Skip to content
Capital Compass
Viewing as guestSign Up Free
§11 + §13 · COST & ROI

Cost & ROI

Every Opus, Sonnet, and Haiku call this AI fund makes is logged here with its dollar cost. Every decision is scored at 7, 30, and 90 days against SPY. The ledger is the product.

30-day spend

$14.18

40 runs · soft ceiling $125

Ceiling utilization

11%

Inside policy

Mean alpha vs SPY (30d)

-3.20%

Hit-rate 11%

Cost per run avg

$0.354

1.4M tokens 30d

Per-cadence spend (30d)

Each run carries its own soft/hard budget from cost.js. Soft = log-warn, hard = abort.

Sunday Brief5 runs
$9.34
$1.868/run · cap $3.50/$5.00
Daily regime7 runs
$1.73
Monthly Brief1 run
$1.37
$1.370/run · cap $5.00/$10.00
Bulletin1 run
$0.80
$0.802/run · cap $1.50/$2.50
Midweek Note5 runs
$0.62
$0.125/run · cap $1.00/$2.00
Watch21 runs
$0.32
$0.015/run · cap $0.50/$1.00

Model mix (30d)

Where the dollars go. Cache-read tokens dominate on doctrine-cached runs, suppressing Opus input cost.

  • Unknown
    $12.13 · 86%
  • Opus
    $1.73 · 12%
  • Haiku
    $0.32 · 2%

Cost timeline (90d)

Daily spend (bars) + cumulative (line). The horizontal reference line is the 30-day soft ceiling.

Outcome scorecard

Every agent decision is scored at 7, 30, and 90 days against SPY total return. Hit-rate is % of decisions with alpha > 0.

7-day

-0.97%

39 decisions · 33% hit-rate

30-day

-3.20%

9 decisions · 11% hit-rate

90-day

—

No decisions scored

By decision kind7d alpha · hit30d alpha · hitn
  • Haiku scan-0.89% · 32%-2.24% · 17%28
  • Regime call-0.17% · 40%-7.30% · 0%6
  • Midweek check-0.67% · 50%-0.79% · 0%5
  • portfolio_pulse-2.38% · 20%—5
  • Sunday deep-0.87% · 33%-7.30% · 0%4
DecisionHorizonReturnSPYAlpha
  • Jun 5 · 2026-06-05-h7d+1.96%+0.41%+1.55%
  • Jun 4 · 2026-06-04-h7d-7.16%-2.41%-4.75%
  • Jun 3 · 2026-06-03-m7d-12.78%-3.61%-9.16%
  • Jun 3 · 2026-06-03-h7d-12.78%-3.61%-9.16%
  • Jun 2 · 2026-06-02-s7d-13.24%-3.44%-9.79%
  • Jun 2 · 2026-06-02-h7d-13.24%-3.44%-9.79%
  • Jun 1 · 2026-06-01-h7d-8.12%-2.50%-5.62%
  • Jun 1 · regime-2026-7d-8.12%-2.50%-5.62%
  • Jun 1 · 2026-06-01-s7d-8.12%-2.50%-5.62%
  • May 29 · 2026-05-29-s7d-7.14%-2.43%-4.71%
  • May 29 · 2026-05-29-h7d-7.14%-2.43%-4.71%
  • May 28 · 2026-05-28-h7d+0.82%+0.41%+0.41%
  • May 27 · 2026-05-27-m7d+5.21%+0.63%+4.59%
  • May 27 · regime-2026-7d+5.21%+0.63%+4.59%
  • May 27 · 2026-05-27-s7d+5.21%+0.63%+4.59%
  • May 27 · 2026-05-27-h7d+5.21%+0.63%+4.59%
  • May 26 · 2026-05-26-h7d+5.69%+1.15%+4.54%
  • May 25 · regime-2026-7d+3.07%+1.11%+1.96%
  • May 22 · 2026-05-22-s7d+5.03%+1.56%+3.47%
  • May 22 · 2026-05-22-h7d+5.03%+1.56%+3.47%

Budgets & ceilings

Per-run caps from cost.js (soft / hard / abort) and the rolling 30-day + 7-day ceilings from STRATEGY §11.2.

WindowActualCeilingStatus
30-day soft$14.18$125.0011%
30-day hard$14.18$250.006%
7-day hard$4.35$50.009%
CadenceSoft / runHard / run30d runs30d total
Sunday Brief$3.50$5.005$9.34
Midweek Note$1.00$2.005$0.62
Monthly Brief$5.00$10.001$1.37
Watch$0.50$1.0021$0.32
Bulletin$1.50$2.501$0.80

Cost transparency is the discipline forcing function. Soft and hard ceilings are codified in STRATEGY §11.2; per-cadence caps are hardcoded in backend/services/strategist/cost.js. Changes require commit + STRATEGY edit.

Outcome attribution methodology is documented in STRATEGY §13: SPY benchmark over the same window, regime calls marked to NAV, single-name picks marked to ticker, hit-rate = % of decisions with alpha > 0 at 30 days.