Draw the graph, route on history, and test policy against history

[blocksifrdev]

The three places the doctrine was still ahead of the code.

1. The graph was ASCII

It is supposed to be the visual model of execution authority — and it is what a developer sees in the first sixty seconds.

pctr graph --svg
pctr graph --svg route.svg payments.transfer   # with the selected route drawn

Columns run principal → agents → tools → actions. Agents carry framework, live trust and a clock mark when evidence is stale; actions are coloured by severity and marked ! when protected; naming an action draws its route in green and dims the rest. Standalone SVG — no fonts, no scripts, no network.

Every graph carries an aria-label describing the route in words. I rendered it to PNG and actually looked at it: the first two passes had tofu boxes where the brand fonts have no arrow or clock glyph, so those are drawn as paths now, and the legend no longer overlaps itself.

2. Routing ignored what happened

resolveRoute now accepts accumulated history and prefers routes that work.

The constraint that matters: history only reorders routes that already passed every admissibility check. A flawless record buys no authority. There's a test that hands an unauthorized agent twenty clean receipts and asserts it still cannot route — optimization happens after admissibility, never instead of it.

3. No way to test a policy against reality

pctr whatif --policy '{"approvalThresholds":{"amount":25000}}'

Executions replayed   10
Decided the same      5
Would tighten         3
Would loosen          1

1 execution(s) that were refused would now proceed

Nothing executes — each recorded receipt is re-decided under the proposed policy. Exits non-zero when a change would loosen anything, so it gates policy edits in CI.

Writing its test surfaced something worth stating outright, and it's now its own test: raising an approval threshold cannot unlock a CRITICAL consequence, because severity comes from what the action can cause, not from the rule that reads it.

Verification

167 tests (11 new), Python 19, AGT parity green at 52 checks.

🤖 Generated with Claude Code

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.