Mission Control
Oversight
The loop runs itself. Your job is judgment — approve what the machine proposes, decide what gets real capital, set the bar.
Loading…
Mission Control
The loop runs itself. Your job is judgment — approve what the machine proposes, decide what gets real capital, set the bar.
Loading…
Every night the machine walks this pipeline left to right. The funnel narrows on purpose — almost everything dies before the pen.
Promotion needs 13 forward paper weeks, of which at least 4 genuinely live (wall-clock, after nomination), plus the quality gates. A date below is the earliest the calendar allows — quality gates can still block after it passes.
3 Mondays to go — time gates met 2026-09-28 at the earliest. Still blocked by: deflated Sharpe 0.92 < 0.95; paper info-ratio n/a not positive.
The most recent backtests, newest first — every attempt is recorded before it is scored, wins and losses alike.
the NFCI aggregates credit, equity and rate stress into a leading gauge, so tightening financial conditions precede risk-asset drawdowns by weeks before they are priced into individual names; macro releases are traded against an expectation, and the expectation is reconstructable from FRED's own vintage archive: the first published value minus a naive forecast built from the prior vintage path is a usable surprise even with no consensus tape, and those surprises accumulate because data-generating processes are autocorrelated and revisions arrive in batches; bond desks reprice within hours while equity flows lag several days as revisions propagate through analyst models and rebalancing calendars, so ACCUMULATED surprise leads equity repricing rather than coinciding with it; alone among the registered signals it reads what each number said ON THE DAY IT WAS PUBLISHED rather than today's revised history, so the surprise lives in the first-print-to-revision gap that no other module can see. Composition: expressed cross-sectionally, the combined score sorts the universe into relative winners and losers. This interaction is plausibly under-arbitraged because these are distinct, slow-moving forces that standard cross-sectional factor models largely ignore.
the NFCI aggregates credit, equity and rate stress into a leading gauge, so tightening financial conditions precede risk-asset drawdowns by weeks before they are priced into individual names; macro releases are traded against an expectation, and the expectation is reconstructable from FRED's own vintage archive: the first published value minus a naive forecast built from the prior vintage path is a usable surprise even with no consensus tape, and those surprises accumulate because data-generating processes are autocorrelated and revisions arrive in batches; bond desks reprice within hours while equity flows lag several days as revisions propagate through analyst models and rebalancing calendars, so ACCUMULATED surprise leads equity repricing rather than coinciding with it; alone among the registered signals it reads what each number said ON THE DAY IT WAS PUBLISHED rather than today's revised history, so the surprise lives in the first-print-to-revision gap that no other module can see. Composition: used as a gate, the first force scales exposure to the rest — capturing a conditional, regime-dependent effect. This interaction is plausibly under-arbitraged because these are distinct, slow-moving forces that standard cross-sectional factor models largely ignore.
Aggregate market read: neutral (+0.01)
How near the best research is to the 90% bar that lets a pattern enter the pen. Bars fall as more trials accumulate — that's deflation, not a bug.
large speculators front-run institutional rebalancing by 1-3 weeks, and at the extremes that becomes crowding: deeply long equity-futures positioning alongside deeply short 10-year-note positioning is an overcrowded risk-on bet that mean-reverts, and it survives because COT publishes with a lag, aggregates heterogeneous trader types, and the reversal window is narrow and regime-dependent; convexity demand from hedgers and liability-driven investors shows up as the LONG end flattening while the belly stays steep, and when it does duration hedgers step back from Treasury purchases and funding conditions tighten in repo and OIS 5-10 days before the NFCI registers it; unlike signal-curve this reads the 20-day CHANGE in curvature rather than the absolute term-spread level, so it sees reversals that happen inside a bull-steepening or bear-flattening regime, and it survives because tracking it otherwise wants dealer-flow and repo data. Composition: expressed cross-sectionally, the combined score sorts the universe into relative winners and losers. This interaction is plausibly under-arbitraged because these are distinct, slow-moving forces that standard cross-sectional factor models largely ignore.
large speculators run crowding cycles that linear aggregation cannot see: CFTC net positioning stretched long in equity futures while simultaneously stretched short in 10-year notes is regime confusion — contradictory macro views held at once — and resolves within 2-3 weeks through forced rebalancing as carry unwinds collide with fading equity momentum; it survives because the professionals holding the positions cannot trade against their own hedges; convexity demand from hedgers and liability-driven investors shows up as the LONG end flattening while the belly stays steep, and when it does duration hedgers step back from Treasury purchases and funding conditions tighten in repo and OIS 5-10 days before the NFCI registers it; unlike signal-curve this reads the 20-day CHANGE in curvature rather than the absolute term-spread level, so it sees reversals that happen inside a bull-steepening or bear-flattening regime, and it survives because tracking it otherwise wants dealer-flow and repo data. Composition: used as a gate, the first force scales exposure to the rest — capturing a conditional, regime-dependent effect. This interaction is plausibly under-arbitraged because these are distinct, slow-moving forces that standard cross-sectional factor models largely ignore.
The machine stops here — these are the calls it isn't allowed to make.
weekly snapshot · 2026-09-07 (next: Monday 13:00)
The machine's standing schedule — this is the work that happens while you wait on the pen. Dots show each job's most recent actual run.
Changing these is a deliberate code change (edit criteria.ts + commit) — friction on purpose, so the bar is never lowered on a whim.
The loop runs on its own — these are for when you want to poke it now instead of waiting for the schedule.
Time gates met. Still blocked by: deflated Sharpe 0.90 < 0.95.
4 Mondays to go — time gates met 2026-10-05 at the earliest. Still blocked by: paper info-ratio n/a not positive.
4 Mondays to go — time gates met 2026-10-05 at the earliest. Still blocked by: paper info-ratio n/a not positive.
Time gates met. Still blocked by: deflated Sharpe 0.90 < 0.95; paper info-ratio -0.26 not positive.
1 Monday to go — time gates met 2026-09-14 at the earliest. Still blocked by: deflated Sharpe 0.80 < 0.95.
Time gates met. Still blocked by: deflated Sharpe 0.90 < 0.95.
2 Mondays to go — time gates met 2026-09-21 at the earliest. Still blocked by: deflated Sharpe 0.91 < 0.95; paper info-ratio -0.17 not positive.
3 Mondays to go — time gates met 2026-09-28 at the earliest. Still blocked by: deflated Sharpe 0.95 < 0.95.
3 Mondays to go — time gates met 2026-09-28 at the earliest. Still blocked by: deflated Sharpe 0.93 < 0.95; paper info-ratio n/a not positive.
3 Mondays to go — time gates met 2026-09-28 at the earliest. Still blocked by: deflated Sharpe 0.92 < 0.95; paper info-ratio n/a not positive.
3 Mondays to go — time gates met 2026-09-28 at the earliest. Still blocked by: paper info-ratio n/a not positive.
3 Mondays to go — time gates met 2026-09-28 at the earliest. Still blocked by: deflated Sharpe 0.92 < 0.95; paper info-ratio -0.18 not positive.
3 Mondays to go — time gates met 2026-09-28 at the earliest. Still blocked by: paper info-ratio -0.05 not positive.
3 Mondays to go — time gates met 2026-09-28 at the earliest. Still blocked by: deflated Sharpe 0.91 < 0.95; paper info-ratio -0.18 not positive.
3 Mondays to go — time gates met 2026-09-28 at the earliest. Still blocked by: paper info-ratio n/a not positive.
3 Mondays to go — time gates met 2026-09-28 at the earliest. Still blocked by: paper info-ratio -0.21 not positive.
3 Mondays to go — time gates met 2026-09-28 at the earliest. Still blocked by: deflated Sharpe 0.93 < 0.95; paper info-ratio -0.17 not positive.
3 Mondays to go — time gates met 2026-09-28 at the earliest. Still blocked by: paper info-ratio -0.24 not positive.
3 Mondays to go — time gates met 2026-09-28 at the earliest. Still blocked by: deflated Sharpe 0.94 < 0.95.
4 Mondays to go — time gates met 2026-10-05 at the earliest. Still blocked by: paper info-ratio n/a not positive.
the NFCI aggregates credit, equity and rate stress into a leading gauge, so tightening financial conditions precede risk-asset drawdowns by weeks before they are priced into individual names; macro releases are traded against an expectation, and the expectation is reconstructable from FRED's own vintage archive: the first published value minus a naive forecast built from the prior vintage path is a usable surprise even with no consensus tape, and those surprises accumulate because data-generating processes are autocorrelated and revisions arrive in batches; bond desks reprice within hours while equity flows lag several days as revisions propagate through analyst models and rebalancing calendars, so ACCUMULATED surprise leads equity repricing rather than coinciding with it; alone among the registered signals it reads what each number said ON THE DAY IT WAS PUBLISHED rather than today's revised history, so the surprise lives in the first-print-to-revision gap that no other module can see. Composition: combined as a weighted tilt, the forces should reinforce when aligned and cancel when opposed. This interaction is plausibly under-arbitraged because these are distinct, slow-moving forces that standard cross-sectional factor models largely ignore.
dominant media narratives spread epidemically and shift aggregate risk appetite semi-independently of fundamentals (Shiller); macro releases are traded against an expectation, and the expectation is reconstructable from FRED's own vintage archive: the first published value minus a naive forecast built from the prior vintage path is a usable surprise even with no consensus tape, and those surprises accumulate because data-generating processes are autocorrelated and revisions arrive in batches; bond desks reprice within hours while equity flows lag several days as revisions propagate through analyst models and rebalancing calendars, so ACCUMULATED surprise leads equity repricing rather than coinciding with it; alone among the registered signals it reads what each number said ON THE DAY IT WAS PUBLISHED rather than today's revised history, so the surprise lives in the first-print-to-revision gap that no other module can see. Composition: expressed cross-sectionally, the combined score sorts the universe into relative winners and losers. This interaction is plausibly under-arbitraged because these are distinct, slow-moving forces that standard cross-sectional factor models largely ignore.
dominant media narratives spread epidemically and shift aggregate risk appetite semi-independently of fundamentals (Shiller); macro releases are traded against an expectation, and the expectation is reconstructable from FRED's own vintage archive: the first published value minus a naive forecast built from the prior vintage path is a usable surprise even with no consensus tape, and those surprises accumulate because data-generating processes are autocorrelated and revisions arrive in batches; bond desks reprice within hours while equity flows lag several days as revisions propagate through analyst models and rebalancing calendars, so ACCUMULATED surprise leads equity repricing rather than coinciding with it; alone among the registered signals it reads what each number said ON THE DAY IT WAS PUBLISHED rather than today's revised history, so the surprise lives in the first-print-to-revision gap that no other module can see. Composition: used as a gate, the first force scales exposure to the rest — capturing a conditional, regime-dependent effect. This interaction is plausibly under-arbitraged because these are distinct, slow-moving forces that standard cross-sectional factor models largely ignore.
dominant media narratives spread epidemically and shift aggregate risk appetite semi-independently of fundamentals (Shiller); macro releases are traded against an expectation, and the expectation is reconstructable from FRED's own vintage archive: the first published value minus a naive forecast built from the prior vintage path is a usable surprise even with no consensus tape, and those surprises accumulate because data-generating processes are autocorrelated and revisions arrive in batches; bond desks reprice within hours while equity flows lag several days as revisions propagate through analyst models and rebalancing calendars, so ACCUMULATED surprise leads equity repricing rather than coinciding with it; alone among the registered signals it reads what each number said ON THE DAY IT WAS PUBLISHED rather than today's revised history, so the surprise lives in the first-print-to-revision gap that no other module can see. Composition: combined as a weighted tilt, the forces should reinforce when aligned and cancel when opposed. This interaction is plausibly under-arbitraged because these are distinct, slow-moving forces that standard cross-sectional factor models largely ignore.
elevated policy uncertainty raises required risk premia and delays real investment, so risk assets should lag defensives until it resolves; macro releases are traded against an expectation, and the expectation is reconstructable from FRED's own vintage archive: the first published value minus a naive forecast built from the prior vintage path is a usable surprise even with no consensus tape, and those surprises accumulate because data-generating processes are autocorrelated and revisions arrive in batches; bond desks reprice within hours while equity flows lag several days as revisions propagate through analyst models and rebalancing calendars, so ACCUMULATED surprise leads equity repricing rather than coinciding with it; alone among the registered signals it reads what each number said ON THE DAY IT WAS PUBLISHED rather than today's revised history, so the surprise lives in the first-print-to-revision gap that no other module can see. Composition: expressed cross-sectionally, the combined score sorts the universe into relative winners and losers. This interaction is plausibly under-arbitraged because these are distinct, slow-moving forces that standard cross-sectional factor models largely ignore.
elevated policy uncertainty raises required risk premia and delays real investment, so risk assets should lag defensives until it resolves; macro releases are traded against an expectation, and the expectation is reconstructable from FRED's own vintage archive: the first published value minus a naive forecast built from the prior vintage path is a usable surprise even with no consensus tape, and those surprises accumulate because data-generating processes are autocorrelated and revisions arrive in batches; bond desks reprice within hours while equity flows lag several days as revisions propagate through analyst models and rebalancing calendars, so ACCUMULATED surprise leads equity repricing rather than coinciding with it; alone among the registered signals it reads what each number said ON THE DAY IT WAS PUBLISHED rather than today's revised history, so the surprise lives in the first-print-to-revision gap that no other module can see. Composition: used as a gate, the first force scales exposure to the rest — capturing a conditional, regime-dependent effect. This interaction is plausibly under-arbitraged because these are distinct, slow-moving forces that standard cross-sectional factor models largely ignore.
Positive = risk-on (tilt toward equities), negative = risk-off (tilt toward defensives). These readings are what the search composes into candidate patterns.
the NFCI aggregates credit, equity and rate stress into a leading gauge, so tightening financial conditions precede risk-asset drawdowns by weeks before they are priced into individual names; convexity demand from hedgers and liability-driven investors shows up as the LONG end flattening while the belly stays steep, and when it does duration hedgers step back from Treasury purchases and funding conditions tighten in repo and OIS 5-10 days before the NFCI registers it; unlike signal-curve this reads the 20-day CHANGE in curvature rather than the absolute term-spread level, so it sees reversals that happen inside a bull-steepening or bear-flattening regime, and it survives because tracking it otherwise wants dealer-flow and repo data. Composition: combined as a weighted tilt, the forces should reinforce when aligned and cancel when opposed. This interaction is plausibly under-arbitraged because these are distinct, slow-moving forces that standard cross-sectional factor models largely ignore.
equity valuations are convex in real rates and the convexity is asymmetric — mandate floors and forced index rebalancing bias the response toward downward rate shocks — but when the Fed balance sheet and financial conditions reach an unusual joint state investors reprice tail risk and the convexity inverts; linear rate models cannot express that regime switch; large speculators front-run institutional rebalancing by 1-3 weeks, and at the extremes that becomes crowding: deeply long equity-futures positioning alongside deeply short 10-year-note positioning is an overcrowded risk-on bet that mean-reverts, and it survives because COT publishes with a lag, aggregates heterogeneous trader types, and the reversal window is narrow and regime-dependent. Composition: used as a gate, the first force scales exposure to the rest — capturing a conditional, regime-dependent effect. This interaction is plausibly under-arbitraged because these are distinct, slow-moving forces that standard cross-sectional factor models largely ignore.
convexity demand from hedgers and liability-driven investors shows up as the LONG end flattening while the belly stays steep, and when it does duration hedgers step back from Treasury purchases and funding conditions tighten in repo and OIS 5-10 days before the NFCI registers it; unlike signal-curve this reads the 20-day CHANGE in curvature rather than the absolute term-spread level, so it sees reversals that happen inside a bull-steepening or bear-flattening regime, and it survives because tracking it otherwise wants dealer-flow and repo data; sentiment extremes are contrarian only when the crowds holding them are CONFUSED: a persistent disagreement between retail survey sentiment and active managers' actual equity exposure means one group is liquidating into the other's accumulation, which resolves as a 2-3 week counter-move; it is the breadth of opinion rather than its extremity that marks exhaustion, so a shared extreme says nothing, and it is under-exploited because fusing two lagged weekly surveys is unglamorous work. Composition: used as a gate, the first force scales exposure to the rest — capturing a conditional, regime-dependent effect. This interaction is plausibly under-arbitraged because these are distinct, slow-moving forces that standard cross-sectional factor models largely ignore.