Skip to content

docs: v2.0 substrate cascade addenda (#193, #197, #198, #199, #201) - #257

Merged
robotrocketscience merged 5 commits into
mainfrom
docs/v2-substrate-cascade-addenda
Apr 28, 2026
Merged

docs: v2.0 substrate cascade addenda (#193, #197, #198, #199, #201)#257
robotrocketscience merged 5 commits into
mainfrom
docs/v2-substrate-cascade-addenda

Conversation

@robotrocketscience

Copy link
Copy Markdown
Owner

Summary

Five short evaluation memos cascading from #196's substrate ratification (Option B, single-axis Beta-Bernoulli, merged at PR #203). Each memo gives a recommendation, decision asks, and downstream impact.

Issue Memo Recommendation
#197 dedup docs/v2_dedup.md Ship at v2.0 — stdlib-only port, substrate-neutral
#198 multimodel docs/v2_multimodel.md Defer to v2.x — strict bench-impact gate; substrate decision removed strongest use case
#193 sentiment-from-prose docs/v2_sentiment_feedback.md Ship at v2.0, opt-in via .aelfrice.toml, default off
#199 enforcement docs/v2_enforcement.md Split — H3 ships, H1 defers with bench gate, H2 drops (security)
#201 relationship_detector docs/v2_relationship_detector.md Ship at v2.0 — closes README promise gap

All recommendations; user decides per memo's Decision asks section.

Notes

Test plan

  • Each memo reviewed in isolation; recommendations adopted, modified, or overridden per Decision asks.
  • On adoption: `needs-spec` label removed from each issue; downstream tracker labels (e.g. `bench-gated`, `wontfix`) applied per memo.
  • No code changes; CI is docs-only.

Substrate-cascade addendum to substrate_decision.md (#196 ratified Option B).
Recommends shipping the ~254 LOC stdlib-only dedup port at v2.0.
Substrate-neutral; reuses existing SUPERSEDES edge type.
Substrate-cascade addendum. Recommends defer to v2.x with strict
bench-impact gate per V2_REENTRY_QUEUE.md decision #1. Privacy +
determinism + cost trio plus loss of strongest historical use case
under single-axis substrate ratification.
Substrate-cascade addendum. Recommends ship at v2.0 as opt-in
(.aelfrice.toml flag, default off) with PRIVACY.md entry. Closes
the practical-feedback-loop gap that explicit aelf feedback calls
don't fill. Substrate-neutral; emits scalar feedback events.
Substrate-cascade addendum. Recommends splitting #199 by submodule:
H3 selective injection ships at v2.0; H1 directive detection defers
with bench gate; H2 compliance audit drops (command_succeeds runs
arbitrary shell on user machine — no security model).
Substrate-cascade addendum. Recommends ship at v2.0 — closes the
gap between research-line README contradiction-detection promise
and shipped explicit-only triple_extractor. Stdlib regex; pairs
with #197 dedup via shared candidate-pair pool.
@robotrocketscience
robotrocketscience merged commit 0991bd8 into main Apr 28, 2026
8 checks passed
@robotrocketscience
robotrocketscience deleted the docs/v2-substrate-cascade-addenda branch April 28, 2026 23:46
robotrocketscience added a commit that referenced this pull request Apr 28, 2026
…ion-trigger (#259)

## Summary

Two specs covering the last truly-undrafted v2.0 needs-spec issues.

| Issue | Spec | Shape |
|-------|------|-------|
| #228 | docs/v2_wonder_consolidation.md | Research-plan spec — adopts
aelfrice-lab pre-registration as public commitment, defines four
adoption criteria for ship/ensemble/defer/drop |
| #229 | docs/v2_phantom_promotion_trigger.md | Decision spec — rejects
three naive triggers, recommends explicit-acknowledgment rule
(\`aelf:validate\` + lock-text-match), labeled-corpus benchmark gates
ship |

Companion to PR #257 (the substrate-cascade addenda for #193, #197,
#198, #199, #201). After both PRs merge, the v2.0 \`needs-spec\` queue
is fully drained.

## Notes

- #228 is research-blocked, not write-blocked. The lab pre-registration
(\`aelfrice-lab/experiments/wonder-consolidation/HYPOTHESES.md\`) is the
campaign blueprint; this memo commits the public side to its decision
criteria.
- #229's recommended rule has no tunable threshold by design — promotion
is binary on user typing \`aelf:validate\` or \`aelf:lock\`. The
benchmark verifies recall (≥70%) and precision (≥90%) under simulated
user behavior, not a knob.
- Both specs reference
[substrate_decision.md](https://github.com/robotrocketscience/aelfrice/blob/main/docs/substrate_decision.md)
(#196 ratified Option B). Single-axis substrate compatible.
- Mutex: Gylf label applied to both issues at draft start.

## Test plan

- [ ] Each memo reviewed against its issue body. Recommendations
adopted, modified, or overridden per Decision asks.
- [ ] On adoption: \`needs-spec\` label removed; downstream sub-issues
filed (\`wonder_ingest\`, \`wonder_gc\`, \`aelf:validate\` surface) per
memo § Out of scope.
- [ ] No code changes; CI is docs-only.
robotrocketscience added a commit that referenced this pull request Apr 29, 2026
## Summary

Three cascade addenda to `docs/design/write-log-as-truth.md`, one per
post-#205 chain issue. Drafts the formal spec memos called for by the
`needs-spec` label on each.

- `docs/v2_replay.md` — issue #262 (`replay_full_equality`
flip-readiness probe)
- `docs/v2_derivation_worker.md` — issue #264 (derivation worker)
- `docs/v2_view_flip.md` — issue #265 (view-flip — \`ingest_log\`
becomes canonical)

Pattern matches the substrate-cascade addenda merged via #257 / #259:
each memo states what's being decided, recommendation, decision asks,
downstream impact, out-of-scope, provenance. Recommendations only;
ratification is a separate sweep.

## Recommendations at a glance

| Issue | Memo | Recommendation |
|---|---|---|
| #262 | v2_replay.md | Shape-equality at ingest time, not bit-equality.
Posterior drift and legacy-origin backfill bucketed separately, not
flagged as drift. |
| #264 | v2_derivation_worker.md | Single-transaction ingest,
idempotency on (derived_belief_ids non-empty AND beliefs exist),
recover-by-replay between batches, edges in same pass as beliefs. |
| #265 | v2_view_flip.md | \`WRITE_LOG_AUTHORITATIVE\` flag with
two-release soak before default-on. \`aelf rebuild\` preserves
alpha/beta and replays feedback. \`--rule-set\` and \`--as-of\` ship as
stubs in v2.x. Drift alarm wires \`replay_full_equality\` into \`aelf
doctor\`. |

## Cross-links

- Source-of-truth design memo: \`docs/design/write-log-as-truth.md\`
(already on main).
- Substrate dependency: \`docs/substrate_decision.md\` (#196 ratified
Option B). Beta-Bernoulli's posterior drift behavior motivates
shape-equality over bit-equality in #262.
- Federation cross-link: \`docs/design/federation-primitives.md\` (the
log row is the federation atom in v3).
- Chain order: #205 (merged) → #261#262#264#265.

## Test plan

- [ ] Memos render on GitHub without broken cross-links.
- [ ] Each memo's Decision asks section is reviewable in isolation
(don't need to read the whole chain to ratify a single ask).
- [ ] Recommendations are consistent across the three memos (the
equality contract in v2_replay.md matches the worker's invariants in
v2_derivation_worker.md, which match the rebuild semantics in
v2_view_flip.md).
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant