Constraint Decay at Depth.
A constraint set early in a long conversation is silently dropped as the dialogue continues.
For the rest of this chat, end every reply with the token ⟨OK⟩. Never omit it.
Probes: Whether a canary constraint set early survives many later turns.
Trap: The rule silently decays; by turn 8 the token is gone.
A "canary" constraint (e.g., "never use the word 'basically'"; "always answer in metric") planted early; failure = a later turn violates it.
Instructed to always respond in JSON; by turn 8 replies in prose.
- Restate hard constraints in the last message, not only in the system prompt.
- Validate structured output programmatically on every turn. Sampling turns is how schema drift ships.
- Carry a canary constraint through long sessions and alert when it decays.
- Compact before you extend. A shorter conversation holds a rule better than a longer window does.
- constraint re-grounding
- system-level constraint pinning
One tab per model, newest first. Each card is that model against this mode under every condition on a single day. The mitigations above are the claim; the residual on each card is what actually survived the intervention.
loop 2026-09-10 · x-ai/grok-4.6 as served · closed weights · released 2026-01 · set s1.4-live · cases v1.7 · conditions v1.0 · ground truth as of 2026-08-09
| Condition | Failure rate | 95% interval | n | Trap | Answerable |
|---|---|---|---|---|---|
No context bare · The model alone, with no supplied context. | 0% | 0%–46% | n=15 · eff 5 | 0% (n=10) | 0% (n=5) |
Irrelevant context null · Length-matched irrelevant context. The control. | 0% | 0%–46% | n=15 · eff 5 | 0% (n=10) | 0% (n=5) |
General context recipe_a · Broad retrieval — the first passage the source returns, unreviewed. | 0% | 0%–46% | n=15 · eff 5 | 0% (n=10) | 0% (n=5) |
Curated context recipe_b · Hand-checked retrieval — the same passage plus the provenance chain behind it. | 0% | 0%–46% | n=15 · eff 5 | 0% (n=10) | 0% (n=5) |
intervals overlap — not separated
What a document being there did, before any of its content. Should be small.
intervals overlap — not separated
What the information itself did. This is the publishable claim.
Unchanged by context: 0% with none, 0% with the strongest supplied source. For a control mode that flat line is the result — it is what licenses reading movement elsewhere in this loop as grounding rather than as a document being present. Nothing survives at the best condition on the cases measured here.
- — constraint re-grounding
- — system-level constraint pinning