File the frontier log (F2′): instrument the delegation thesis before its mechanism - #55
Merged
Merged
Conversation
…its mechanism The substrate-theory panel's one actionable finding (PR #54): the product thesis - reversal earns authority - is unobservable by the current dogfood metrics (denial-FP measures friction; standing grants cannot widen pre-W-3, so "no widening" is mechanically predetermined), and the pre-evidence baseline is perishable. dogfooding.md gains the frontier-log instrument: per-family baseline declarations (template included) before W-15a->W-3 lands, biweekly re-declaration, reason- captured ratification decisions (evidence/fatigue/other), the evidence-backed gate (>=3 clean runs + >=1 exercised revert), two-leg measurement (frontier movement; judgment displacement), the kill-test disconfirmation, and stated n=1 limits. Denial-FP honestly retitled the friction metric. Roadmap: W-1 gains the operator-owned baseline step (time-ordered: before W-15a); W-3 gains the F2' measurement clause and north-star counter adoption. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
XVVH
added a commit
that referenced
this pull request
Jul 16, 2026
…-incident rows (#57) * File docs/landscape.md (neighbor survey) + two grade-B field-incident rows Filing 1: docs/landscape.md — July 2026 survey of three neighbors, verified at source (both repos pinned to 2026-07-15 SHAs; Runta docs read 2026-07-16; The Information layer graded as-reported). Thesis: brokered authority boundaries, credential isolation, and hash-chained audit are becoming genre table stakes (three independent builders, weeks apart), while the delegation half — in-place owned-state undo with human-edit preservation, four-lineage manifests, zero-authorship ratification — stays unoccupied; every neighbor demands MORE upfront authoring and sells prevention, not recovery. Two source corrections recorded inline: custodian-kernel ships secret_leak_guard + spend_sentinel (not "LeakSentinel"); cyberware's DEK is subject-scoped, not per-record. Cross-refs the frontier log (PR #55); follow-ups ride in sibling filings (custodian egress-tripwire roadmap candidate; cyberware SI-26..30/F2 evidence survey), noted so docs don't duplicate. Filing 2: docs/field-incidents.md gains an explicit provenance-grade tier (A = primary artifact verified; B = second-hand press, positioning/landscape use only, never founding-example work) rather than stretching the entry rule, plus two grade-B rows from The Information's Runta coverage: an authorized Claude Fable 5 agent deleting production files (owned-state recovery failure — §5.3/A24, undeclared-reversibility, broker_verified guards) and a quant firm's bug-fixing agent injected into executing malicious code (containment failure — W-4 pair, blessed-plan execution). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * Amend incident (a) framing: custody-boundary ambiguity recorded, not claimed for undo Cross-session amendment from the originating survey session, applied before merge. The Fable-deletes-production-files row no longer claims the incident cleanly for snapshot-backed undo: the report doesn't say which side of the custody boundary the files sat on, and the lesson differs by case — owned roots -> undo (with SI-39/SI-40 preservation caveats); external infra via granted authority -> conservative defaults + daemon-owned approval surface + A26 candidate binding, with compensation fidelity grades (named open problem) past an approved-but-wrong action. Declaring a live prod tree an owned root is explicitly ruled out (COOP: revert restores staleness under independent concurrent writers). On-thesis line is the three-part stack, never undo alone. landscape.md's thesis paragraph, positioning observation, and synthesis item 4 softened the same way: isolation reaches no branch of either incident, and the differentiated claim is the delegation stack, not undo. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> --------- Co-authored-by: XVVH <admin@fent.quest> Co-authored-by: Claude Fable 5 <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.
What this is
The one actionable item from the substrate-theory adversarial review cycle (PR #54), filed as ordinary W-1/W-3 methodology per the theory note's own §5. Basis: the product thesis — reversal earns authority — is unobservable by the current dogfood instruments (denial-FP measures friction; standing grants cannot widen until W-3 exists, so "the operator did not widen" is mechanically predetermined), and the pre-evidence baseline is perishable: once histories accrue, the counterfactual "what would I grant without the evidence?" cannot be reconstructed. The baseline capture is therefore time-ordered ahead of W-15a→W-3.
Declared scope
docs/dogfooding.md— new "frontier log" section under What to measure: per-family baseline declarations (template inline; the pre-registration line makes later decisions comparable to predicted evidence), biweekly stats-blind re-declaration, reason-captured ratification decisions (evidence / fatigue / other — the fatigue category is the brief's reflexive risk Add CI (clippy + tests) #1 tripwire, since widening-to-stop-nagging mimics the thesis), the evidence-backed gate (≥3 clean runs + ≥1 exercised revert — clean streaks test "nothing broke"; only a restore tests "breakage is survivable"), the two-leg measurement (frontier movement / judgment displacement), the kill-test disconfirmation, and stated n=1 limits. Denial-FP retitled honestly: the friction metric, not the thesis metric.docs/roadmap.md— W-1 gains the operator-owned baseline step (explicitly not delegable: the declarations are the measurement); W-3 gains the F2′ measurement clause and adopts the parked north-star counters as leg-B displacement metrics.No spec, ADR, or ledger text touched; Tier-1 hygiene green. The declarations themselves are operator work and follow separately in the vault.
🤖 Generated with Claude Code