Shape the ritual signal.

The workshop checks four case-insensitive signal families: ritual, agent, onchain or record, and experiment or prove. Each family contributes at most one count, regardless of repetitions. A text containing ritual, agent, record and prove produces four out of four. At one hundred pixels per signal these bars compare four hits with a four-signal ceiling. These are substring checks, not semantic evidence. Substring-containing words such as agentic or proven can also match. Nothing is posted or written to a chain. A question mark or the literal sequences what-space or how-space choose one fixed question template. Otherwise the other template is used. Bars compare two possible templates with one displayed draft at two hundred pixels per count. With no signal hits the count is zero, yet the UI still displays all four suggested signal labels. Zero supported hits and four displayed suggestions must not be confused. The tool rewrites with fixed text; it does not run a language model or establish that the draft is supported. Export returns one JSON review with source text, the most recent fixed draft, selected signal names and an empty external actions array. Bars compare one file with zero external actions at two hundred pixels per count. Editing without forging leaves the prior draft in memory. Reset clears visible output and storage but retains that memory, so export can pair the reset source with the earlier draft. Reload uses saved source and forges it again. These original state boundaries should be reviewed before treating an exported draft as fresh evidence.