Findings.
Seeds, locks, laws, evals, and misses we will defend. Process writeups sit under Blog.
Entries dated before 2026-10-11 keep the lab’s former name, Fudoshin Research, as a record. Fudoshin is now a program inside Vektra Research.
- The loop closes on wait, not on QA card is not a loop. At Pi = 0.60 the wait-loop sheds and the official hint cuts. Q is audit. Tail still open (miss 0.328 after the mean is repaired). Simulated plant, not live.
- All-hot tandem reading overstated wait 23%; corrected and simulatedFirst draft said over the 400 ms mean SLA. Traffic equations (rho_l = lambda tau_l) say under: W_net = 345 ms, confirmed by discrete-event simulation to 0.6%. Survived: 30% of jobs still miss the deadline. Stage D.
- Operator instrument for the Lineage EquationPublished budget is Q_eff = u Q with zone on u. One window in, a card out. Additive Q_eff is historical writing only.
- Stage E synthetic weight IDLinear identification of Q on 300 synthetic perturbations beat equal weights (R^2 0.942 vs 0.875). Not live operational data.
- nu*^4 identifiability missFull-quadratic Lineage form makes momentum weights practically unidentifiable; Stage E weights come from a linearized instrument.
- Contiguous proof stagesOfficial maturity is the highest contiguous stage on the A-I ladder; skipped rungs do not count.
- Lineage Equation v2Q^2 = Pi-tilde^2 + M-tilde^2 is the official capacity invariant: algebraic identities and exact gradients hold; live prediction is not claimed.
- Agent Mathematics as the S1 spineThe active S1 program is Agent Mathematics; the open mathematics packet is now a Fudoshin Research program under Vektra Industries.
- S1 + Decision laws as first spineDefault agenda is S1 thinking frameworks plus falsifiable decision-collapse laws before product shells.
- Name lock: Fudoshin ResearchPublic lab face is Fudoshin Research under Vektra Industries; scientific object is mind-under-load.