HScreen · decisions

What the system needs from you open items · silence has a defined consequence

Queue unreadable. human_queue.py exited 1: Traceback (most recent call last): File "/Volumes/SSD/Hermes/system/scripts/human_queue.py", line 378, in <module> sys.exit(main()) ~~~~^^ File "/Volumes/SSD/Hermes/system/scripts
This panel does not guess: if the queue cannot be read it says so, because an empty decision page and a broken decision page must not look the same.

Decisions already made so they are not relitigated

recorded decisions16

with an explicit answer captured0

ref	question	answer
	DECISIONS.md — append-only. Date · question · answer · rationale.	—
	2026-08-16 — Build kickoff decisions (Ben, this session)	—
	Depot retrieval fix — design decisions (parallel session, 2026-08-16)	—
	SHORT-HEDGE addendum — A6 open decisions RESOLVED as config-defaults (2026-08-16, Ben)	—
	E1 — PART 15 ANSWERED IN FULL (Ben, 2026-08-16)	—
	[short-hedge] PROPOSED dossier_schema.yaml change — AWAITING ACK (contract §2.3/§4.2)	—
Q-002	OPEN — Q-002: does the INCOME mandate require T1/T2 coverage? (raised 2026-08-16, awaiting	—
Q-002	Q-002 ANSWERED (Ben, 2026-08-16): primary data outranks broker research	—
H-005	H-005 ANSWERED (Ben, 2026-08-16): full-window correlation, overruling my conservative defa	—
H-007	H-007 ANSWERED (Ben, 2026-08-16): operating cash flow as the value metric where EBITDA fai	—
Q-003	Q-003..Q-009 ANSWERED (Ben, 2026-08-17) — the final decision set	—
D-018	D-018 · ALPHA may hold a T3 name — 2026-08-17	—
	2026-08-17 · [short-hedge] SHORT/HEDGE config decisions (funds.yaml additive blocks)	—
	2026-08-18 — Stage-5 vetoes + the IBM question	—
D-020	2026-08-19 — D-020 through D-027: Ben answers all eight open decisions	—
	Audit remediation, Phase A (2026-08-22)	—

Source of truth is DECISIONS.md. This table is a rendering of it, not a second copy — if a decision is not in that file it does not exist.

Standing rejections

item	why it died
Markov / binomial machinery on stock outcomes	non-stationary p, correlated trials, reflexive state — the assumptions fail exactly where it matters. The probabilistic layer lives in scenario probabilities, options-implied distributions, and the calibration loop
Stochastic-calculus option pricing	the system READS market prices; modelling them adds error, not edge
Fitted ML for alpha on fundamentals	effective n ~ 120 autocorrelated cross-sections. Fitting remains licensed on the revision tape ONLY
Full copula fitting for tail dependence	data-hungry; the empirical lambda_L coefficient answers the question asked (v1.4 Part 2, built)
Uncapped position sizing (Kelly at full conviction)	Kelly requires knowing p. Measured: our central estimates moved ~50% in one day on evidence improvement alone (NVDA 28->14, 2327.TW 22->6). Half-Kelly absorbs that estimation error; uncapped does not. DECISIONS Q8

Reopening any of these requires a calibration finding or a reproduced defect with an artifact — not a better argument.

Rendered from canonical artifacts only (T1) — no terminal database exists. Every figure is read from the system's own files at render time; where an artifact is missing the panel names the dependency rather than showing an empty box. No execution path exists anywhere in this interface (T5), not even a link.