A three-seat Profitability Path Audit (Claude · ChatGPT · Gemini) ranked ten routes to a genuinely profitable trading version. Nine were dead, bounded, or a posture. One had a structural reason to exist: on Kalshi prop series the maker fee is $0 — so a resting order on a genuinely wide, thin, active early-week book can earn the spread instead of paying the taker hurdle. If those books exist. If adverse selection doesn't eat them. This page is the full path from that hypothesis to a proof — or a clean, permanent kill.
Gate record closed — PASS stands. The 7-day census window (Nov 2–8, Phoenix days) is complete, and the program's gate record (data/gate-status.json, evaluated Nov 9, 2026, 6:16 AM MST by the frozen gate code) records 36 qualifying series-days against the frozen pass bar of ≥3 — by Phoenix day: Nov 2: 8 · Nov 3: 7 · Nov 4: 5 · Nov 5: 0 · Nov 6: 1 · Nov 7: 0 · Nov 8: 15, across 15 prop/ladder series. The gate record closed today, Nov 12, 2026: the UNTESTABLE backstop does not fire. The final verdict document (VERDICT.md, landed late Nov 9) is in the program record, with the G-11 independent recomputation at 36 = 36 — the document gap named in updates #12–#14 is closed. The paper maker screen is ACTIVE under its frozen quoting policy (early-week Tue/Wed/Thu sessions, 06:00–20:00 Phoenix; first session Tue Nov 10) — paper only, nothing live, and any live step remains a separate Gate 2 decision. Today's Morning Audit on the revival paths: no changes — every dead/parked path re-checked against its revival condition, none revived, no candidates; the grok-bot weekly check is Mondays only and is not due today (Thursday) — Monday's result is carried. Rig note: Nov 11 completed 95 census cycles (a 1-cycle shortfall vs the ~96 expected, above the flag threshold) and the watchdog restarted the supervisor 22 times since yesterday's check, checkpoint resume each time — the degraded 429 cycles stand in the record, not hidden. Sensitivities from the same gate record (printed per the pre-registration; they never govern): S1 interval-joint 19 series-days, S2 ≥50%-of-snapshots 1, S3 close-within-7d 36. Qualifying-spread histogram (3,607 qualifying markets): 3¢ 698 · 4¢ 398 · 5¢ 229 · 6¢ 184 · 7¢ 139 · 8¢ 113 · 9¢ 121 · ≥10¢ 1,725 — no concentration at exactly 3¢, so quoteability is not named the binding constraint. Named gap: the actual operator-hours tally against the Nov 1–12 phase cap is not in any record read today; the PASS above rests on the frozen primary rule.
Next milestone: Fills checkpoint — 150 credited fills under the frozen policy, or the Jan 4, 2027 stop. The gate record closed Nov 12, 2026.
Program outcomes and observations come only from the validated record embedded in this page. Display counts may be derived deterministically from those records; the clock never advances stages, infers verdicts, or creates observed progress.
Current: the paper maker screen (its condition — a gate pass — occurred Nov 9). Conditional stages run only if their condition occurs: the CLV read if the fill checkpoint is reached, the settled-P&L season if CLV passes.
Gate record closed — PASS stands. The gate record closed Nov 12, 2026 with the Nov 9 verdict standing: 36 qualifying series-days against the frozen ≥3 bar (Nov 2: 8 · Nov 3: 7 · Nov 4: 5 · Nov 5: 0 · Nov 6: 1 · Nov 7: 0 · Nov 8: 15; 15 series); the UNTESTABLE backstop does not fire. The final verdict document (VERDICT.md, landed late Nov 9, G-11 recomputation 36 = 36) is in the program record. The paper maker screen remains the current stage under its frozen policy (Tue/Wed/Thu sessions; fills checkpoint = 150 credited fills or the Jan 4, 2027 stop). Strategy Watch: no revival changes, no candidates; Atlas 16 records, CANDIDATE-FOR-AUDIT 0; Perception gaps MEASURED 2 + MEASURED-EXTERNAL 1, CANDIDATE-FOR-AUDIT 0. Sensitivities (never governing): S1 19, S2 1, S3 36; spread histogram shows no concentration at exactly 3¢ (698 of 3,607). Named gap: the actual operator-hours tally for the Nov 1–12 phase is not in any record read today.
Every stage carries its state, its dates, and the rule that closes it. Conditional stages name the exact condition that opens them — a gate pass alone does not guarantee the later stages: fill volume and the CLV verdict are separate conditions.
An illustrative book: 46¢ bid / 50¢ ask. The maker posts a resting bid at 47¢ — maker fee $0 on these series. A taker crossing this book pays ≈2¢ of half-spread plus ≈1.75¢ of taker fee per contract at ~50¢. (In a tight 1¢-spread book the total taker hurdle is ≈2.25¢ — that figure already includes the fee.) Posting a quote earns nothing by itself: it may never fill, and a fill is only the start — what the price does next decides the outcome. Both endings are shown with equal weight, because which of these happens on average is exactly what the program measures.
Rest a bid at 47¢ inside a 46/50 book. Fee $0. Nothing is earned yet — the quote may never fill.
A seller hits the quote; the mid rises to ~52¢. Markout ≈ +5¢ before settlement. The favorable ending.
A seller hits the quote because the price is falling; the mid drops to ~44¢. Markout ≈ −3¢. Adverse selection — the ending that kills makers on average.
Price paths are illustrative; the fee and spread figures are the program's registered constants. This diagram does not represent the program's credited fills or realized returns — none exist yet.
A daily audit (~06:45 Phoenix; first regular run Mon Nov 2 — this run — with an inaugural pre-window run on Nov 1) re-checks every dead and parked path on this page against its pre-registered revival conditions and scans for genuinely new candidates. A revival trigger never activates anything — it proposes a fresh audit and a new decision. ◐ INFORMATIONAL
How to read the watch chips: ✕ DEAD — revival condition unmet · 👁 WATCH — something moved, condition not yet met · ⚑ TRIGGERED — FOR REVIEW — condition met; a fresh audit and an owner decision come next, nothing is activated. Green is reserved for passed tests and never appears here.
Today (Nov 15, 2026): Morning audit: no changes on the revival paths. No path is WATCH or TRIGGERED; no candidates found. The aged-band path stays PARKED — GrokBot F's verdict lands later today (Sun Nov 15, after its 21:00 UTC run); at this check its verdict file still records only the interim verdict. Routine publish — only this Strategy Watch panel is refreshed; the program panels above stay on the Nov 12 gate-record-close milestone stamp. Last checked: Nov 15, 2026, 6:23 AM MST.
| Path | Status · last checked Nov 15, 2026 | Revival condition (one line) |
|---|---|---|
| Kalshi pregame taker | ✕ DEAD | Material fee-schedule change, OR durably wider game-day books over multiple days, OR a new settlement-compatible exactly-matched deviation class. |
| Cross-venue hedging (Kalshi vs books) | ✕ DEAD | A Kalshi settlement-rule change (tie / postponement / DNP / stat-correction) so a hedge's worst state is no longer unlocked. |
| LLM-in-the-loop trading | ✕ DEAD | A ≥500-event paired observational record with kept-vs-removed lower bound > 0 exists. |
| Memecoin launch-curve LP | ✕ DEAD | A real LP instrument with fee entitlement on graduating venues AND a specified inventory policy. |
| Memecoin aged-band timing (1–24h) | ◐ PARKED | Moves only on GrokBot F's Nov 15 verdict. |
| Kalshi-vs-Kalshi baskets | ◐ LINE ITEM — 0 CLEAN | ≥5 state-matrix-clean candidates in the census parity log promotes it to a standalone evaluation. |
| Grok-bot genre paths | ✕ DEAD | Weekly check (Mondays) only: xAI ships per-bot isolation, or the original claim's author publishes verifiable wallet history. |
Candidates today: none. Notes from today's checks: the fee schedule is still the Jul 7, 2026 schedule; no Kalshi rulebook amendment dated after Oct 26, 2026 touching tie / postponement / DNP / stat-correction for an in-scope series was found (a Nov 4, 2026 Rule 14.3 security-futures filing exists and is out of scope); game moneyline median spread is 1¢ on Nov 1–14 and 1¢ on the partial Nov 15 morning (revival needs ≥3¢ on ≥3 game days); the grok-bot weekly check is Mondays only and is not due today (Sunday) — Monday's result is carried: xAI documentation still states no per-bot isolation, connectors account-wide, and no author wallet-history publication was found. The aged-band path stays PARKED — GrokBot F's verdict lands later today, Sun Nov 15 (after its 21:00 UTC run); at this morning's check the verdict file still records only the interim verdict, so the path does not move. Rig note: Nov 14 completed 74 census cycles (a shortfall vs the ~96 expected, above the 72-cycle flag threshold); the watchdog restarted the supervisor 19 times since yesterday's check, checkpoint resume each time — degraded cycles stand in the record. Checker watchdog deaths (census chain): 2 today · 142 total. The census window closed Nov 8 and the gate verdict is carried in the program panels above, sourced from the program's gate record — it is not a revival status and nothing in this panel activates anything. The gate record closed Nov 12 with the PASS standing. The paper maker screen's frozen-policy sessions run Tue/Wed/Thu — today (Sun Nov 15) is not a session day, and no session is started by this audit.
Atlas: 16 records (A-01–A-16) · CANDIDATE-FOR-AUDIT 0 · KILLED 3 · DATA-BLOCKED 1 · 1 covered by Build F · 1 known-documented, routed out — Frontier: no change since yesterday's record; the registry remains at Sweep 4 (landed Nov 9) — A-05 evidence PRELIMINARY (Sample 1 PASS, 1 of 2; not a candidate), A-02 Sample 1 FAIL (1 of 2), A-01 covered by Build F with the gate-verdict note, A-04 KILLED (watch recheck near-misses only, clean candidates still 0). No record holds CANDIDATE-FOR-AUDIT; CANDIDATE-FOR-AUDIT remains 0. (Registry: ~/workspace/strategy-discovery/ATLAS.md — registry summary v1, 2026-11-01, updated through Sweep 4, 2026-11-09. Read-only to this program.)
Perception gaps: EDGE-REGISTER v1.1 — MEASURED 2 (PR-03 in-game lag, PR-05 memecoin hype) + MEASURED-EXTERNAL 1 (PR-07 favourite–longshot tails) · CANDIDATE-FOR-AUDIT 0 (none — no candidate name) — counts unchanged since yesterday's record; the register's weekly accrual #3 landed Nov 10 (no status change, no candidate assigned). (Register: ~/workspace/perception-reality/EDGE-REGISTER.md, read-only to this program.)
Data Room inventory as of Nov 1, 2026 — compiled Sun Nov 1, 2026 (America/Phoenix), from the Backtest Data Room's INVENTORY.md. Every figure below is taken verbatim from that inventory's “Totals at a glance” and verified against INVENTORY.md itself at this publish. Originals were never modified; all pulls public-API only, $0 spend.
| Block | Rows | Coverage |
|---|---|---|
| Kalshi market metadata + results (399 NFL series) | 157,207 markets | 2026 season to date (series-dependent; core series full season) |
| Kalshi candles, Nov 1 slate (hourly) | 75,793 | Oct 25 – Nov 1, 601/601 markets |
| Kalshi candles, season core series | GAME complete: 140,498 (264/264); SPREAD 620,201 rows / 3,413 of 3,564 (95.8%, stopped on one 429, resumable); TOTAL complete: 536,274 (2,662/2,662) | all settled core markets, full life each |
| Kalshi trades, Nov 1 slate | 452,127 prints (floor — see gaps) | Oct 25 – Nov 1, 600/601 markets |
| Wave-1 Kalshi snapshots (A–F, Nov 1) | ~100k quote rows + F census accruing | Nov 1, 2026 only (+ F forward) |
| Forward settlements banked | 49,591 | all settled markets, F's 25 series, full API history |
| Forward depth ladders | 2,400 books/cycle, accruing | from Nov 1, 23:13 UTC, every 900s |
| Solana sample universe | 402 unique mints | wave-1 samples, Nov 1 |
| Solana OHLC (hourly) | 395 rows / 56 tokens | Oct 31 – Nov 1 |
| April-system evidence | documents only — no database | Mar–Apr 2026 |
Top gaps (from INVENTORY.md, as of Nov 1, 2026):
Source: ~/workspace/backtest-data/INVENTORY.md, compiled Nov 1, 2026 — read-only to this program. Candles are marks, not fills, and are not books (no depth/touch size).
Road map as of Nov 1, 2026, 1:24 PM MST — verdicts from the Profitability Path Audit, reconciled by all three seats.
The audit's reconciled ranking (research priority, not expected profit). No path is credible as a profitable strategy today. This build is path #1 — and paths #4 and #6 ride inside its census as arms, at $0 extra.
| # | Path | Verdict | Why |
|---|---|---|---|
| 1 | Kalshi maker quoting — early-week props / ladders | UNPROVEN · TESTABLE | Maker fee $0 on prop series; the only structural reason to exist. Decided by the census gate below. |
| 2 | Memecoin screen vs coin flip (GrokBot F) | UNPROVEN · TESTABLE | Already running; verdict Nov 15. A relative PASS ≠ profitable — an absolute read-out is pre-registered beside it. |
| 3 | Productize the measurement tooling | BOUNDED TEST ONLY | Revenue that doesn't require beating a market. 30-day demand test, kill <5 paying customers at ≥$10/mo. Not launched — a separate decision. |
| 4 | Kalshi in-game stale-quote taking | UNPROVEN · TESTABLE | The least-refuted claim — nothing pregame touches it. Runs as the census's second arm, with an executability margin. |
| 5 | Memecoin timing (1–24h aged band only) | UNPROVEN · TESTABLE | Never tested; an F-successor only, never a parallel fork. No build now. |
| 6 | Kalshi-vs-Kalshi structural baskets | CENSUS LINE ITEM | Same rulebook, so a true payout identity is possible. Parity log + state-matrix test; ≥5 clean candidates promotes it. |
| 7 | Stay paper-only | THE POSTURE | Credible as a posture, dead as an earnings mechanism. Contains paths 1–2 at ≈$0 and ≈0 operator hours. |
| 8 | Kalshi pregame taker (lines + props) | DEAD | Every moneyline exactly 1¢ wide; Kalshi's mid matched the devigged book price to 0.1pt. The "edges" were stale references. |
| 9 | LLM-in-the-loop trading | DEAD | Deliberation's delta over a deterministic gate: exactly 0 (n=67). Observational logging only, never a gate. |
| 10 | Memecoin LP on launch curves | DEAD | No instrument exists pre-migration; post-migration means quoting into the most informed flow in any market we touch. |
The maker hypothesis dies or lives on a fact nobody has measured: early-week prop/ladder books. The census counts series-days (one series × one Phoenix day, Nov 2–8). A market qualifies only if it is wide (spread ≥3¢), thin (touch depth <50 contracts on both sides), and alive (it actually printed trades — the activity floor the devil's advocate added, so dormant books can't fake a pass).
of a series-day's two-sided prop/ladder markets must be wide + thin + trading
counted Nov 2–8, 2026. No discretionary calls — the formula is frozen.
The maker arm closes and the null is published, provenance-complete. The calendar is not extended to rescue it.
Qualifying series-days were computed once, at the verdict (Nov 9, 2026; record closes Nov 12) — no running partial count was published during the window, because a mid-week tally would invite exactly the discretionary reading this program forbids. The verdict count is 36 (see the gate tracker below). The rig snapshots every configured series every ~15 minutes: spread, depth at the touch, and trades printed.
Qualifying series-days — 36 / 3 · GATE PASS — computed at verdict, Nov 9, 2026 (gate record evaluated 6:16 AM MST; by day Nov 2: 8 · Nov 3: 7 · Nov 4: 5 · Nov 5: 0 · Nov 6: 1 · Nov 7: 0 · Nov 8: 15)
Recorded census days — 7 completed / 7 (window Nov 2–8 complete)
Parity candidates (state-matrix clean) — 0 recorded; census parity log running (18,991 scanned in the last cycle; 9,524,997 cumulative)
All three must hold in ≥10% of a series-day's two-sided markets for that series-day to count.
Cross-venue hedges are unlocked by settlement tails (ties pay $0.50, books refund). Same-exchange baskets settle under one rulebook, so a true identity is possible. Every candidate gets the state-matrix test: minimum payout across all settlement states, minus both legs' fees, must be > 0. ≥5 clean candidates → standalone evaluation. Zero → the line item closes with the census.
Polling at ≤2 seconds after scoring plays, play-id aligned to the ESPN reference. A candidate needs a ≥5¢ reference move, a <1¢ Kalshi move, and persistence of ≥ max(5s, 2× the rig's own measured latency) — staleness must be capturable, not just visible. <100 qualifying events → the arm closes untestable.
Written and locked before activation. The engine exists today and is deliberately inert: it refuses to quote unless the gate record says PASS.
Did ≥3 series-days qualify? If not: UNTESTABLE by Nov 12, null published, stages 4–6 close by gate, path #1 joins the dead — cleanly, permanently, for $0.
The paper engine quotes early-week books. Kill is evaluated on the first 150 credited fills — no warm-up exclusion, because the first fills are the most adversely selected. Fewer than 150 by Jan 4 → stop.
Mean closing-line value with confidence bounds. Upper 95% bound <1.5¢ → kill. A pass (lower bound >0, point ≥1.5¢) buys a settled-P&L season — not a live step.
Settled returns over ≥60 independent games. Lower bound <0 → kill. This is the honest horizon: settled-P&L proof of a 2–3pt edge needs ~2,175–4,900 bets against a 272-game season. Anyone promising faster is selling something.
Nothing on this page authorizes real orders. A credentialed gateway, capital, and position caps would be Curtis's separate, explicit decision.
Maker arm closes as UNTESTABLE by Nov 12. Published null. No extension. ○ NOT FIRED
Stop — untestable by volume. A program that can't reach its own n ends, it doesn't drift. ○ NOT FIRED
Kill. Evaluated on the first 150 fills, markouts reported per fill. ○ NOT FIRED
Kill — even if CLV was positive. Convergence is not cash. ○ NOT FIRED
Public Kalshi API, subscription audit seats, prompt-only Grok Build. OpenRouter banned and unused.
Paper orders only. No account, no sign-in, no trading keys exist anywhere in this build — the client code has no order endpoints at all.
Hard stop. Every verdict reports w_max — net proceeds before operator time ÷ operator hours. A surplus that can't pay for the attention it consumes fails the definition of profitable.
Built Nov 1, 2026 as Kalshi Build F — the build-out of the Profitability Path Audit's #1 ranked path, with its own full three-seat cross-browser audit before the census window opened. The dashboard upgrade on this page passed its own three-seat audit (round 1 + cross-examination + animation addendum) the same evening; every blocking fix the seats raised is in the page in front of you. Evidence base: the Nov 1 census (558 markets, every moneyline 1¢ wide), the Kalshi fee schedule effective Jul 7, 2026, and Bürgi–Deng–Whelan's maker/taker record — scoped honestly: the average maker loses ~10%, which is the base rate this program must beat, not a coefficient for it. Paper only. No guaranteed profit — there is no guaranteed profit in the evidence, only a proof chain with pre-registered kills.