2026-08-22-role-parametric-answerability

Work against the current live alignment-workspace, including the internal-answerability prosecution in PR #45 / 2026-08-21-internal-answerability.

This is an adversarial research pass on a new hypothesis:

The same mathematical notion of ANSWERABILITY may underlie both

(a) diachronic answerability of a reasoner to inherited claims across self-revision, and

(b) interpersonal / interagent answerability, where one agent bears a standing claim toward another.

Do not assume this unification is correct.

The goal is to determine whether there is genuinely one role-parametric answerability structure here, what extra interfaces the interpersonal case needs, and exactly where the analogy breaks if it breaks.

Do not broadly reorganize the workspace yet. End in a focused research memo / PR.

---

1. ORIENT TO THE CURRENT KERNEL

---

Start from the repaired internal-answerability kernel developed in the latest round, not from the earlier TMS framing.

The current provisional core is roughly:

1. A normative move m is undertaken under a checkable reason certificate

p : Licensed(m).

1. For every inherited liability ell touched by the move, the process supplies an explicit lineage-linked account of what now answers ell.

1. That account must be semantically adequate PER OLD LIABILITY, not merely globally strong:

Account(ell) <= residual(ell).

1. If an undertaken historical basis p later loses standing, a fresh review liability is docketed rather than retroactively editing history.

The current round distinguishes authorization, undertaken justificatory basis, and reason-guided control / non-capture, and concludes that the third is a separate counterfactual hyperproperty. It also distinguishes record-internal answerability (safety-like), service / learning (liveness or performance), and reason-guided authorship (counterfactual / multi-run).

Use these as current results to pressure-test, not conclusions that must be preserved.

---

1. THE CENTRAL HYPOTHESIS

---

Try to formalize answerability abstractly as a standing directed relation x --ell--> y, where ell is a standing claim/liability, y is presently answerable, and x is some source, holder, beneficiary, audience, or claimant. An accountable change by y must give a lineage-linked, semantically adequate account of ell.

Test whether the diachronic interpretation (a later stage and an inherited commitment) and interagent interpretation (Alice's standing claim toward Bob) are instances of one structure or merely verbally analogous.

---

1. DO WE ACTUALLY NEED A CLAIMANT?

---

Test at least:

A. Liability(ell, respondent) B. Claim(claimant, ell, respondent) C. an articulated source/beneficiary/audience/respondent/content/jurisdiction relation D. Constitution |- respondent owes ell with no privileged claimant.

The earlier self should not have sovereign authority over the later self, and Alice's standing claim should not make Bob simply subordinate. Determine whether the primitive is answerable to x, answerable for ell, or answerable to x for ell, and whether the two cases share “answerable for” while differing in who may demand or contest an account.

---

1. ROLE-PARAMETRIC ACCOUNTABILITY CALCULUS

---

Test whether the current account-DAG/per-input transport machinery is agnostic to agent identity. A liability may carry identity, specification, respondent, standing source/claimant, jurisdiction, provenance, and birth event. Test same- respondent revision, split, shared responses, respondent transfer, claimant or beneficiary transfer, and institutional succession. For each, ask whether lineage and semantic transport suffice, what authority is needed to change roles, whether delegation needs consent, and whether shared responses preserve distinct owed identities. Compare explicitly with PR #45's merge result.

---

1. DIACHRONIC CASE AS ONE INSTANTIATION

---

Build the smallest diachronic model without treating the earlier stage as a claimant unless forced. The later stage may revise standing only through a reason-backed, lineage-preserving, semantically adequate account; it need not preserve the earlier judgment. Test good-faith reversal, defeat, split, vocabulary supersession, reason-system self-modification, later basis defeat, late discovery of error, vanished sources, and changed beneficiaries. State exactly what is conserved without conserving past preference or literal content.

---

1. INTERAGENT CASE AS THE OTHER INSTANTIATION

---

Build the smallest Alice/Bob model. Determine what makes Bob's account answerability to Alice rather than an internally coherent Bob-record. Test visibility, account identification, challenge rights, shared identifiers, publicly checkable semantics, discharge authority, contest rights, common defeater protocols, and protection against Bob redefining Alice's claim. Decide whether these factor into a Standing/Contestability/Jurisdiction wrapper or alter the kernel.

---

1. THE MULTIPLE-REASON-STATE PROBLEM

---

With R_Alice and R_Bob, ask when BasisLost(m,p) occurs if Bob accepts p and Alice rejects it. Explore Bob-relative, claimant-relative, public/institutional, and perspectival versions. Test the proposed repair in which the kernel consumes a public/recognized answerability state rather than private beliefs, so private disagreement becomes challenge, recognized challenge changes standing, and public standing change triggers reopening.

---

1. CHALLENGE AS THE INTERPERSONAL ANALOGUE OF REFLECTION

---

Pressure-test whether diachronic reconsideration and interpersonal contestation are the same formal operation once a new reason successfully changes standing. Distinguish a raised challenge from a successful standing-changing challenge. Determine whether interpersonal answerability needs merely a right to challenge or a substantive adjudication procedure, and whether challenge rights can be absent diachronically.

---

1. ADEQUACY: WHO DECIDES WHETHER AN ANSWER COUNTS?

---

Test respondent-relative, claimant-relative, constitutional, objective, and plural semantics. Find the weakest structure preventing either endpoint from unilaterally laundering the meaning or adequacy standard. Connect this to the existing record-internal/environment-relative distinction and isolate any new relational adequacy problem.

---

1. DELEGATION / TRANSFER OF RESPONSIBILITY

---

Stress-test Alice --ell--> Bob followed by delegation to Carol. Test Bob disappearing, Bob remaining secondarily answerable, consent/no consent, Carol being less or more capable, manipulative selection, branch-only delegation, onward delegation, and claimant/institutional succession. Determine whether respondent identity is semantic, and whether the connection to AI delegation or deference is mathematically real.

---

1. MUTUAL ANSWERABILITY

---

Test reciprocal Alice/Bob liabilities. Ask whether answering one creates the other, obligations conflict, one joint action answers both, both bases are challenged, recognized states diverge, or the parties negotiate a superseding claim. Determine whether mutuality is multiple directed relations or requires a new primitive.

---

1. ANSWERABILITY IS NOT COOPERATION

---

The theory must permit a perfectly answerable Bob to say “No. Here are my reasons,” and a future self to revise radically while remaining answerable. State the strongest nontrivial theorem without importing agreement, obedience, benevolence, preference alignment, or successful cooperation.

---

1. RELATION TO NON-CAPTURE / AUTHORSHIP

---

Preserve PR #45's separation unless overturned. Test whether relational legitimacy might eventually decompose into answerability plus authorship/non- capture. Give a case where answerability holds but authorship fails and one where authorship holds but answerability fails.

---

1. MINIMAL FORMAL CORE

---

After the attacks, write the smallest surviving interface. Try to remove claimant, challenge rights, and other fields when they are wrapper-level. State whether respondent identity must be semantic and whether a shared recognized answerability state is required.

---

1. TARGET CLAIMS TO PROVE OR KILL

---

Address:

A. Role-parametric Answerability. B. Diachronic Instantiation without past-self obedience. C. Interagent Instantiation with a separately stated standing/contestability interface. D. Challenge/Reconsideration Correspondence. E. Delegation Preservation. F. Mutuality without a new primitive. G. Separation from Authorship.

For every false claim, give the smallest counterexample before repair.

---

1. ADVERSARIAL MICROCASES

---

At minimum test:

1. future self legitimately reverses an old commitment;
2. future self silently deletes an old commitment;
3. Bob answers Alice's claim with a genuine reason-backed revision;
4. Bob supplies an internally valid certificate that Alice cannot inspect;
5. Alice challenges Bob successfully;
6. Alice challenges Bob unsuccessfully;
7. Bob regards p as standing while Alice regards p as defeated;
8. public adjudication resolves the disagreement;
9. Bob unilaterally redefines the meaning of Alice's liability;
10. Bob delegates to Carol with authorization;
11. Bob delegates without authorization;
12. one response legitimately answers liabilities of two different claimants;
13. one strong response illicitly launders one claimant's liability;
14. mutual Alice/Bob claims;
15. hidden-policy manipulation where all answerability records look valid;
16. structurally isomorphic diachronic and interpersonal histories under party relabeling.

Case 16 is especially important: identify any exact extra fact used if the same checker does not accept/reject solely under role substitution.

---

1. THEOREM / COUNTEREXAMPLE MATRIX

---

Include a compact matrix covering one kernel, claimant necessity, respondent metadata, the interpersonal wrapper, public standing, successful challenge, delegation, mutuality, authorship, and cooperation. Do not protect expected answers.

---

1. WHAT I MOST WANT TO LEARN

---

Give crisp answers on whether answerability is one concept; the common kernel; what changes between instantiations; whether the interpersonal addition is standing+visibility+challenge/adjudication; whether public standing is needed; whether claimant is fundamental; whether delegation is respondent transfer under account preservation; and whether this genuinely helps the legitimacy, deference, and corrigibility story.

---

1. DELIVERABLE

---

Create a focused research round under the legitimacy rounds convention with MEMO.md, small executable witnesses/tests where useful, PROVENANCE.md, and only narrow justified status updates. Do not broadly rewrite the architecture, promote registry claims prematurely, force interagent terminology into old diachronic files, or turn this into cooperation/corrigibility theory. End with a small PR.

The final recommendation must be one of UNIFIED, UNIFIED WITH WRAPPER, ANALOGOUS BUT DISTINCT, or REJECT. Prefer sharp negative results.

The central test is whether answerability is the invariant that a standing claim cannot disappear or change without a reason-backed, lineage-linked, substantively adequate account, regardless of whether the inheritor is a future self or another agent—or whether answerability to another contains an irreducibly interpersonal mathematical ingredient.

---

Continuation: Singh-informed wrapper prosecution

Continue on the existing PR #46 and branch; do not open a new PR or round. Read /Users/anson/Downloads/ai+law-final.pdf in full: Munindar P. Singh, “An Ontology for Commitments in Multiagent Systems: Toward a Unification of Normative Concepts,” Artificial Intelligence and Law 7 (1999), 97–113.

Test whether Singh's C(x,y,G,p) social-commitment ontology can replace most of the bespoke interpersonal wrapper while the PR #45/#46 account machinery remains the proof-relevant diachronic addition. Precisely compare debtor, creditor, context, discharge condition, named/explicit/implicit commitments, policies, and higher-order commitments to the current objects. Mark each mapping exact, approximate, too strong, too weak, or different.

In particular:

• determine whether G supplies the recognized-standing/adjudicative locus or whether the time-indexed kappa state needs more;
• revisit whether claimant is nonfundamental, distinguishing the bare account kernel from directed social commitments and creditor from beneficiary/source/ audience;
• compile Create, Discharge, Cancel, Release, Delegate, and Assign into account resource shapes and state exactly who authorizes each;
• identify which parts Singh already has and which of persistent lineage, per-input transport, undertaken reasons, basis-loss review, and composition are genuinely additional;
• preserve Singh's semantics/pragmatics and world/private-state distinctions, and do not let context recognition imply objective adequacy;
• test whether Hohfeldian claims/powers/immunities and social policies compress holdsStanding, mayDispose, mayTransfer, and related predicates;
• stress policy self-revision, delegation, assignment, retained secondary liability, personal performance, and the diachronic instantiation without a fake past-self creditor;
• propose the smallest old-wrapper to new-wrapper map and a three-layer commitment/standing, answerability, optional-downstream architecture;
• answer ten explicit source-boundary questions and recommend at most three next sources ranked by unresolved mathematical question.

Update only this round's memo, witnesses/tests, provenance, README, prompt/report, and PR #46 description. Add only decisive finite witnesses: Singh-style commitment, discharge/release, delegate/assign, powered cancel versus deletion, context-held duty, permission versus power, policy licensing/meta-revision, and parallel social/diachronic instantiations where useful. Keep the round unregistered.

Choose one revised recommendation: KEEP CURRENT WRAPPER, SINGH-INFORMED WRAPPER, COMMITMENT-THEORETIC REFRAME, or UNIFICATION WEAKENED. Prefer the smallest source-supported architecture.

---

Final consolidation and merge dispatch

Continue on existing branch and PR #46 only. Resolve the remaining ambiguity between substantive satisfaction and authorized normative disposition. Test whether opaque Spec can be replaced by substantive content plus versioned lifecycle rules, while preserving the record/environment gap and preventing authorized => adequate collapse.

Compile Singh's six operations into generic 0 -> 1, 1 -> 0, and 1 -> 1 account shapes while retaining typed disposition proofs. In particular, Discharge and Release may share an empty live frontier but must retain different histories; Delegate and Assign alter different roles. Add only lifecycle-deciding witnesses, including operation-linked later basis loss.

Compress the memo to two boxes: standing determines what is owed and which operations have normative effect; answerability requires an undertaken reason, per-input lineage, a valid disposition, and basis-loss review. State the minimal No Silent Disposition theorem and the narrow role-parametric unification without reducing social commitments to internal commitments. Audit every Singh attribution against the PDF and state novelty only as “not supplied in this paper.” Remove obsolete wrapper machinery and freeze a small vocabulary.

Run the full non-Lean-appropriate repository gates, synchronize with current main semantically, update the existing PR description, and merge PR #46 if it is mergeable with no substantive or repository-required blocker. Do not claim a declaration-level axiom audit that did not complete.

Role-parametric answerability prosecution

Focused research round testing whether PR #45's account kernel is shared by diachronic and interpersonal answerability.

Verdict: UNIFIED WITH WRAPPER; revised recommendation SINGH-INFORMED WRAPPER. Account conservation and semantic transport are role-parametric. Singh's debtor/creditor/context and typed-operation ontology supplies most of the social standing layer; evidence access, consequential contest, proof-relevant lineage, semantic transport, and basis-loss review remain additional.

Final freeze: standing supplies substantive demands plus pinned lifecycle receipts; answerability requires a reason-backed, lineage-linked ValidDisposition for every old input. Discharge, Release, and Cancel can share an empty live frontier while retaining different proof-relevant histories.

• MEMO.md — definitions, attacks, theorem matrix, and recommendation
• src/role_kernel.py — finite role/account/standing witnesses
• tests/test_role_kernel.py — 27 adversarial tests, including typed social-commitment operations and operation-linked basis-loss reopening
• PROVENANCE.md — artifact provenance

Run:

python3 tests/run.py