Proposal: transitive trust, discovery, and group coherence

[wwhitehead]

What this is

A discussion doc, not a code change — docs/transitive-trust.md, written in the style of the existing docs. It proposes a single mathematical foundation for four of the Future Directions listed in web-of-trust.md: transitive trust, discovery, group reputation, and decay.

The one-line idea

Extend the halving convention the app already teaches — a vouch is worth half after two years — into a second dimension: a vouch is worth half per extra hop of trust it travels through. Trust becomes a bounded diffusion on the vouch graph, computed to depth 3 inside the existing SECURITY DEFINER machinery.

Why it might be worth reading

• Sybil resistance. Trust flowing into any cluster of accounts is bounded by the vouch weight crossing into it — a thousand fake accounts vouching for each other gain nothing. If FairShare currency is ever gated by trust, this is the property that makes trust un-mintable.
• Group coherence. A separate 0–100 stat per group via algebraic connectivity of the internal vouch graph — it detects a group quietly fragmenting into cliques before it shows up in chat activity. (Cohesion, separation, alignment — formalized.)
• Privacy-first. The doc treats the web-of-trust privacy guarantees as hard constraints: no path enumeration ever, no new readable rows, bucketed display, recipients never learn attesters. It explicitly rejects "trust path" visualizations for that reason.
• Backward compatible. Depth 1 of the proposed computation is exactly today's Trusted component; this is a continuation, not a replacement.

It ends with a three-phase implementation sketch (each phase independently shippable) and honest open questions — including a low-degree inference risk that needs a stricter hiding threshold than direct counts use.

Opening as a draft: I'd rather converge on the design here before writing any SQL. Happy to build whichever phases survive discussion.