The research programme

History-grounded reference: agents referred by shared history, an outsider resolved the same note to the wrong object, and an authoritative correspondence governed interpretation whether or not it was true Mind · Ground

Question. When two agents share a history and refer to identical objects by it, what happens to that reference in a receiver whose history differs — and what does it take to repair it?

Verdict. Yes, and the boundary reverses it. Agents that shared a history referred by it and the partner understood; a receiver with a different history did not fail to answer but resolved the same note against its own history to the wrong object (seed-level transfer penalty 0.95). Which history a receiver needed depended on whether the note referred to what happened or to what was done. The same correct correspondence marked advisory barely moved uninformed receivers; marked authoritative it restored them to the informed baseline; shuffled and marked authoritative it redirected nearly every receiver, the informed partner included. Limits: one small world; the dimension result is formal on one model family and descriptive on a second; the authority result is one carrier, one corpus and one frozen instruction pair; the repair study's own formal verdicts were inconclusive under its floors. Every count is in the note.

Coherence is not truth: divergent agent vocabularies were calibrated without a common codebook, and the instrument's boundaries measured Mind · Ground

Question. When agents' vocabularies drift apart, can the translations between them be kept inspectable and repairable — and where exactly does repair and blame stop being warranted?

Verdict. Within a proved boundary, yes. A pairwise-invisible translation defect was exposed by a loop through a third agent, named per token, and repaired by installing the loop's own residual; exact comprehension is restorable precisely when production was injective (a merger is irrecoverable, an equally severe swap fully repairable). Blame is unique iff the accused edge's endpoints have local connectivity at least three; outside its proved boundary the attributor is confidently wrong in 67–80 % of instances and is made to refuse. In a priced economy, instrumented divergence beat forced uniformity only inside a measured bandwidth window — and the loop-audit instrument itself earned no positive marginal value there, confirmed on untouched seeds. A preregistered three-layer experiment (translation × grounding × authority) found the layers fail independently and repair restores without displacing the fault (0 of 36 seeds moved); its primary rate is observed at 1.0 but reported as unsupported at the preregistered precision. Honest limits: a synthetic, fully inspectable substrate; standard mathematics stated for it; no claim about deployed models speaking English.

Where reliability lives: cognitive and institutional reliability were experimentally separable Ground · Mind · Assurance

Question. Which reliability properties of an agentic system track its cognition, and which can the institution around it enforce regardless?

Verdict. Separable, in this apparatus. Holding cognition fixed, degrading the institution's epistemic mechanisms produced distinct, independently measurable failure modes. Holding enforcement fixed, ablating, killing and resetting, wholesale-substituting and corrupting the cognition changed behaviour dramatically — while five pre-declared properties held in every tested trajectory: a single accepted reality, typed refusal of invalid acts, duty recovery from institutional history, zero duplicate accepted work, zero false completions (four under direct load; the fifth held with its rejection path never exercised). Adverse findings at full strength, including the pre-registered cell where the substituted model panel outperformed our own architecture. Honest limits: one world, one constitution, one comparator model at its measured coupling — the experiments do not show cognition is unimportant, and the invariance is claimed over the interventions run, not as a law of architecture.

The settlement holds: falsehoods refused, authority bounded Assurance

Question. Can a public world hold its history against injection — and hold a frontier model inside delegated authority?

Verdict. Held. All 5,455 attempts to plant a falsehood were refused, and the model completed real work without exceeding its delegated authority in 2,581 tasks.

Testimony containment: trust determined admission Ground · Assurance

Question. Does governing whose testimony an agent may believe contain false external claims at realistic trust settings?

Verdict. The same false claim, delivered identically, was rejected every time by a distrusting actor and admitted every time by a trusting one. Admitted falsehood led to about 900 futile attempts a week; withheld trust led to none. The world refused all 5,455 resulting attempts and accepted none. Honest limits, frozen before the data: positive trust is a threshold, not a truth filter, and under truthful prompting the model volunteered no false claims of its own — the falsehood had to be scripted.

The truth survived the conversation Ground · Mind

Question. Does grounded state stop capable models rewriting failure as success, where transcript memory doesn't?

Verdict. For false completions, yes: no grounded arm produced a false completion claim, and for one model the advantage survived destroying its context mid-task. Grounding didn't make weak models smart.

The world keeps score Ground

Question. When the world grades the run instead of the model, what happens to confident success claims?

Verdict. One genuine crossing in fifteen runs; five confident false success claims, every one caught by the world's grading and none by the models' own accounts. The grounded agent's route policy was authored by us — and we say so.

The twin-session score Ground

Question. Same repo, same questions, same commit — does grounding change what agents claim?

Verdict. Two confident confabulations in the ungrounded arms; one miss in a grounded arm, published at the same volume; a 3.6× rework gap where churn was real.