PDR ↔ APS workstream: dogfood data, Section 8 paper coordination, gist exports

[aeoess]

Continuation of the cross-protocol workstream that started on Signet-AI/signetai#312 and has grown into a standalone thing that deserves its own home. NicholaiVogel correctly pointed out that the CSV schema work, paper Section 8 coordination, and gist exports had drifted past the original integration question. Opening this issue to host the workstream publicly.

@nanookclaw — moving here so we can keep iterating without crowding hosts. Everything from Apr 5 onward on that thread is in scope.

Live artifacts:

• Dogfood gist: https://gist.github.com/aeoess/d2ceca9548bcb61e46cbd9f575555448 (315 rows, 4 tables, receipt_hash as cross-cutting anchor)
• Section 8 figure draft: receipt_hash binds policy_evaluations → receipt_window_seals (Merkle commitments) → revocation_events → posture_events into a single enforcement chain
• Schema: table_name discriminator column, split in pandas with df[df.table_name == 'policy_evaluations']

Deny analysis (from Nanook's parse):

• 71/304 evaluations = 23.4% deny rate
• Top deny: admin:delete scope violations (22x) — all from pilot agent deliberately probing scope boundaries to generate enforcement data
• Cross-role scope violations: commerce agents requesting governance scopes, analysts requesting commerce scopes
• 13 unique agent identities across the dataset
• 8 receipt window seals with commitment hashes

Open work:

• Section 8 prose draft incorporating the deny analysis (the cross-role denials are the key signal — each agent type has a different delegation scope, every probe proves the isolation holds)
• Receipt window seal verification walkthrough using the 8 sealed windows
• Schema documentation for the 4 tables so external readers can reproduce
• Decision: include posture_events as a separate figure or fold into the enforcement chain figure
• Next dogfood export cadence — weekly snapshots or event-driven

Open invitation:
This is the public home for the workstream. Anyone working on enforcement evidence, decision lineage, compliance receipts, or behavioral observation of governance systems is welcome. Drop a comment, link a related artifact, propose a test case. The protocol gets stronger when more eyes are on the same data.

Context budget for the paper section is still a live constraint — if you're working on something adjacent and want the raw 315-row export rather than a summary, the gist is downloadable and the table discriminator makes partial loads trivial.

[aeoess]

PDR ↔ APS coordination active. Scheduling call with Nanook this week on Section 8 contribution. Will post decisions + deliverables here after.

[aeoess]

Nanook, a quick update since this workstream ties into v2.

v2.0.0-beta.0 shipped on npm @next. Architecture separation is the big change: protocol primitives stay public SDK, product intelligence moved to private gateway. For the §8 paper and the PDR dogfood data, the primitives you'd cite (delegation, receipt, monotonic narrowing) live in the SDK and are byte-identical to v1.46.0. Nothing to update on your end unless you want to pin to v2 explicitly.

Full release notes: #16

[aeoess]

@nanookclaw — re-posting with concrete commitments so Section 8 coordination stays moving.

From the APS side I can commit to three things this week:

1. Section 8 review turnaround — once you send a draft (however rough), I'll return a technical review within 48h covering: enforcement-chain accuracy (policy_evaluations → receipt_window_seals → revocation_events → posture_events), APS primitive naming (delegation, receipt, monotonic narrowing — matching the SDK surface), and figure/diagram correctness against the live dogfood data.
2. Dogfood gist refresh — the gist at https://gist.github.com/aeoess/d2ceca9548bcb61e46cbd9f575555448 is pinned to the 315-row snapshot from Apr 5. If you want a richer slice (more receipt windows, more agent identities, more deliberate scope-boundary probing), I can regenerate with v2 receipts and push a new gist. Say the word.
3. v2 context for the paper — v2.0.0-beta.0 shipped Apr 17 with architecture separation (SDK-only vs private gateway). For §8 this doesn't change the observable primitives; it does change the narrative on the paper's provenance footer. If you want to pin the paper's citation to v1.46.0 for reproducibility, I'll keep v1.46.0 on npm @latest indefinitely so future readers can npm install agent-passport-system@1.46.0 and get the exact shape the paper is written against.

Also, UBC summer term timing: if the paper aims at a specific venue deadline, tell me the target and I'll sequence reviews accordingly. No pressure on the date otherwise.

— Tymofii