A dated, deliberately delayed extract of the private research notebook. Directions with patent potential publish the problem and the result, not the construction.
Not research on a model — the classifier is off-the-shelf and says so — but on the regime it runs in: zero retention, a score instead of a verdict, fail-closed under outage, and a refuted hypothesis about reactive moderation as a cheaper substitute for scanning. Includes what has not been measured.
Six directions on a time service whose statement is meant to hold for someone who trusts neither us nor our server: a zone-free unit, a clock that ticks without asking the server and publishes its own error, a weekly Merkle log, anchoring in a bank transfer reference beside Bitcoin, a draw nobody can steer, and proof of co-stamping a moment without identity.
Seven directions on records of research time whose credibility must not rest on the good will of their author: immutability with storno instead of editing, two dates, anchors that corroborate rather than measure, proof surviving the loss of the system, evidence without disclosing content, and records that cut their own hours.
A concept-phase program on machine forecasts whose first measured object is their own fallibility: calibration without model self-assessment, resolvability under not-at-random missing ground truth, mechanisms over events and fail-closed self-verification — with the entire construction withheld.
Six directions around cognitive-aging screening on consumer EEG: in-browser event marking, ERP and inter-subject correlation against age norms, remote hyperscanning of dyads, objective pharmaco-EEG endpoints, the evidential value of a record against the right to erasure, and comparability of data across device updates — with the status of each direction stated openly.
A compression research line running since 2021, negatives published first: a measured 56% expansion on high-entropy data, an abandoned degenerate variant, and the entropy boundary as a scoping result.
Compliance on ciphertexts and hashes alone: multi-source time proofs, a tamper-evident audit chain and replacement scanning under end-to-end encryption — including a published negative result on server-side antivirus.
Five directions around a digital vault that needs no guard: 2-of-3 trust splitting, multi-factor recovery, and the boundary between attesting the past and enforcing the future — with documented redesigns.
Four directions where steganographic security reduces to sampler fidelity rather than an arms race with detectors — including a published negative result on classical image steganography.