Two decisions, one tree: what you may say — gated by the identity you can prove — and what makes them act, the play that tier permits. The router proposes; the Gate disposes — now on persuasion too.

Exhibit 1 — Pass 1 · the permission router: identity key → personalization tier wired

Unchanged spine, one rule made explicit: when the company won’t resolve, default down the identity ladder to the individual — never sideways to a guessed sector.

Exhibit 2 — Industry needs a confidence gate — assertion is earned, not defaulted wired

Why this matters: location is ~city-accurate on fixed-line broadband, so we treat it as generally correct (and still allude, never recite). Industry is a second-order inference (IP→company→PDL→sector) — each hop loses confidence, so it must clear a bar before it changes the page. A confident-looking sector on a guess is the failure mode we just removed.

Exhibit 3 — Pass 2 · permission × play: what drives action at each tier proposed integration

Each tier is a budget of provable facts. The play spends exactly that budget — no more (that’s creepy/unprovable), no less (that’s a wasted send). One trigger, a loss-framed implication, matched-peer proof, one soft ask. loss angle + AI-tell Gate now wired in creative.py · full play orchestration proposed

Mapping back to your one-liner: “can’t figure out the company → default to individual” is the spine of Pass 2 too — Tier 2’s play degrades to Tier 1/0’s the instant the firmographic budget empties. The play always has a provable fallback move, so the page never blocks; it spends a smaller, honest budget.

Exhibit 4 — The ask, by temperature: drive action with one CTA, friction-matched to trust proposed integration

Rule: exactly ONE ask per message · ≤15 words · escalate commitment only as the lead warms · never a calendar link in a cold first touch.

Sequence, not single-touch: default 4–7 total touches, each carrying a new angle (never “just bumping this”), and reuse the same anchor signal across email · voicemail · call so the channels reinforce one identity. The CTA is the only place the message asks for a finite resource — so it’s the last thing to escalate. Wired: the no-cold-calendar-link veto (verify_message) and the never-single-touch / no-repeat rules (verify_sequence) are enforced; the staged escalation itself is the lead-temperature policy layered on top.

Exhibit 5 — The agent graph: a deterministic router, LLM leaves, one Gate wired spine

Every visit runs this graph. The spine is deterministic and reproducible; only two nodes call an LLM (Research · Copy agent); the Gate is the single point of disposition; and any blocked line falls back to provable copy — so the page degrades, never blocks. Drag to pan · ⌘/Ctrl-scroll to zoom · click a node for its spec · trace a tier.

↩ Fallback path: any line the Gate rejects — on either axis — → the deterministic angle template ships instead (angle_copy). The page never blocks; it degrades to what’s provable and clean. Only two nodes are LLM (Research, Copy); the spine is deterministic, and the Gate stays the single point of disposition. Axis B keeps Provenance’s rule intact: a line that can’t be proven OR can’t be defended as non-spam is structurally unable to ship — never hand-excluded (Constitution Art. II).

Exhibit 6 — Every node explained: input → branches → output → default → code

The same agent graph as a table — the accessible, printable form of the diagram above. Click a node in Exhibit 5 and its row flashes here.

Read top-to-bottom, the tree never widens a claim it can’t back and never asks for more than trust earned: identity gates permission, permission gates the play, the play gates the ask, and the Gate gates the send. Every branch has a provable default, so the worst case is always honest, never broken.