Larry Williams: the printed systems, run forward

What we granted, and what we actually tested

Granted, not measured: Williams printed executable rules and his own results. The subject is two printed systems, not the trader or his current record. The championship record and the later family result are granted as attributed accounts, not independently certified by us; no primary audit document is on file.

Tested: Two of eight reproduction-grade book families, at one contract, on Yahoo Finance 24-hour daily futures composites after publication. Both primary systems use two trade-handling treatments of the same prices. The range-stop sensitivity and the second coin use Arm-A only. Costs and a Yahoo ^IRX cash benchmark are included.

§1 · The locked pre-registration

Committed 028efc0 on 2026-06-13, before the probe was run ; execution was committed as 20ba677 later that evening. The post-hoc control is dated 2026-09-18.

The falsifiable question

Do the two printed rules clear a stated after-cost floor, cash and randomized-entry controls on post-publication data, while the two data treatments agree?

• A yes looks like: Both treatments agree, the rule clears cash and friction, exceeds its control threshold, has plausible cadence and does not depend on one winning year. This still would not substitute for the printed-cell reproduction gate.
• A no looks like: A rule below the floor yields a screen-tier finding only. Data-treatment disagreement would make the screen inconclusive.
• Ambiguous residue: Positive returns can fail a control. A stop-reading sign flip must be shown. Unrun printed-cell reproduction caps every conclusion.

Our stated prior: A faithful screen is worth distinguishing from our earlier bot

The protocol was a falsifiable screen, not a numerical probability forecast. The decision criteria and ambiguity rules were written before execution. The later same-fill control is explicitly outside that commitment.

The lock, the actual run and later additions

Untested scope and when it was identified

1. Other six book families, discretionary filters and money management; the bot previously built in his name.
2. His own historical windows and pit session: the paid reproduction gate was deferred in the lock.
3. Independent source data was intended but not obtained; Arm-B is the documented same-vendor substitute.
4. Direction re-flipped at the fill: identified after the second coin, remains unrun.

The goalposts — all required to proceed

The table reports each executed arm separately. Clause D is descriptive, below. The code uses a fixed $1,500 risk proxy even for the range-stop C ratio; it is not that arm's actual median stop. One of seven fidelity checks (calendar long-only) is structural; six use synthetic price fixtures.

§2 · The headline arithmetic

§3 · Was it the path? The stress tests

Stress rows are not independent experiments or pass/fail gates. A result with only a few trades, including the calendar COVID window, retains its denominator.

§4 · The controls

POST-HOC 2026-09-18: the same-fill coin was added months after the June screen. It never becomes a pre-registered arm by being included here. The original at-open control and result file remain unchanged.

The second coin samples touched day/side pairs at the breakout price. Conditional on a single-side touch, the fill already fixes breakout direction. It tests sequencing and price, not an independent direction choice. It permits overlapping positions; the real rule does not. Its primary form includes both-touched days; applying the real tie-break materially lowers the apparent percentile. The 60-bar cap drops no same-fill pair on this dataset. Neither a median comparison nor this post-hoc percentile is a new registered validation gate.

§5 · The strongest objection and what remains unrun

The strongest objection is correct: a pit-session rule was run on a 24-hour Yahoo composite, the stop sentence has two readings, and the printed-cell reproduction was not run. The alternative stop is shown because it changes the sign. The second coin changes the story about the driver. The unrun Pinnacle experiment remains the clearest way to lift the ceiling.

Other implementation limits: daily bars cannot order every intraday touch; stop fills use their level despite gaps. Real trades are sequential while null trades may overlap. Null walks stop after 60 bars. Terminal unresolved positions do not become completed ledger trades. Cash compounds the Yahoo ^IRX discount-rate Close divided by 100 and 252, forward/back-filled to futures dates; it is an approximation, not a total-return T-bill series. Arm-B is not independent-price replication.

§6 · Their own terms

The printed figures are his attributed cells, not ours and not a matched decay comparison. The first edition identifies the rule, session, windows and zero-cost assumption. Two systems out of eight were tested. No page scan is redistributed.

The complete 12-row interpretation ledger below comes from the pre-run specification. It records choices, not successful reproduction of the printout. Read it with the actual Arm-B deviation and run-stage choices above.

| OV | Item | Value / choice | Rationale | Re-validation |
|----|------|----------------|-----------|---------------|
| OV-001 | Fig 4.1 stop interpretation | money stop **$1,500** PRIMARY; **50%-prev-range** = sensitivity arm | book p61: "$1,500 or 50 percent of the previous day's range" — money-vs-range ambiguity; Williams' systems typically code a money stop | flips if repro gate favors the range form |
| OV-002 | Continuous-series method | back-adjusted (Arm-A) / non-adjusted-roll-accounted (Arm-B) | yfinance roll undocumented (DATA_PROBE) | Pinnacle non-adjusted at repro gate |
| OV-003 | Day-session approximation | daily-bar OHLC (yfinance Globex composite; no pit split) | free data has no intraday pit session | Pinnacle day-session series at repro gate |
| OV-004 | T-bond friction | $4 round-turn + 1 tick ($31.25) slippage | modern discount-broker norm; book was $0 | broker fills at deploy (T4) |
| OV-005 | S&P (ES) friction | $4 round-turn + 1 tick ($12.50) slippage | modern E-mini norm; book was $0 | broker fills at deploy (T4) |
| OV-006 | Contract point values | ZB $1,000/pt ($31.25 per 1/32); ES $50/pt ($12.50 per 0.25) | exchange specs | n/a (fixed) |
| OV-007 | Fig 10.1 repro window | printout **08/14/87–08/31/98** (NOT prose "1982") | printout is ground truth over prose | n/a |
| OV-008 | Screen OOS span | 2000-09 (yfinance inception) → present | 1999→2000-09 gap has no free data | Pinnacle/IBKR fills the gap at deep run |
| OV-009 | Position sizing | fixed 1 contract | matches book printouts (`Max contracts held 1`) | n/a |
| OV-010 | Bailout exit | first profitable opening after entry; multi-day holds | book p61/p148 | n/a |
| OV-011 | Fig 10.1 stop timing | $1,500, NOT on entry day, active all days after | book p148 verbatim | n/a |
| OV-012 | Repro-gate friction | **$0 / $0** (match book frictionless printouts) | reproduce his cells under his assumptions | n/a — friction added only in screen/OOS |

net **$73,468.75** · gross **$203,156.25** · **651**
  trades · **80%** profitable · winners 533 · largest win $3,968.75 · **avg trade $112.86** · max
  closed-out DD **−$10,031.25** · account size required **$13,031.25**. [p64 printout]

net **$73,437.50** · gross **$103,250.00** · **129**
  trades · **85%** profitable · winners 110 / losers 19 · largest win $6,750.00 · max closed-out DD
  **−$3,325.00** · profit factor **3.46** · account size **$6,950.00**. [p148 printout]

§7 · The complete ledger — all 10301 rows