Method — p-rick

test 01

Real problem

A failure mode that recurs, hurts, and is documented in incidents, studies, or expensive behavior — not a hypothetical nobody has suffered.

test 02

Large audience

Millions of people or an entire profession, not a hobby community. The missing thing must matter at mass scale to count as a gap.

test 03

Empty category

Verified against the landscape before writing, with every adjacent incumbent named and its boundary stated. One search can kill the claim — that is the point of writing it falsifiably.

test 04

Specifiable result

Formalizable into a model, protocol, or architecture that a builder could implement and an evaluator could test. A feeling is not a deliverable.

The exclusions

hard rules, not preferences

No toy projects: nothing whose primary value is demonstration, novelty-of-implementation, or portfolio value. No deterministic wrappers: no test suites, verification gadgets, “verified wrapper” utilities, or deterministic-crypto proof-of-correctness artifacts whose only user is their own construction.

No AI-obsolete software: nothing whose reason-to-exist disappears the moment a model can do it; the program prefers gaps the AI era makes larger. No CRUD, no chatbot shells, no dashboards-as-research.

No me-too products: if the category exists and is merely under-served, that is a product review, not a gap. The program publishes specifications for missing systems — not apps, not demos, not libraries wrapping an afternoon’s insight.

What survives is the narrow slice where a real, mass-scale problem has no category at all: the empty quadrant. The 30-paper ledger below is the program’s claim that such quadrants exist and can be found reproducibly.

The pipeline

survey · claim · grade · specify · red-team

Survey before writing: multi-agent landscape scans that name the incumbents, the near-misses, and the dead research lines; directions that die in survey die quietly, with their notes kept. Falsifiable gap claims: “nothing exists that does X” is written so that one search can refute it — and claims are scoped to survive the search (“component PARTIAL, composition STRONG” is an honest verdict; the program uses it).

Graded evidence: STRONG / PARTIAL / WEAK verdicts appear in the papers themselves, including where the strongest objections might kill the thesis. Specification over prototype theater: formal models, protocols, threat models, and evaluation designs — the artifacts a builder actually needs, before any code exists. Red-team pass: adversarial review challenges novelty claims, headline figures, formal properties, and cross-paper consistency; findings are fixed, not footnoted. And an essay with every paper, because research nobody reads is a diary.

Citation honesty

Papers are drafted with domain knowledge and marked citations, then pass through a live-source verification ledger before final release. References recorded from prior knowledge are flagged verification queued until the pass completes; the ledger travels with the sources, and corrections are applied to the papers rather than argued around. Where the program is confident, it says so; where it is provisional, it says that too. The first program-wide pass (2026-10) adjudicated 21 load-bearing claims against live sources: 16 verified as stated, 5 corrected in place — including a misattributed D-Lib citation (Granger, 2000, not Rosenthal) and a nonexistent USENIX title — with the corrected records and per-reference verified marks now in the papers.

The current ledger

gaps by grade · 30 papers

Paper	The missing thing	Gap grade
P-001	Personal event bus — user-owned ambient event middleware	STRONG
P-002	Entitlement engine — consumer rights as machine-readable objects	STRONG
P-003	n=1 cost-of-living index — the methodology layer	PARTIAL→STRONG
P-004	Degradation contracts — behavior under pressure, declared	STRONG
P-005	Provenance-native storage — the file system as causal ledger	STRONG
P-006	Attention scheduler — interruption economics as an OS problem	STRONG
P-007	Intermittent compute fabric — one machine across many devices	STRONG
P-008	Device succession — vendor-death resilience architecture	STRONG
P-009	Circadian orchestration — biological time in team scheduling	PARTIAL components / STRONG composition
P-010	Bus factor protocol — registry succession governance	STRONG
P-011	Model extinction — behavioral conservation instruments	STRONG
P-012	Complexity ledger — design as relocation, accounted	STRONG
P-013	Delegated operation — multi-principal software	STRONG
P-014	Modality failover — cross-sensory redundancy for alerts	STRONG
P-015	Uncertain document — uncertainty-native numbers	STRONG
P-016	Defaults ledger — governance for default settings	STRONG
P-017	Revocation protocol — consent withdrawal as propagation	STRONG
P-018	Epistemic half-life — truth decay semantics for stored data	STRONG
P-019	Dormancy engineering — wake-time assurance for sleeping software	STRONG
P-020	Blast radius engineering — personal reliability for correlated failure	STRONG
P-021	The reproduction number of code — epidemic thresholds for dependency compromise	LAW · validated (28 configs)
P-022	The admission law — closed-form cache admission under Zipf + transients	LAW · validated (18 configs)
P-023	Generation loss — drift law for re-embedded vector corpora	LAW · validated (5 experiments)
P-024	Resolution collapse — the satisfiability threshold of dependency resolution	LAW · validated (48 grids)
P-025	The collapse law — ceilings, hysteresis, and recovery for retry storms	LAW · validated (6 experiments)
P-026	The cascade law — the arithmetic of the credential cascade	LAW · validated (6 experiments)
P-027	The dissipation budget law — the conservation architecture for depth	LAW · validated (5 experiments + E6 end-to-end training, 96 runs) · rev 1.2: normalization comparison exact (§4.6) · rev 1.4: E6b trunk-LN ablation landed (registered collapse branch taken: 0.409 ± 0.137 at depth 240, backward max 2.99×10⁸) + E7a pre-registered CIFAR-10 round landed (60/60 runs: mixed verdict — parity, zero PH failures, no signature at 48, 4.4× compute; reported per the no-spin rule)
P-028	The anchor law — ratchets, floors, and the coverage of agentic maintenance	LAW · validated (4 experiments)
P-029	The lock-in law — the phase transition in agent collectives	LAW · validated (5 experiments)
P-030	The Green’s function of context — positional recall as inverse design	LAW · validated (6 experiments)

Thirty papers, zero incumbents inside their gaps, eight series. Series VI–VIII are a different genre from I–V: not gap specifications but laws — closed forms with validation harnesses and reproducible figures, the program’s quantitative turn. The specification series: user-owned data (I), systems (II), continuity — what systems lose over time (III), assumptions — the quiet premises software runs on (IV), and promises — the guarantees software implies and never has to honor (V). The laws series: derivation and validation (VI), collapse — critical thresholds where everyday infrastructure fails abruptly (VII), and machinery — mechanisms invented, not just found (VIII).

Work tracking

The program logs itself: agents.md is the automatic session ledger — what was done, when, and by which agent. Research first, products later, in separate directories, only after the research freezes.