M0-01R — Prove explicit observation lineage and baseline/override projection #7
Labels
No labels
bug
duplicate
enhancement
help wanted
invalid
question
wontfix
No milestone
No project
No assignees
1 participant
Notifications
Due date
No due date set.
Dependencies
No dependencies set.
Reference
Creatidy/model-intelligence#7
Loading…
Add table
Add a link
Reference in a new issue
No description provided.
Delete branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Authorization and Predecessor
Adrian explicitly authorized this ONE new bounded issue and its implementation PR; this issue is owner-selected upon creation. Freshly fetched canonical develop is
6edd31d491a41a92e73ebfc6d9ee81c284129f8a, containing merged PR #6. One normal checkout, ordinary issue branch; no new git worktrees.Refs #3 and PR #4 (#4), failed M0-01 experiment frozen at
303f5fe118a592f3cf5e7ea47c5955625850ea28, unmerged STOP_REVISE. Use only architectural/adversarial evidence and provenance; no patching or cherry-picking its domain implementation. Repeated failures demonstrated that values, timestamps, row presence and aggregate snapshot history cannot safely reconstruct revision/cessation.Objective and Accepted Scope
Prove whether the frozen #1 A/B/C cases become substantially simpler with explicit observation/claim identity and provenance, explicit revision lineage, normal/default baseline terms, bounded typed temporary overrides, and historical evidence separated from effective projection. Small typed Python architecture proof, not a service or live market truth; prefer zero runtime dependencies.
Observations: explicit observation_id, source identity/authority/type/reference, observed/published time when known, retrieval time and distribution/license. Payload equality never supplies identity; re-retrieval never supplies revision or freshness renewal.
Claims: explicit claim_id, observation reference, subject, typed payload, validity/applicability and optional
revisesclaim ID. Separate model/capability/pricing, plan/offer economics and execution-surface families; no generic EAV. Unknown lineage stays independent. Reject self-reference and other impossible lineage. Never infer revision from source, time, value, snapshot disappearance or another source.Offer projection at T: baseline head plus applicable override heads, field-wise typed optional override fields. Omitted fields inherit baseline. Bounded overrides support future starts, natural expiry, explicit revision (shortening/extension/term change), explicit withdrawal, and later independent campaigns. Independent disjoint fields may compose; overlapping incompatible values/boundaries remain conflicts without recency/source/lexical/trust winners or implicit discount/multiplier composition. Authority is provenance. Conditions and zoned windows are explicit; private quota state is excluded.
History retains expired/superseded claims. Current projection exposes baseline head(s)/active overrides/effective terms/current unresolved conflicts, not expired/superseded campaigns. Future heads may be queried separately. Conflicting unresolved applicability boundaries stay visible without applying expired terms.
A: stable model/provider identity, contextual capability/benchmark/pricing evidence, independent sources and preserved disagreement, no universal score/winner; another source's equal value is independent of explicit revision.
B: synthetic plan identity/generation, baseline, promotions/restrictions, interval/timezone/window/conditions and explicit lineage.
C: advertised/observed/selectable/enforceable for headless, structured output, workspace editing, model selection and physical-model observability/enforceability; provider-managed surface may have no known physical-model ID. Unknown is not false.
Semantic changes must be explainable solely by identity, explicit lineage and applicability: model observed/evidence revised, price changed, override activated/revised/expired, effective offer changed, execution surface changed, conflict detected/resolved. No
_claim,_corrected,_ceasedor equivalent inference heuristics.Deterministic Acceptance
Cover baseline-only, inactive future campaign, activation, declared-field inheritance, automatic expiry/fallback, active revision, shortened end, extension, term changes, withdrawal before original expiry, historical retention/current exclusion, and later independent activation.
Port valid predecessor invariants: missing row is not cessation; older same-source history cannot close newer claim; another source's matching value cannot suppress explicit revision; superseded history cannot become current correction; retrieval-only changes silent; stale/superseded remain so on retrieval; unresolved conflicting boundaries visible; expired != deleted; current excludes expired/superseded overrides; unknown != false; equal cross-source values do not imply lineage. Discard accidental heuristic expectations.
Hard Stop and Exclusions
Return STOP_REVISE if solving this needs guessed lineage/cessation, value/time/row heuristics, generic DSL/rules machinery or large provider-specific hierarchy. Report modules, production LOC, public domain types, private helpers and runtime dependencies, with explanation relative to PR #4.
No REST/MCP server/database/SQLite/Postgres/scheduler/background jobs/event sourcing/generic rules engine/DSL/provider clients/UI/Scarcity Router/creatidy-autonomy, private accounts/credentials/runtime state, release/deploy/main changes or merge.
Validation and Delivery
Deterministic network-free lineage, baseline/override lifecycle, conflicts, A/B/C and adversarial tests; full
make check; package build; installed-wheel smoke;git diff --check. Record exact base/head and checks. One ordinaryissue-<number>-<topic>branch from fetched develop, normal Adrian-authored commits/pushes, ONE develop PR with Refs (no automatic closing). Issue stays open.Automatically run repository-native /finish-pr: fresh foreground pr-reviewer native task, exact frozen whole-PR review, direct findings, at most three remediation rounds including fixes to fixes, fresh review after each. No owner relaying or substituted self-review. Terminal report only READY_TO_MERGE, STOP_REVISE, OWNER_DECISION_NEEDED or finite BLOCKED. Never merge.
Implemented M0-01R in #8 (Refs #7), ordinary branch
issue-7-explicit-lineage, same normal checkout. Base6edd31d491a41a92e73ebfc6d9ee81c284129f8a; implementation HEAD567c3aa43b1247e646d3852e340889860eb7310e, canonical push/refs and clean status verified under effective Adrian identity. No predecessor domain code reused; #3 / PR #4 remain architectural/adversarial evidence only.Explicit observation/claim identity and revision links; immutable historical evidence; baseline + bounded typed field overrides; separate current/future/history; unresolved conflicts with no authority/recency winner. Synthetic A/B/C and all required adversarial/lifecycle cases. 638 production LOC across 3 modules (including docstring init), 22 public classes/2 aliases, 2 private helpers, zero runtime dependencies. More explicit typed data, no inferred revision/cessation machinery.
make checkPASS (57 tests, strict basedpyright 0 errors/warnings, locked sync, Ruff lint/format and whitespace).uv buildPASS; isolated offline installed-wheel smoke PASS (projection/expiry/history, py.typed, zero dependencies, installed outside src). Full base delta whitespace/ancestry checked. Native /finish-pr bounded review/remediation starts automatically; no independent approval yet. Issue remains open; no merge/main/release/deploy.Automatic /finish-pr for PR #8 terminated BLOCKED: fresh foreground native
task(subagent_type="pr-reviewer")returnedUnknown agent type: pr-reviewer is not a valid agent type. Repository reviewer file exists but is unavailable to this runtime;kilo debug agent pr-revieweralso failed because thekiloexecutable is absent. No independent verdict, 0 completed reviews, 0/3 remediation rounds. No self-review/alternate agent/external controller/owner-relaying substitute.Final canonical/local/MCP HEAD is
567c3aa43b1247e646d3852e340889860eb7310e; develop/base/merge-base unchanged at6edd31d491a41a92e73ebfc6d9ee81c284129f8a. Clean normal checkout, pushed PR #8 remains open/unmerged; issue stays open. 57 passing checks/tests, build and isolated offline wheel smoke documented in the delivery update. Blocker is native reviewer availability, not failed architectural semantics or merge approval.Owner-authorized successor proof: issue #9 #9 and PR #10 #10. Fresh branch from canonical develop; PR #8 remains unmerged at
c70bbd380aand is not patched/cherry-picked. R2 separates latest-known lineage from effective-at-T, durable baselines from bounded overrides, and explicit evidence additions from projection endpoint deltas. Native bounded review follows automatically.