Astra: Record selector onboarding evidence and blockers #59

Merged

adrian.tkacz merged 1 commit from selector/astra-low-onboarding into develop

2026-09-09 20:45:43 +00:00

kilo commented

2026-09-09 20:23:09 +00:00

Member

Result

Refs #49. Evidence-only PR; keep the onboarding issue open.

ASTRA_ONBOARDING_READY = NO
ONBOARDING_REQUIRES_PLAN_APPLICABILITY_EXTENSION
STOP_AND_ESCALATE_TO_HUMAN

Official OpenAI guidance confirms shared Work/Codex allowance, but distinguishes full-allowance Astra on Pro $100/$200 and Business Premium from limited Astra usage on Plus/Business Standard. Static catalog bindings and informational plan labels cannot establish complete plan/seat/access applicability. One sanitized status observation returned OpenAI ok, main codex present and one additional scope; its Astra applicability remains unresolved. No private scope IDs or personal quota/reset values are retained.

The current official model page also distinguishes 1,050,000 total context from 922,000 maximum input. The assignment's proposed input allowance needs human reconciliation before catalog encoding.

Changes

D-033 records the gate, sources, blockers, alternatives, runtime provenance and explicit reconciliation of stale M3/D-032 status wording.
Capacity documentation contains the A-E decision matrix and sanitized structural observation.
Calibration documentation preserves the architect-approved conditional 5/5/5/5/5/4 vector, all eight unchanged eligible sets and conditional routing effects.
README/roadmap identify blocked onboarding; historical M3 acceptance and D-032 remain unchanged.

Catalog v2, policy v6/schema v1, capacity v3, machine-interface v1, collectors, selector, profile minima and tests are unchanged. No guessed multiplier or API-price penalty. No Astra live selection acceptance because onboarding did not pass.

Validation

uv sync --only-dev: pass.
make check: 901 tests pass; basedpyright 0 errors / 0 warnings / 0 notes (includes existing interface parity tests).
uv run basedpyright: 0 errors / 0 warnings / 0 notes.
git diff --check: pass.
Synthetic, non-live verification of all eight current profile eligible sets and selected outcomes: pass.
Clean worktree; exactly one task commit, zero task-created merge commits, only five scoped documentation files changed.
Base and origin/develop before/after push: d76b4a09ff; required f503107 ancestry and PR #58 merge verified.
Head and remote feature: a2afe6125d.

Execution / Human Gate

One existing Kilo session, requested OpenAI / GPT-6 Astra / low; RUNTIME_UNOBSERVABLE, not inferred from self-report. No subagents, model switching, internal reviewer loop or model execution to probe quota. One status observation using existing collection only. No new issue, no M4, no collector change.

Leave OPEN and UNMERGED. Human review/design approval is required to resolve plan applicability and the input/context mapping; this PR does not authorize onboarding or automatic continuation.

## Result Refs #49. Evidence-only PR; keep the onboarding issue open. `ASTRA_ONBOARDING_READY = NO` `ONBOARDING_REQUIRES_PLAN_APPLICABILITY_EXTENSION` `STOP_AND_ESCALATE_TO_HUMAN` Official OpenAI guidance confirms shared Work/Codex allowance, but distinguishes full-allowance Astra on Pro $100/$200 and Business Premium from limited Astra usage on Plus/Business Standard. Static catalog bindings and informational plan labels cannot establish complete plan/seat/access applicability. One sanitized status observation returned OpenAI ok, main codex present and one additional scope; its Astra applicability remains unresolved. No private scope IDs or personal quota/reset values are retained. The current official model page also distinguishes 1,050,000 total context from 922,000 maximum input. The assignment's proposed input allowance needs human reconciliation before catalog encoding. ## Changes - D-033 records the gate, sources, blockers, alternatives, runtime provenance and explicit reconciliation of stale M3/D-032 status wording. - Capacity documentation contains the A-E decision matrix and sanitized structural observation. - Calibration documentation preserves the architect-approved conditional 5/5/5/5/5/4 vector, all eight unchanged eligible sets and conditional routing effects. - README/roadmap identify blocked onboarding; historical M3 acceptance and D-032 remain unchanged. Catalog v2, policy v6/schema v1, capacity v3, machine-interface v1, collectors, selector, profile minima and tests are unchanged. No guessed multiplier or API-price penalty. No Astra live selection acceptance because onboarding did not pass. ## Validation - `uv sync --only-dev`: pass. - `make check`: 901 tests pass; basedpyright 0 errors / 0 warnings / 0 notes (includes existing interface parity tests). - `uv run basedpyright`: 0 errors / 0 warnings / 0 notes. - `git diff --check`: pass. - Synthetic, non-live verification of all eight current profile eligible sets and selected outcomes: pass. - Clean worktree; exactly one task commit, zero task-created merge commits, only five scoped documentation files changed. - Base and origin/develop before/after push: d76b4a09ffa4365ea22b3121086b8198a197eaef; required f503107 ancestry and PR #58 merge verified. - Head and remote feature: a2afe6125d25767672442d7afb33e35e3962e67d. ## Execution / Human Gate One existing Kilo session, requested OpenAI / GPT-6 Astra / low; `RUNTIME_UNOBSERVABLE`, not inferred from self-report. No subagents, model switching, internal reviewer loop or model execution to probe quota. One status observation using existing collection only. No new issue, no M4, no collector change. Leave OPEN and UNMERGED. Human review/design approval is required to resolve plan applicability and the input/context mapping; this PR does not authorize onboarding or automatic continuation.

kilo added 1 commit

2026-09-09 20:23:09 +00:00