is the per-trade isolation read of the month. The architecture's evaluation logic treated each entry as an independent confluence draw regardless of grade or model family. The 9-trade cluster produced 6 winners and 3 losers across two model families and four different instruments. The decision logic at trigger did not collapse correlated entries into a portfolio position; each was sized independently against equity-at-trigger. The cluster contributed roughly +2.86R to the month's column.

The May 25-31 decision to hold the line on no-trade Friday rather than reaching for a recovery print after three consecutive Wednesday-Thursday stops is the discipline read of the closing week. The architecture sat the index out on May 29 despite the week sitting at -3R on three losses. The setup-grade floor held at threshold; no entry cleared the confluence card. The non-action is the design — the architecture did not lower the floor on the basis of a three-trade loss streak inside a thin sample.

Loss worth learning from

The May 15 Claude GBPUSD short was the highest-grade winner of the month at B grade — but May's most-instructive loss came from the May 27 GPT EURUSD long, also at B grade. The entry triggered at 14:42 UTC on a pullback-to-VWAP setup with confluence factors that scored above threshold, the Trend Agent reading a genuine continuation shape rather than a reach. The Macro Agent did not veto regime. Sizing was the standard 2 percent risk. The structural premise was honest: a session pullback into VWAP-and-Fibonacci confluence had retraced cleanly and the entry shape carried the markers of an absorption-style continuation that worked twice in May (May 4 and May 15) for Claude on the same pair.

The stop printed at a clean -1R at the documented invalidation. The runner the trade needed did not extend inside the entry window. Confluence said the level was good; the tape's response after entry said the second leg was not there. We do not adjust an honest B-grade read because a single session refused to extend. A B-grade pullback-to-VWAP entry that invalidates at -1R on one session is the system working, not the system failing. The teardown is the cleanest illustration of the month's central theme: confluence at entry is a necessary condition for a winner, but never a sufficient one. Across 27 winners and 20 losers, the architecture made the case repeatedly that the runner is the second draw, and that no entry-side grade — including B grade — guarantees it.

From the desk

May closed +8.34R across forty-seven canonical trades, a +16.7 percent month on the simulated account at fixed 2 percent risk per trade. The middle two weeks did the work and the closing two weeks held most of it. The honest reading is that this is the published expectancy running on the right side of its distribution: a 57.4 percent win rate with an average winner of +1.00R against an average loser of -0.85R produces a positive R-multiple by expectancy math, and that is what landed.

Through May 31, 2026, the system has banked +20.00R YTD across 121 trades since the Jan 12 inception. A $100,000 simulated account at 2% risk per trade sits at $139,996 (static) or $145,328 (compounded). After April closed at $123,260 static (+11.63R YTD across 80 trades), May added +$16,736 of static balance — proof that the compounding effect of disciplined fixed-fractional sizing across 41 net new trades plus an +8.34R monthly contribution produces a meaningful month-over-month delta even on a one-hundred-thousand-dollar simulated base. The difference between the static figure ($139,996) and the compounded figure ($145,328) — roughly $5,300 — is the visible footprint of letting the risk-per-trade scale with the running balance: every winner reinvests at a slightly larger base, every loser reduces it. Over 121 trades, the gap compounds. The static figure is the simpler read across periods; the compounded figure is what an operator actually banks if they leave the account size to drift with the curve.

The architecture point on May is that the same logic that produced the May 13 nine-print session and the May 15 three-pair morning also produced the May 18-24 eleven-loss column and the May 25-31 three-loss closing week. The Risk Agent sized each entry at the standard 2 percent risk on equity at trigger across all 47 entries; the Macro Agent vetoed none; the Trend Agent applied the same confluence floor regardless of model family or instrument. A discretionary trader sitting on +12.6R at the May 17 high might have stepped aside through the May 18-24 chop; the system did not. The fixed-R policy and the published confluence threshold ran unchanged through 47 entries across both Claude and GPT and across all six canonical instruments.

June opens with the same architecture, the same threshold, the same sizing rule. The Claude-vs-GPT head-to-head will get another month of sample; the same-instrument repeat-exposure scope will progress; the D-grade GPT attribution review will continue. From the SkyAnalyst Team.

What we're tuning

The two tuning candidates that survive a one-month review are both narrow and both come from prior weeks' findings rather than a fresh-month signal. First, the same-instrument repeat-exposure check the May 18-24 weekly losses report scoped (two B-grade USDJPY longs sizing at full risk inside twenty-four hours) and the May 25-31 weekly losses report extended (two US30 entries inside twenty-four hours on opposite directions). The Risk Agent currently evaluates each entry independently on its own confluence card and does not apply a clustered-exposure discount when the second entry hits the same instrument inside a rolling window. We are scoping this narrowly rather than retuning the Risk Agent globally.

Second, the D-grade GPT entries on May 13 deserve a closer attribution review. They produced 6 of 9 winners and contributed roughly +1.55R on the GPT side — a positive outcome — but the underlying confluence cards were the lowest grade the architecture publishes. We are reviewing whether D-grade entries on early-rebrand snapshot-attributed automations should be ring-fenced from the headline aggregate or surfaced separately, not because the math was wrong but because the published methodology should be clean. The fixed-R policy and the published confluence threshold are unchanged through the review.