Direction backprop#1
Merged
Merged
Conversation
…ation plumbing Wire the reject path of verify_insight to journal.add_rejection so every candidate that fails the register gate is appended to the rejection_log (canonical_form, pareto_axes, gate reasons, closest peer, alias signal) — organic negative signal for future discrimination work. Add the no-op config knobs for the selective-verification pipeline: alias-gap routing (reject/fasttrack thresholds + audit-back rate), paraphrase-perturbation variant count, and committee dissent threshold. All defaults are no-ops until the corresponding phase ships. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…926)
Add the carry-lessons-across-time layer CE was missing. Every N cycles
(default 5) the engine clusters related investigations — connected
components of the entry graph, reusing engine/graph.py — and distills each
cluster into a FRONTIER EDGE: the unexplored synthesis / contradiction at
its boundary. The edge is injected into the next introspection as something
to push PAST, never to confirm.
Adapted to CE's novelty objective rather than Arbor's scalar one, with
deliberate guards against the failure modes:
- abstraction runs on the cross-family verifier model, never the primary,
so the generator can't author and then consume its own prior;
- "frontier edge to exceed" framing resists convergence onto the engine's
own beliefs (CE's value is divergence / OOD synthesis);
- direction_max_injected caps how many priors reach introspection;
- suppress_direction_insight (+ --suppress-direction) kills a bad prior;
- direction_backprop_enabled is a global off switch.
Append-only storage with an overlap-based supersedes chain (a grown cluster
replaces the prior head it overlaps) — never mutates prior records.
Omits Arbor's frontier pruning: for a novelty objective there is no clean
falsification signal, and pruning on a weak proxy (repeated low surprise)
would kill the directions that pay off late.
New: engine/direction_backprop.py, ABSTRACT_DIRECTION_PROMPT, journal
storage + accessors, 6 [engine] config knobs, --abstract-directions /
--show-directions / --suppress-direction admin ops, standalone tests, and
a design spec under docs/superpowers/specs/. README documents the feature
and notes the adversarial verifier as CE's role-analog of Arbor's held-out
merge gate (and why CE rejects a scalar dev/held-out gate).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Took some ideas from Arbor