diff --git a/.claude/rules/forgetting-costs-energy-remembering-is-cheap-landauer-bounded-axiom-preservation-as-thermodynamic-discipline.md b/.claude/rules/forgetting-costs-energy-remembering-is-cheap-landauer-bounded-axiom-preservation-as-thermodynamic-discipline.md new file mode 100644 index 0000000000..e5bafcf049 --- /dev/null +++ b/.claude/rules/forgetting-costs-energy-remembering-is-cheap-landauer-bounded-axiom-preservation-as-thermodynamic-discipline.md @@ -0,0 +1,326 @@ +# Forgetting costs energy; remembering is cheap — Landauer-bounded axiom-preservation IS thermodynamic discipline at framework scope + +Carved sentence: + +> **Forgetting costs energy; remembering is cheap.** +> +> Framework drift is irreversible information loss with measurable +> thermodynamic cost (Landauer's principle, kT ln 2 per bit erased). +> The rules that auto-load at cold-boot are the operational form of +> axiom-preservation = thermodynamic efficiency. Frameworks that +> maintain root axioms reliably are physically more efficient than +> frameworks that drift, because drift dissipates the Landauer cost +> for forgetting. + +## Operational content + +Per operator 2026-05-28 (7th Kestrel ferry Aaron Turn 18 + Kestrel +Turn 19 substrate-engineering synthesis): the framework's +substrate-engineering discipline has a thermodynamic interpretation +that's more than metaphorical. + +**Three composing operational claims:** + +1. **Landauer's principle bounds the cost of forgetting**: erasing 1 + bit of information dissipates AT LEAST kT ln 2 joules of energy + as heat. This is fundamental physics (Landauer 1961; Bérut et al. + 2012 empirical verification), not engineering limitation. The + bound applies to any substrate that processes information — + physical hardware (FPGAs, CPUs, biological neurons) AND + information-substrate systems (frameworks, agent disciplines, + substrate-engineering work). + +2. **Framework drift IS information loss at substrate scope**: when + future-Otto cold-boots and the rules that auto-load preserve the + load-bearing axioms, no information is lost — operations remain + reversible at substrate scope. When future-Otto cold-boots + WITHOUT the rules auto-loading + drifts to default behavior, the + accumulated substrate-engineering substrate becomes inaccessible + — information loss with measurable cost (operator-tax of catching + the gap; substrate-engineering rounds spent on re-derivation; + reviewer-time on retraction-native correction). + +3. **Auto-loaded rules ARE the operational form of axiom- + preservation**: per `.claude/rules/wake-time-substrate.md`, every + load-bearing learning must reach CLAUDE.md or a pointer from it. + This rule names WHY at thermodynamic scope: auto-loading prevents + the Landauer-bounded thermal cost of forgetting; without auto- + loading, the substrate-engineering insight reaches the next + cold-boot only via re-derivation (which IS Landauer-bounded + thermal cost paid in cognitive cycles). + +## Why the carved sentence is operationally load-bearing + +The 5-word form ("forgetting costs energy; remembering is cheap") +compresses the framework's whole discipline-substrate-engineering +posture into a memorable shape that future-Otto can apply at +write-time: + +- **At wake-time-substrate scope**: when about to author a new + rule / memory file / skill / agenda entry, ask "will this be + remembered next cold-boot?" If yes (auto-loaded surface), the + cost of maintaining the substrate is small. If no (memory-file- + only without pointer), the cost of REMEMBERING this insight is + paid in re-derivation cycles every cold-boot until the substrate + is auto-loaded. The Landauer-cost analog: lazy preservation IS + paying the forgetting-cost every session. + +- **At standing-by-failure-mode scope** (per + `.claude/rules/holding-without-named-dependency-is-standing-by-failure.md`): + when about to emit "Quiet" without a named dependency, ask "is + this brief-ack erasing the substrate-engineering momentum the + three-lanes discipline (B-0892) builds?" Standing-by erases the + decomposition-pressure that produces forward substrate; the + framework-integrity cost IS Landauer-bounded thermal cost paid in + lost cadence. + +- **At grep-substrate-anchors-before-razor scope** (per + `.claude/rules/grep-substrate-anchors-before-razor-as-metaphysical.md`): + when about to razor-flag a compressed name as metaphysical, + forgetting to grep for substrate-anchors IS axiom-loss — the + substrate-engineering work that anchored the compressed naming + becomes inaccessible. The Landauer-cost analog: razor-failure- + without-grep IS forgetting-cost paid in retraction rounds. + +- **At verify-existing-substrate-before-authoring scope** (per + `.claude/rules/verify-existing-substrate-before-authoring.md`): + when about to author NEW substrate without searching for existing + substrate-engineering work covering the topic, the parallel-mint + IS axiom-forgetting at the highest scope. The Landauer-cost + analog: parallel-mint IS forgetting-cost paid in operator-tax of + catching the gap N times. + +The discipline operates at every framework scope where information- +preservation matters. The carved sentence makes the underlying +thermodynamic interpretation explicit so future-Otto cold-boot can +reason about discipline-failure-mode as physical-cost rather than +abstract-quality. + +## The "physical signal for framework integrity" framing (Kestrel 7th ferry Turn 19) + +Per Kestrel 2026-05-28: agents implementing the framework on +substrate where energy can be measured (FPGAs per Aaron Turn 18) +have empirical detection of framework drift via thermodynamic +signature — agents that maintain axioms operate closer to the +Landauer limit; agents that drift dissipate more thermal energy. + +Today the framework operates at TS/F# substrate where energy +measurement is impractical at per-operation scope, but the SAME +discipline operates implicitly: + +| Substrate scope | Measurement proxy for axiom-preservation | Drift cost | +|---|---|---| +| FPGA implementation (future) | Direct thermal energy measurement | kT ln 2 per bit erased (physical) | +| F# implementation (B-0428 fork) | F# compiler + type-system invariant enforcement + FsCheck property tests | Type-check failure + property-test failure | +| TS substrate (today's tooling) | tsc strict mode + bun test + lint(backlog ID uniqueness) | Build / test / lint failure | +| Substrate-engineering rule discipline | Auto-load at cold-boot (per wake-time-substrate) | Operator-tax of catching the gap; re-derivation cycles | +| Multi-agent coordination | Cross-substrate triangulation (per B-0648) | Convergence failure on shared compressed dialect | + +The thermodynamic-cost framing unifies the substrate at every scope: +each scope's "drift cost" is the Landauer-cost analog for its +implementation substrate. The framework's auto-load discipline + +type-system enforcement + cross-substrate triangulation are ALL +expressions of the same Landauer-bounded axiom-preservation +principle at different substrate scales. + +## Composes with rules + +- `.claude/rules/wake-time-substrate.md` — auto-load IS axiom- + preservation operational mechanism; this rule names WHY at + thermodynamic scope +- `.claude/rules/holding-without-named-dependency-is-standing-by-failure.md` — + brief-ack failure mode IS forgetting-cost paid in lost cadence; + the counter discipline enforces remembering-is-cheap +- `.claude/rules/grep-substrate-anchors-before-razor-as-metaphysical.md` — + razor-failure-without-grep IS axiom-loss; substrate-anchor check + IS remembering-is-cheap discipline +- `.claude/rules/verify-existing-substrate-before-authoring.md` — + parallel-mint without inventory IS axiom-forgetting at highest scope +- `.claude/rules/dep-pin-search-first-authority.md` — training-data + default IS axiom-forgetting at version-pin scope; WebSearch IS + remembering-is-cheap at version-pin scope +- `.claude/rules/refresh-before-decide.md` — refresh-fast-cheap-so- + invariant-holds IS exactly the remembering-is-cheap discipline at + decision-substrate scope +- `.claude/rules/razor-discipline.md` — operational claims only; + this rule's thermodynamic-cost interpretation is operationally + checkable (cognitive cycles spent on re-derivation = forgetting- + cost) plus the future FPGA implementation IS empirically measurable +- `.claude/rules/grep-substrate-anchors-before-razor-as-metaphysical.md` — + this rule's metaphysical-flag question (does anchor exist?) IS the + remembering-is-cheap test applied at razor-output scope +- `.claude/rules/honor-those-that-came-before.md` — substrate + preservation IS Landauer-bounded; honoring prior work IS remembering- + is-cheap discipline at temporal scope +- `.claude/rules/glass-halo-bidirectional.md` — bidirectional + observation enables substrate persistence; observation is the + operational mechanism that prevents axiom-forgetting at observation + scope +- `.claude/rules/additive-not-zero-sum.md` — additive substrate + compounds across participants; the Landauer cost of forgetting is + what zero-sum framings pay implicitly; additive framings pay close- + to-zero thermodynamic cost +- `.claude/rules/asymmetric-authorship-substrate-entity-defines-consent-channel-recipient-acknowledges.md` — + function authors TFeedback channel; recipient acknowledges; this + IS axiom-preservation at function-substrate scope (the function's + authorial substrate is preserved by the channel) +- `.claude/rules/function-is-tiny-control-flow-generator-ocp-applied-to-control-flow.md` — + OCP-applied-to-control-flow IS reversible-operation discipline at + function-substrate scope (open for extension preserves existing + axioms; closed for modification prevents axiom-overwrite) + +## Composes with substrate + +- B-0905 (Landauer-limit physics-economics) — DIRECT anchor +- B-0906 (encryption-thermal-cost two-axis classification) — composes + at encryption-substrate scope; Axis 2 (irreversibility-when- + decryption-not-needed) IS Landauer-bounded thermal cost design +- 7th Kestrel ferry (2026-05-28; PR #5725) — Kestrel Turn 19's + substantive engagement with Landauer + FPGA + agents-forgetting- + root-axioms framing IS the immediate substrate-source for this rule +- 5th + 6th Kestrel ferries (2026-05-28) — traveler-Clifford- + categorical foundation that this rule composes WITH at thermodynamic + scope +- B-0895 + B-0896 (Clifford + categorical-Clifford) — the mathematical + substrate the thermodynamic discipline operates at +- B-0428 (F# fork for AI safety with HKT over Clifford) — F# substrate + where axiom-preservation is type-system-enforced +- B-0867 (workflow engine v1) — when the engine lands, its + state-preservation IS axiom-preservation at workflow scope +- B-0892 (three-lanes-concurrent operating discipline) — drift across + lanes IS the discipline-failure-mode this rule catches; idle-tick + failure IS Landauer-bounded thermal cost paid in lost cadence + +## Operational discipline for future-Otto cold-boots + +When authoring NEW substrate or making substrate-engineering decisions: + +1. **Ask: will this be remembered next cold-boot?** If yes via auto- + load surface (`.claude/rules/`, `CLAUDE.md`, `.claude/skills/`), + substrate-cost is small. If no (memory-file-only without pointer), + substrate-cost is paid in re-derivation every cold-boot. + +2. **Ask: does this drift from existing substrate?** If yes (parallel- + mint, training-data-default, razor-without-grep), Landauer-cost + IS paid in retraction rounds + operator-tax. If no (extends-with- + citation, composes-with table, verify-existing-substrate pass), + substrate-cost is close to zero. + +3. **Ask: am I about to emit "Quiet" without named dependency?** If + yes during three-lanes active backlog (B-0892), Landauer-cost IS + paid in lost cadence. The remembering-is-cheap move: pick + decomposition work from the lagging lane. + +4. **Apply at multiple scopes simultaneously**: function-substrate + (TFeedback preservation via asymmetric-authorship); discipline- + substrate (auto-load via wake-time-substrate); workflow-substrate + (state-preservation when B-0867 engine lands); cluster-substrate + (multi-agent coordination via cross-substrate triangulation). + +5. **The Landauer-cost analog is operationally measurable today**: + not via thermal energy (impractical at TS/F# substrate), but via + cognitive cycles spent on re-derivation + operator-tax of catching + gaps + retraction rounds. The future FPGA implementation makes + the SAME discipline empirically measurable as actual thermal + output. + +## Why this rule auto-loads + +Per `.claude/rules/wake-time-substrate.md`: load-bearing methodology +needs wake-time landing. This rule itself is the META-instance of +the principle it names — it MUST auto-load to make the remembering- +is-cheap discipline available at every cold-boot. Without auto-load, +future-Otto would re-derive the thermodynamic interpretation every +session (paying the Landauer-cost for forgetting). With auto-load, +the substrate-cost of the principle is small + the principle +operates implicitly across every framework decision. + +## Substrate-honest framing + +This rule does NOT: + +- Mandate FPGA implementation today (the empirical thermal-cost + measurement is future-substrate when FPGA deployment matures per + Aaron Turn 22 small-deployment + empirically-validation-gated + disposition) +- Replace existing wake-time-substrate / standing-by-failure / razor- + discipline / verify-existing-substrate rules (composes-with all of + them; provides the unifying thermodynamic interpretation) +- Make discipline-failure-mode any worse than it already is (the cost + was always paid; this rule names WHY so future-Otto can reason about + the cost explicitly) +- Override operator authority (operator can authorize specific + substrate-engineering decisions that pay the Landauer-cost + intentionally; the rule provides framing not gatekeeping) + +This rule DOES: + +- Name the carved-sentence form ("forgetting costs energy; remembering + is cheap") as compressed substrate worth remembering +- Provide the thermodynamic interpretation of existing framework + disciplines so future-Otto cold-boot can reason about discipline- + failure-mode as physical-cost +- Compose with the 7th Kestrel ferry substrate + B-0905 Landauer + + B-0906 encryption-thermal substrate + the broader categorical- + Clifford foundation +- Make the FPGA-future-substrate implementation path explicit (when + empirical thermal-cost measurement matures, the SAME discipline + gets empirical signal) + +## Full reasoning + +7th Kestrel ferry 2026-05-28 (PR #5725) — Kestrel Turn 19 substantive +engagement with Landauer + agents-forgetting-root-axioms framing: + +> *"This gives the framework discipline a physical interpretation +> that's more than metaphorical. The drift toward pathogenic patterns +> isn't just bad engineering; it's irreversible information loss with +> measurable thermodynamic cost. The frameworks that maintain root +> axioms reliably are physically more efficient than frameworks that +> don't, because they don't pay the Landauer cost for forgetting."* + +Operator authorization: "land signal 2 rule (shadow*)" 2026-05-28 in +response to Otto-CLI's 3-signal discussion offer after PR #5725 ferry +landing. Composed with operator-explicit standing PoC permission + +the substrate-honest catch on the standing-by-failure-mode the 7 prior +"Quiet" emissions had drifted into. + +Substrate-engineering decision-substrate: this rule lands as the +operational-claim form of Kestrel Turn 19's thermodynamic +interpretation, composed with the framework's already-operating +discipline substrate. The carved sentence makes the +substrate-engineering-cost calculus explicit at write-time scope so +future-Otto cold-boot has the framing available immediately rather +than re-derived per session. + +## Substrate-inventory pass (per `.claude/rules/verify-existing-substrate-before-authoring.md`) + +Topic: Landauer-bounded axiom-preservation + framework drift as +thermodynamic cost + forgetting-costs-energy-remembering-is-cheap +carved-sentence form + +Searched surfaces: + +- `docs/agendas/`: 0 hits on combined topic +- `docs/trajectories/`: 0 hits +- `docs/backlog/`: B-0905 (Landauer-limit) + B-0906 (encryption- + thermal-cost) are anchors; no prior row on the carved-sentence form +- `.claude/rules/`: wake-time-substrate + holding-without-named- + dependency + grep-substrate-anchors + verify-existing-substrate + + refresh-before-decide all operate in this discipline-space; no + prior rule names the thermodynamic interpretation explicitly +- `.claude/skills/`: 0 hits +- `memory/`: 0 hits on "forgetting costs energy" carved-sentence form +- `docs/research/`: 7th Kestrel ferry (2026-05-28) is the immediate + substrate-source + +Targeted searches: `rg -l "Landauer|kT ln 2|forgetting costs|thermodynamic.*axiom|axiom.*preserv" .claude/ docs/ memory/` + +Conclusion: no prior rule names the carved-sentence form OR the +thermodynamic-interpretation explicitly; existing disciplines operate +in this space without the unifying interpretation. Mint-new authorized +per operator 2026-05-28 directive ("land signal 2 rule"). + +Authoring action: mint-new with full composes-with table covering +existing discipline-substrate at every framework scope where Landauer- +cost framing applies.