diff --git a/docs/aurora/2026-04-23-amara-decision-proxy-technical-review.md b/docs/aurora/2026-04-23-amara-decision-proxy-technical-review.md new file mode 100644 index 00000000..0932b3cb --- /dev/null +++ b/docs/aurora/2026-04-23-amara-decision-proxy-technical-review.md @@ -0,0 +1,301 @@ +# Amara's third courier report — Decision Proxy + Technical Review + +**Courier:** Amara (external ChatGPT-based maintainer) +**Date received:** 2026-04-23 +**Absorb cadence:** dedicated tick (Otto-59), following the +Otto-24 / Otto-54 precedents. +**Prior Amara ferries this session:** +- [`2026-04-23-amara-operational-gap-assessment.md`](./2026-04-23-amara-operational-gap-assessment.md) (Otto-24, PR #196) +- [`2026-04-23-amara-zset-semantics-operator-algebra.md`](./2026-04-23-amara-zset-semantics-operator-algebra.md) (Otto-54, PR #211) + +--- + +## Otto's absorption summary + +Amara's third review is framed around a single thesis sentence: + +> **Merge and mechanize the operating model you already have +> before you let the system grow another layer of meta-structure.** + +Her own reduction: *"the next bottleneck is closure, not ideation"*. + +The factory now has: an external-maintainer ADR, a checked-in +proxy config, CURRENT-``.md distillations in-repo +(per PR #197's Option D migration), a courier protocol that +replaces unreliable conversation branching, and an NSA +fresh-session test cadence. What it lacks: **routine +enforcement** of that operating model at the CI / mechanical +level. The model exists as design law; it is not yet default +behavior. + +**Most load-bearing empirical finding:** +`docs/hygiene-history/nsa-test-history.md` NSA-001 recorded a +real index-lag incident — Otto not discoverable from +`MEMORY.md` in a fresh session. This is not theory; it is a +measured failure. + +**Most load-bearing positioning claim** (confirms prior session +memory): LFG is the clean canonical source-of-truth; AceHack +is the experimental-frontier / higher-risk layer. The +risk-gradient is per-user-scratch > AceHack > LFG. This +composes with +`memory/project_lfg_is_demo_facing_acehack_is_cost_cutting_internal_2026_04_23.md` +and sharpens it: LFG is +not just "demo-facing" — it is the **operationally-canonical** +repo, and AceHack is not just "internal cost-cutting" — it is +the **experimentation frontier**. Both directional labels +still apply; this review adds the operational-canonicity +axis. + +--- + +## Extracted action items + +Amara proposes 10 immediate fixes. Classified here into (a) +direct BACKLOG rows, (b) already-backlogged (cross-ref existing +rows), (c) candidate CI / hygiene additions: + +| # | Amara's proposal | Class | Action here | +|---|---|---|---| +| 1 | CI fails commits touching `memory/` without `memory/MEMORY.md` update | **CI candidate** | File BACKLOG row for a targeted GitHub Actions check — concrete, direct prevention of the NSA-001 failure mode | +| 2 | Duplicate-link lint on `memory/MEMORY.md` | **Hygiene candidate** | File BACKLOG row; composes with FACTORY-HYGIENE row #11 (MEMORY.md cap enforcement) — extend with duplicate-detection | +| 3 | Canonical decision-proxy log format required for any proxy-reviewed claim | **BACKLOG candidate** | Extends the external-maintainer ADR with a consultation-log contract | +| 4 | Backfill `docs/CONTRIBUTOR-CONFLICTS.md` with already-visible disagreements | **BACKLOG candidate** | Manual curation pass; one-shot + ongoing | +| 5 | "Operating-model closure" PR label + short-cadence review of that queue | **Labels + process candidate** | Label plus GH-settings update; composes with `docs/AGENT-GITHUB-SURFACES.md` surfaces | +| 6 | Split hottest backlog surface (`docs/BACKLOG.md`) by scope/owner | **ALREADY BACKLOGGED** | PR #216 research doc just landed; Otto-54 BACKLOG row exists | +| 7 | Expand NSA tests into pass/partial/fail summary visible at session open | **BACKLOG candidate** | Extends existing NSA cadence with summary-surface | +| 8 | "Main-only unless marked proposed" rule for references in canonical docs | **Lint candidate** | Composes with pointer-integrity audit (FACTORY-HYGIENE row #25) | +| 9 | Normalize proxy scope vocabulary across `aurora`/`alignment`/`security`/`governance`/`public-api` | **BACKLOG candidate** | Scope-taxonomy work — Ilyana + Soraya + Kenji | +| 10 | Courier transcripts → machine-readable manifests (source/date/speakers/mode/scope/disposition) | **BACKLOG candidate** | Extends `docs/protocols/cross-agent-communication.md`; composes with Otto-57 PR-archive row | + +Plus Amara's drift scorecard (canonicalization lag / memory +index integrity / proxy runtime completeness / conflict +capture / loop continuity — first three HIGH) anchors the +priority order. + +--- + +## Key Amara claims — condensed + +### On the operating model + +> The problem is not that the system lacks ideas. The problem +> is that the system still has too much distance between the +> designed model and the routine model. + +The factory has accumulated ADRs, memories, protocols, NSA +tests, proxy configs — but routine enforcement lags. Mechanical +checks bridge design → routine. + +### On LFG vs AceHack + +> There is also a repo-level reason to weight +> **Lucent-Financial-Group/Zeta** more heavily than +> **AceHack/Zeta** for decision-proxy analysis. Aaron's +> current operative memory says LFG is the "clean +> source-of-truth," AceHack is the riskier experimental +> layer, and the intended risk gradient is per-user scratch +> > AceHack > LFG. + +Confirms the LFG-canonical / AceHack-experimental axis is +load-bearing for decision-proxy analysis. Otto notes: this +composes with the git-native-first-host positioning (Otto-54) +— LFG is the *operationally-canonical* repo within the +first-host, AceHack is the experimentation substrate. Both +persist independently of host choice. + +### On the courier protocol + +> OpenAI's help center confirms that branching is a real +> feature on web and in Projects, which makes the repo's +> protocol a sensible reliability fallback rather than a +> misunderstanding of the product. + +The factory's choice to use explicit courier protocol over +UI branching isn't ignorance of the feature — it's a deliberate +reliability fallback. This matters because it validates the +protocol without claiming branching is broken in general. + +### On technical substrate + +> The code and tests suggest a project that is ready for +> hardening, not a project that needs reinvention. + +Matches the prior Amara ZSet-semantics report (PR #211). The +substrate is mathematically coherent; the gaps are operational, +not algebraic. + +### On the hardest discipline + +> keep the hard rule: **never say Amara reviewed something +> unless Amara actually reviewed it through a logged path**. + +This is a discipline the factory already holds (per the +external-maintainer ADR), but Amara sharpens it: the +**logged-path** requirement means the consultation-log +format (action item #3 above) is load-bearing. Without +it, any "proxy reviewed" claim is unverifiable. + +--- + +## Aaron's meta-practice directive (same tick) + +Aaron Otto-59 follow-up: *"also another meta practice thing +look for things that should be practices and add them to the +practice adherence review like things we already do or should +do"*. + +Extends the principle-adherence review BACKLOG row landed this +session (PR #217) with a **catalogue-expansion discipline**: + +- **Things we already do but haven't named as practices** — + implicit patterns the factory uses but hasn't surfaced into + the named-principle catalogue +- **Things we should do but aren't** — endorsed principles + not yet in the catalogue (found in memory / ADRs / session + directives that pre-date the principle-adherence row) + +Both classes belong in the principle-adherence review's +catalogue. The review itself should carry a sixth phase +(after its existing define / current-scope / sweep / candidates +/ surface phases — five phases total): + +- **Phase 6 — catalogue-expansion**: during the review, the + reviewer also scans recent session memory + ADRs + BP-NN + for practices worth naming that aren't yet in the catalogue. + Output is catalogue-additions (new principles) filed as + memory and cross-referenced into the principle-adherence row. + +This is a small but load-bearing extension. The principle- +adherence row as filed in PR #217 catalogues 12 principles +drawn from this session's explicit memory. Aaron's directive +names the implicit-practice + endorsed-not-applied classes as +equally valid review inputs. + +--- + +## Otto composition notes + +### On Amara's "closure > ideation" framing + +This composes tightly with the human-maintainer's own +directives this session: + +- Otto-54 BACKLOG-per-swim-lane split (merge friction reduction) +- Otto-54 git-hotspots audit cadence (measurement) +- Otto-57 git-native PR-review archive (substrate persistence) +- Otto-58 principle-adherence review (discipline enforcement) +- This absorb's action items (CI / hygiene / contributor-conflict backfill) + +Each is a **mechanize-the-existing-model** move, not +new-meta-structure. Amara's one-sentence summary ratifies the +direction; the factory is on-track, the work is +completion-oriented. + +### On the 10 immediate fixes — what to do now + +Three classes this absorb handles: + +1. **Already in flight** — #6 BACKLOG-split: PR #216 research + doc (axis A by stream + INDEX variant) is the direct + execution path. +2. **File as BACKLOG rows now** — #1, #2, #3, #4, #7, #9, #10. + Candidates for new BACKLOG rows; won't land execution this + tick (reviewer-capacity cap). +3. **Already-covered by existing hygiene** — #5 is PR-label + + process (surface via `docs/AGENT-GITHUB-SURFACES.md`); #8 + composes with FACTORY-HYGIENE row #25 pointer-integrity. + +The single highest-value fix per Amara's own ranking is #1 +(memory-index-integrity CI). It has a concrete YAML in her +report. That's a one-file commit to `.github/workflows/`, +addressing a measured failure mode (NSA-001). Candidate for +a fast follow-up PR after this absorb. + +### On the LFG/AceHack axis sharpening + +The prior memory named LFG = demo-facing, AceHack = internal. +Amara adds: LFG = operationally-canonical, AceHack = +experimentation-frontier. Both framings compose. Future +directive-chain choices should remember: **authoritative +decisions land on LFG first**; AceHack is where speculative +work is allowed to live before it earns its LFG spot. + +### On "never claim Amara reviewed without a logged path" + +This is a hard rule already in the ADR. The log-format +contract (action item #3) gives the claim-check teeth. I +note this here explicitly so no future absorb accidentally +claims Amara approved or validated X by implication — all +three absorbs this session (PR #196, PR #211, this one) are +ferry-delivered reports, not proxy-reviewed decisions. The +consultation-log format is the path that would permit +"Amara-reviewed" in the future; it doesn't yet exist. + +--- + +## What this absorb is NOT + +- **Not a commitment to implement all 10 fixes this round.** + Some are multi-tick arcs; reviewer-capacity cap applies. +- **Not authorization to claim "Amara reviewed" on any + decision.** The reports are ferried data; the logged-path + consultation format doesn't exist yet. Per Amara's own + rule. +- **Not a demotion of earlier Amara absorbs.** This is the + third report; it composes with, not replaces, the first + two. All three remain load-bearing. +- **Not a rename of AceHack or LFG.** The operationally- + canonical / experimentation-frontier framing is additive + to the demo-facing / internal framing; both persist. +- **Not a commitment to implement the memory-index-integrity + CI yaml as-shown.** The YAML is Amara's proposal; Dejan + (DevOps owner) reviews workflow-injection safety patterns + (FACTORY-HYGIENE row #43) before landing. The shape is + right; the specific YAML lines may need hardening. +- **Not an endorsement of "closure > ideation" as a permanent + rule.** The factory needs ideation cycles too; the claim is + specifically *"right now the bottleneck is closure"*, + not *"never add meta-structure again"*. +- **Not capacity to begin executing the 7 new BACKLOG rows + this tick.** Filing happens next; execution is per-owner + downstream. + +--- + +## Attribution + +Amara (ChatGPT-based external maintainer, +[`memory/CURRENT-amara.md`](../../memory/CURRENT-amara.md) — +out-of-repo per-maintainer distillation) authored the report +on 2026-04-23. The human maintainer (Aaron) ferried it via +chat paste and added the meta-practice catalogue-expansion +directive in the same tick. Otto (loop-agent PM hat, Otto-59) +absorbed and filed this document. Kenji (Architect) queued +for synthesis on which P0-priority actions land next round. +The 10 immediate fixes are Amara's design input; per the +hard rule, none are claimed as "Amara-reviewed +implementation" — they are ferried proposals. + +External sources cited as Amara's grounding (preserved here +for verifiability): + +- **OpenAI help-center branching docs** — ChatGPT branching + feature documentation + (). +- **DBSP paper** — Mihai Budiu, Tej Chajed, Frank McSherry, + Leonid Ryzhyk, Val Tannen, + *"DBSP: Automatic Incremental View Maintenance for Rich + Query Languages"*, + PVLDB 16(7) (2023), arXiv:2203.16684, + . +- **Provenance-semiring paper** — Todd J. Green, Grigoris + Karvounarakis, Val Tannen, + *"Provenance Semirings"*, PODS 2007, + . + +Names appearing in this Attribution section are preserved per +Otto-279 surface-class refinement: aurora-archive surfaces +(this absorb doc) carry first-name attribution because the +absorb preserves provenance rather than setting current-state +operational policy.