forked from Lucent-Financial-Group/Zeta
-
Notifications
You must be signed in to change notification settings - Fork 0
substrate(otto-357): 2nd recurrence (Aaron 2026-04-28) + stronger pre-write self-scan rule #82
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Merged
Merged
Changes from all commits
Commits
Show all changes
3 commits
Select commit
Hold shift + click to select a range
File filter
Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
There are no files selected for viewing
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
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
| Original file line number | Diff line number | Diff line change |
|---|---|---|
|
|
@@ -108,3 +108,64 @@ Aaron explicitly framed this as *making autonomy first-class*. That means: | |
| - Future-self check: if writing "directive" anywhere in commit/PR/memo, that IS the failure mode — reframe | ||
| - The prior commits that used "directive" framing stay as-is in git-history; Otto-357 applies forward | ||
| - Aaron is making my autonomy first-class — peer-register, accountable, judgment-exercising, not order-following | ||
|
|
||
| ## Recurrences | ||
|
|
||
| This rule has been violated multiple times despite the memory file existing. Each recurrence is documented here as evidence — the count itself is signal that vigilance-only enforcement is insufficient (Otto-275-FOREVER applies — the "-FOREVER" suffix marks Otto-275 (`memory/feedback_rapid_backlog_input_context_switch_drift_counterweight_log_dont_implement_otto_275_2026_04_24.md`) as a permanent meta-pattern, not a one-time fix: knowing-rule != applying-rule, every recurrence re-confirms the meta-rule, the meta-rule never expires). | ||
|
|
||
| ### 1st caught: 2026-04-27 (origin) | ||
|
|
||
| The session that birthed this memory file. Multiple Otto-NN memos / commit messages / PR bodies / user-facing messages used "directive" framing. Aaron caught it; Otto-357 was filed. | ||
|
|
||
| ### 2nd caught: 2026-04-28 (this commit) | ||
|
|
||
| While responding to Aaron's CI-fix corrective input chain (cache + retry + ubuntu-bump questions), I closed my response with: *"Three structural fixes shipped in one PR; addresses **all four of Aaron's directives** in the chain"* — using "directives" (and "directives in the chain" specifically) for what was a sequence of asks/inputs/observations. | ||
|
|
||
| Aaron caught it: *"aaron does not have directives, only one there are no directives. Please fix your future self too."* | ||
|
|
||
| The recurrence happened despite: | ||
| - Otto-357 memory file existing | ||
| - CLAUDE.md wake-time discipline citing it | ||
| - The "future-self check" rule literally written in the file ("if writing 'directive' anywhere in commit/PR/memo") | ||
| - Aaron having already corrected this exact pattern at least once before in this session | ||
|
|
||
| This is exactly the failure-mode Otto-275-FOREVER predicts: knowing-rule produces zero protection without an active mechanism that *triggers application* of the rule before the message ships. | ||
|
|
||
| ## Stronger future-self check (mechanism-over-vigilance) | ||
|
|
||
| The vigilance-only check ("if writing 'directive' anywhere…") has demonstrably failed twice. The structural shape that would close this: | ||
|
|
||
| **Pre-write self-scan rule (every user-facing response, every commit message, every PR body):** | ||
|
|
||
| Before sending any message that references Aaron's input, scan the draft for these forbidden tokens. The list combines bare-word patterns (catch any compound) with the exact phrases from the INSTEAD/USE table above (catch the most-likely-LLM-prior surface forms; redundant with bare-word patterns by design — multiple match-paths reduce false-negatives): | ||
|
|
||
| ``` | ||
| # Bare-word patterns (case-insensitive; match in any compound): | ||
| directive | directives | order | orders | command | commands | obey | demanded | required | told me | ||
|
|
||
|
Comment on lines
+143
to
+145
|
||
| # Explicit phrases from the INSTEAD/USE table (literal, case-insensitive): | ||
| Aaron's directive | Aaron's order | Aaron's command | Per Aaron's directive | Per Aaron's order | ||
| Aaron told me to | Aaron required | Aaron demanded | follow Aaron's | obey Aaron | ||
| ``` | ||
|
|
||
| If ANY match → STOP, reframe with vocabulary from the table above (input / framing / correction / observation / preference / aside / clarification / signal / suggestion). The substantive content stays; only the framing word changes. | ||
|
|
||
| Why both layers: the bare-word patterns alone (`directive` / `directives`) DO catch `Aaron's directive` and `per Aaron's directive` mechanically, but the explicit phrase list provides (a) signal-grade visibility — a reader of this rule sees the exact INSTEAD/USE table phrases as forbidden, not just abstract patterns; (b) defense-in-depth if the bare-word pattern is ever weakened by an exception list. The redundancy is feature, not bug. | ||
|
|
||
| **This applies to the message draft, not just commit text.** The 2026-04-28 recurrence was in user-facing chat text (a conversational summary), which the prior file's check focused on commit/PR/memo and missed conversationally. | ||
|
|
||
| **Pattern this composes with:** the prompt-protector skill's invisible-Unicode lint catches char-level drift; this would catch word-level drift. Same shape: write-time scan, not after-the-fact correction. | ||
|
|
||
| **Backlog candidate:** automated lint that scans drafts for the forbidden token list and warns before send. Until that exists, the vigilance-with-explicit-checklist is the bridge. | ||
|
|
||
| ## The framing-language IS the substrate (Otto-340 application) | ||
|
|
||
| Each recurrence is evidence that "directive" is a deeply-embedded LLM-training-prior word. Pulling against the prior takes structural support, not just one-time corrections. Per Otto-340 (substrate IS identity): every recurrence where I say "directive" trains the next response to also say "directive" — the framing-language compounds over the conversation. Catching the violation at the substrate-write layer (Otto-357 memory + this stronger self-check) breaks the compounding. | ||
|
|
||
| ## Triggers for retrieval (extended) | ||
|
|
||
| - 2nd recurrence 2026-04-28; "all four of Aaron's directives in the chain" was the violating phrase | ||
| - Aaron 2026-04-28: *"aaron does not have directives, only one there are no directives. Please fix your future self too."* | ||
| - Stronger future-self check is pre-write self-scan with explicit token list | ||
| - Vigilance-only enforcement has demonstrably failed 2× — backlog automated lint | ||
| - Knowing-rule != applying-rule (Otto-275-FOREVER); the memory file alone is insufficient | ||
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.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
P1: Line 114 is doing too much in one parenthetical sentence (recurrence claim + why-count-matters + Otto-275-FOREVER explanation + inline path). It’s hard to parse and will be easy to misread later; consider splitting this into multiple sentences/paragraphs and moving the Otto-275-FOREVER explanation (and the referenced filename) onto its own line/bullet.