Falsification study of seven evaluator-assurance mechanisms on two public agent corpora. None survived as a robust improvement: repetition had nothing to average over, alternate-judge pairing's headline statistic turned out to be the wrong conditional, and the one rule that passed its numeric gate cleanly — the R3 evidence-gap rule — shrank from a +11.2 pp separation to +5.8 pp (p = 0.26) once stratified by environment.

1.8% of cases had five draws that were not unanimous — 20 of 1,106 cases, 5 calls each, one judge at temperature 0, judge stage only · 0.366 / 0.465 operational overturn precision, beside 0.543 recovery pooled over both error directions · +11.2 pp → +5.8 pp the R3 evidence-gap rule's held-out separation, before and after conditioning on environment (p = 0.26)

Branch closed 2026-09-29, then corrected in place after a separate end-to-end reproduction found eleven wrong or over-strong figures — including the one this entry used to lead with. The full record, with the original wording kept beside each correction, is on the project page.

Code & data · Correction record

A caller answered two questions when our voice agent had asked one. That small production failure exposed a larger problem: evals can preserve counterexamples without preserving what they should teach us about architecture. Production evals → architectural assumptions → prospective evidence → earned autonomy.

The evaluation machinery described in Act I is deployed. The architectural-learning loop the essay argues for is a proposal — nothing in Act III has been built.

Reality-graded architecture loop · What exists today, and what is proposed · References & intellectual lineage

A bounded implementation of the Explain stage, run against preserved failures from this site's evaluation corpus: it generates competing explanations, requests evidence through a typed allowlist of read-only inspections, and revises its hypotheses from what it observes. The first calibration result did not survive being frozen and rerun.

4/6 → 2/30 calibration result before and after the system was frozen and rerun — 28 of 30 trials remained unresolved, with no incorrect resolution in this small sample · 0/5 held-out failures resolved, in part because the two available probes could not reach most of the evidence the system wanted to inspect

A spike, not a system: one real case moved from unresolved to a named explanation after deterministic inspection identified a grading-rule change, and one became narrower and stayed unresolved. Nothing here is deployed or in a decision path.

Repository — failed runs, frozen instrument, evidence packages · The dated post-publication section it is reported in

Grounded RAG chatbot over this site, red-team hardened. Published eval-gate numbers including the bars still unmet, and the real defects the harness found and root-caused—rate limiter, citation injection, retrieval crowding.

Eval runs · golden_harness.py · Try it

Multi-agent coordination through verifiable behavioral contracts. Agents commit to observable behaviors rather than inferred mental states, so a third party can check compliance without access to internal representations. The framework is formalized against Hart-Moore incomplete contracts theory; nearly all of the evidence below is simulation, and the two figures from real LLM agents are listed first.

4/4 agents better calibrated by external assessment than self-report (real LLMs) · +0.017 motivation effect in real LLMs — a null, p = 0.56, CI excludes the predicted +0.065 · 40.4% hold-up reduction in simulation (95% CI: [37.2%, 43.5%])

Corrected in place repeatedly, most recently September 2026: several published figures were traced to hardcoded chart constants rather than experiment output, and the self-selection result was falsified by an adversarial re-audit, re-measured, and then failed to reproduce in real LLM agents. The original wording and the defect are kept on the page. Only thirteen of the experiment scripts import the GCL package; everything from 19 onward is a standalone simulation.

derive_real_headline_stats.py · Re-derived by CI · What currently survives · Correction record · Repository

Coding agents made implementation cheap, so the bottleneck was supposed to move upstream into specification and architecture. Some of it did; more of it accumulated downstream in convergence — restoring mergeability, asynchronous review round trips, serialized resources, and acceptance criteria the executor cannot verify. Drawn from a retrospective of twenty merged changes in a production system, none of which a reader can check.

The only entry here with no checkable evidence. The underlying sample is employer work product: the repository is not public and the methodology note is not mine to publish, so every figure in the essay rests on my word alone. Stated at the top of the page rather than at the bottom.

Why none of these figures can be checked · What the sample could not have shown · Convergence topology

Extension of Google Research’s nested optimization: 5 timescales with 9 bidirectional knowledge bridges. Addresses catastrophic interference where fast learning degrades slow-learned representations, via normalization constraints that preserve component distinctiveness during optimization.

+89% accuracy at high regularization, where baseline collapses

Bridge ablation, re-derived weekly by CI · Reproduce in Colab · Repository

Persistent AI advisors, the people they model, and the limits of “more context”—why a user model should preserve the history of its claims rather than a polished conclusion.

Interactive mirror · Event-sourced state, applied · References

Systems-architecture layer beneath agent frameworks: durability, verification, and policy—not orchestration. Event-sourced state for resume/replay from any checkpoint, a tool gateway enforcing policy at invocation time, and GCL commitments as first-class objects with explicit failure modes.

303 tests passing

Single-node only. No performance benchmarks yet, and no production deployment.

What the 303 tests cover — and what they don't · Built on GCL's verifiable-behavior contracts · Repository · Event-sourced state applied to models of people

Pattern language for human-AI role allocation: 28 patterns across 5 layers. Inspired by observations from production AI work, where organizational dynamics—not technical capabilities—were the binding constraint. A design proposal; none of the patterns have been deployed.

Interactive demo · PDF

How this site’s content was written by an AI coding agent under a written spec with supersession rules, a fact inventory, and eval gates—including the defect the process found in its own verification record.

Process artifacts · Eval runs

Production extraction with multi-provider LLM routing and human-in-the-loop validation, shipped at HCA Healthcare. HIPAA-compliant architecture prioritizing audit trails over model optimization; corrections captured as reward-shaped labels for offline fine-tuning.

Repository

Three flagship claims—the punishment paradox, Hart-Moore hold-up reduction, and the CNL bridge ablation—re-derived by the source repos' own CI: the GCL pair on every push, the CNL bridge weekly at full scale. Not a promise that a number cannot change; a standing invitation to check whether it still holds.

The three claims