Decision journal — Mode A, graded against four counterfactuals
Mode A (human decides) + Mode B (machine decides), one ledger · brain human:murat · PRODUCT_EXPERIMENT
loading… — a receipt, not a result
Process metrics
loading…
Where the counterfactual prices come from
loading…
Execution-repo state (read-only mirror)
loading…
Graded decisions — four counterfactuals each
a lesson is withheld until its own row resolved
| Name | Mode | Decided | Horizon | Actual | Held to horizon | Held to next review | Engine's own pick | SPY | Taxonomy |
|---|---|---|---|---|---|---|---|---|---|
No graded rows yet. The log is deliberately empty: seeding it with invented decisions would put fabricated human labels into the only labelled decision dataset this programme has. Rows arrive from POST /api/journal/thesis and from the conviction journal when it carries a falsifier and a catalyst. | |||||||||
regret = counterfactual − actual; positive means the alternative was better. A leg reading n/a had no declared price source — it is not a zero.
Pending decisions (written at decision time, resolved at their own horizon)
| Name | Source | Decided | Resolves | Next review | Budget | Falsifier |
|---|---|---|---|---|---|---|
| No decisions logged. | ||||||
Does the grader work? — the planted cases, run live
loading…
Loss budgets — declared before the first position (invariant 19)
| ref | book | judged at | expected losers | note |
|---|
An idea is retired by its book's scoreboard, never by its own first loss.