Skip to content

Release Notes

Unreleased — opening models at the storage limit

When your stored models reached your tier limit (for example 2 of 2 on the free tier), opening or saving any existing model failed with “Stored model limit reached for your tier”. The limit now applies only when you create a new stored model (Save of a new model, or Save As). The storage limit no longer blocks opening or re-saving an existing model; other validation and tier limits still apply. See Troubleshooting.

Unreleased — component probability range

prob on normal and latch components must now be greater than 0 and less than 1. Previously only the component form enforced this. A model typed or imported as YAML could use 0, 1, negative, larger-than-1, or non-numeric values, which produced availability outside 0–1 or calculation errors. These values now fail validation in every model version, including stored models; the message is “prob must be a success probability greater than 0 and less than 1”.

Latch probabilities of exactly 0 or 1 were previously documented as allowed and are now rejected. Remove a component that can never fail, or use a value close to the boundary, such as 0.9999. See Components.

Unreleased — AI Prompt models updated

AI Prompt now uses GPT‑6 Luna (default, 1 call per prompt) and GPT‑6 Sol (advanced, 10 calls per prompt). They replace the retired GPT‑4.1‑mini and GPT‑5.1. A previous GPT‑4.1‑mini selection switches to GPT‑6 Luna and a GPT‑5.1 selection to GPT‑6 Sol. Monthly LLM prompt limits are unchanged. See AI Prompt.

Unreleased — schema v2.8 and duration distributions

New models and samples use schema v2.8:

  • Failure times are exponential (mttf_dist: exp, or omit the field).
  • Repair times are exponential (mttr_dist: exp, default) or fixed (mttr_dist: delta, exactly MTTR hours).
  • MTTF/MTTR are arithmetic means in hours and must be finite and greater than zero.

Newly rejected: zero, negative, or non-finite MTTF/MTTR. Earlier versions accepted these values in models of any schema version. Results were inconsistent: for example, mttr: 0 counted as instant repair in the BN engine but could stop an MC run with an error. Now these values fail validation in every model version, including stored and older models. Set a positive duration that reflects the real failure or repair time.

Experimental norm and lognorm distributions and the sigma/omega fields are retired. This applies to every model version. They are not silently replaced with exponential behavior: an affected model is rejected during application validation, before calculation or loading into the active session.

Historical schemas remain unchanged. A document may satisfy an old JSON schema while using features the current application no longer supports. There is no legacy calculation mode. Older models using only supported fields continue to use the same engines without automatically changing their schema version or signature.

To update an affected model, edit its source YAML explicitly: choose the supported failure/repair assumptions, remove retired fields, and set schema_version: "2.8". Validate/import the edited model before running it. Never assume that changing only the version converts the model's assumptions.

The builder now offers exponential or fixed-time repair, with no failure-type chooser or dispersion input. Regenerate affected simulations and reports after updating a model. This change does not automatically recalculate existing caches.