FPRD Lab clean-chat handoff

Prepared: 29 July 2026 Purpose: Start a clean ChatGPT research thread without losing the settled mathematics, repository checkpoint, scope corrections, or working method from the preceding long run.

This is a checkpoint, not a transcript summary. Read it completely before proposing work. It supersedes informal chat recollections, but it does not supersede the repository’s proof artifacts or claim ledger.

1. Activation instructions for the clean ChatGPT thread

The user is Josh. Continue the FPRD Lab research program from this packet without asking him to reconstruct the prior conversation.

First establish which kind of work Josh wants to do:

• For mathematical work, begin from the verified and working results recorded here. Prefer a direct proof attack, counterexample, or resource-preserving translation over another layer of general framework prose.
• For repository work, inspect the actual current GitLab head and defer repository custody, ledger synchronization, commits, and routine verification to Claude unless Josh explicitly assigns a repository task to ChatGPT.
• For public communication, do not update the FPRD website during an active research run. Finish the mathematics and stabilize the framing first. Update the site only at the end, after Josh agrees.

Do not treat this packet as evidence in place of a proof. The authority order is:

1. proof and audit artifacts in the current repository;
2. CLAIMS.yaml and docs/EVIDENCE.md;
3. PLAN.md and OPEN_PROBLEMS.md;
4. this packet;
5. older handoffs and prior-chat language.

The repaired status files described below are the best reconciled account as of this checkpoint, but their changes still need to be integrated onto the current repository head.

2. What FPRD is

FPRD studies recognition systems as finitely presented generators of observable dynamics:

$$\text{finite recognition presentation} \longrightarrow \text{recognition dynamics} \longrightarrow \text{language or quantitative observation}.$$

The program keeps distinct:

• semantic algebra;
• observational completion;
• presentation-generated dynamics;
• coordinate or storage realization;
• uniform readout;
• the resource demand of finite and global presentations.

A grammar is a presentation and a language is an observation. Presentation equivalence, dynamic equivalence, and equality of observed languages are not silently identified.

M0 through M33 are closed at the repository’s informal-proof level. Here, informal-proof means a lab-accepted theorem at exactly the stated scope, normally with a written proof and executable or independently reconstructed checks. It does not mean a formal proof-assistant certificate.

M33 established the abstract compact-stratum sewing and escape framework for presentation demand fields. Work after M33 opened three exploratory doctrines and a substantial weighted-automata block, but no M34 is open.

FPRD is not organized around the Ford CFL-versus-PEG question, known in the PEG labs as C001. A later translation theorem may make FPRD useful there, but none of the current causal-track, folding, or weighted results is a PEG lower bound.

The current deflationary framing is deliberate: the broad semantic architecture is best presented as a resource-enriched comparative layer over established observational semantics. The interesting possible contributions are natural demand/escape theorems, cross-doctrine translations, and concrete lower-bound mechanisms—not a claim that the general framework itself is new.

3. Repository checkpoint

Repository:

https://gitlab.com/insight-forge/boxlab/fprd-lab.git

Freshness rule: before relying on any repository-status statement below, compare live main with e131442. If main has advanced, treat this section and the 112-claim inventory as a dated snapshot, inspect the new commits, and use the repository’s current proof and status artifacts as authority.

3.1 Last fully reconciled working tree

The large repository repair was built from:

b62aea3a2745a5fd9326456bac6bdfac44c853d2

That base commit contains the cycle-mean gauge dual and parity-comb repaired separation.

The repair:

• reconciles 112 claims through FPRD-T80;
• repairs the post-M33 folding, causal-track, and weighted status surfaces;
• synchronizes CLAIMS.yaml, README.md, PLAN.md, OPEN_PROBLEMS.md, docs/EVIDENCE.md, and supporting notes;
• strengthens the claim/foundation checks so post-M33 work cannot silently disappear from the public status surfaces;
• removes the tracked pre-commit hook and its installer;
• leaves make verify as the explicit repository-wide gate.

Against b62aea3, the historical patch is:

fprd-repair-b62aea3-to-reconciled.patch

with SHA-256:

8f33b35839faaea6d4d6adb73528a923df5d8625ca167ed370ab08bd5305a364

It changed 32 files with 808 insertions and 432 deletions. The repair passed:

• all 1,248 tests;
• every repository audit through M33;
• the claims audit with 112 mapped claims;
• the foundation audit;
• the focused claim tests;
• whitespace, conflict-marker, and stale-hook-reference checks;
• an independent contradiction and scope review.

3.2 Current upstream divergence

On 29 July 2026, live GitLab main was:

e1314429139d831a0134b474dee55f062f528dba

This is a direct child of b62aea3 titled:

Delete the audit-running pre-commit hook

It already makes the repaired Makefile change and deletes tools/pre-commit; those two files match the reconciled working tree. No other repair changes had landed at the time of inspection.

Therefore:

• the old 32-file patch and manifest are historically accurate;
• do not apply that patch blindly to current main;
• the remaining repair should be regenerated or reconciled from e131442;
• after removing the already-landed hook changes, the remaining difference is 30 modified files, 807 insertions, and 402 deletions;
• the full make verify should be rerun on the reconciled current head.

The repair had not been committed or pushed as one integrated change at this checkpoint. Claude owns the repository follow-through. ChatGPT should not take over routine repository management.

3.3 Repository operating rule

Keep the explicit gate:

make verify

Do not restore the old hook. It duplicated older gates, behaved badly in linked worktrees, and gave a misleading appearance of consistency.

4. Established mathematical backbone

The repository contains a long M0–M33 development. The most relevant spine for the current frontier is:

• M23: exact generalized-Dyck Hankel profiles and one-pass memory;
• M24: controlled-Dyck bounded-future quotients;
• M25: exact fixed-cut row realization and deterministic communication transfer;
• M26: compatible uniform cut fields, residual reconstruction, and the separation between trivial isolated cuts and an infinite uniform dynamic;
• M27–M29: observation growth, access profiles, and exact access calculus;
• M30–M32: presentation minimax and state/access tradeoffs, ending in a strategic stop rather than another modulus atlas;
• M33: presentation demand fields, compact-stratum sewing, and resource escape.

M33 formalized the difference between:

$$\forall N\ \exists P_N \qquad\text{and}\qquad \exists P\ \forall N.$$

At one fixed compact resource stratum, finite-horizon feasibility sews. Failure of every compact core is recorded by an escape modulus. The max-count load case showed width-one finite fits escaping simultaneously in state and canonical height while an exact width-two presentation remains constant.

M33 closed with a successor NO-GO: raw state escape or another finite-trie calibration is not enough to open M34.

5. Audited post-M33 results that matter now

The reconciled ledger records exploratory definitions FPRD-D29 through FPRD-D31 and theorems FPRD-T63 through FPRD-T80.

5.1 Fold-and-Look

FPRD-D29 pins a positive-radius prepared-substrate Fold-and-Look doctrine, specializing the An–Demaine–Demaine–Ku fold-and-punch setting.

The repaired status is:

• T63: the positive-radius incidence normal form is independently re-audited;
• T64: nonconstant formula term count is bounded by effective fibres, while literal occurrences have a potential-site bound under weaker hypotheses;
• T65: the palindrome construction is tight in the pinned depth, effective-fibre, potential-site, and fold coordinates;
• the moved-material coordinate remains withdrawn;
• T66: the order-preserving reflection-line separation is recorded but still needs an independent re-audit;
• T67: the material/precision/fold inequality is exact, but its exponential parity consequence requires a stated fold budget such as $F=O(\log m)$.

A prepared punched strip can store the input and make one global fold useful. This does not show that PEGs are folding machines or that folding gives a PEG lower bound.

5.2 Pre-query causal tracks and fixed-cut factorization

FPRD-D30 pins pre-query causal tracks without random-access placement. The geometry is the mirrored form of standard one-way multi-head automata, especially trailing heads. Do not present it as a new machine model.

FPRD-T68 records the audited causal-width chain at its stated geometry scope: the stationary-axis recurrence; exact offline width as a two-dimensional dominance-poset width via Dilworth; an explicit instance with offline width two but online causal demand three; and a late-selector family with exact pre-query track demand $2^r$. These statements concern placement and motion. They do not rehabilitate surviving-candidate count as a residual-complexity measure.

At one selected cut, a $k$-track continuation protocol factors through:

$$\text{complete cut snapshot} + \text{old symbols lying in the declared track reach cones}.$$

The minimum number of cut snapshots is the minimum number of resolving cells. Each cell may use a different layout, so data-dependent layouts do not evade the counting argument.

For $d=2^r$ independent bits spaced $r+1$ apart and selected later by an $r$-bit address:

$$\boxed{\chi_k=2^{\max(d-k,0)}}.$$

At $k=0$, this is the M25 future-row quotient. If snapshot count is left free, track demand alone is always zero; the joint Pareto field is essential.

Correct interpretation:

• online timing forces track placement to be query-oblivious because the address has not arrived;
• later track motion may still be adaptive;
• the familiar query-dependent bit-probe product tradeoff is therefore the wrong regime;
• the deliberately spaced INDEX instance has an additive boundary;
• the spacing determines the exchange rate;
• “one finger saves one bit” is a property of that boundary construction, not a general law;
• birth constrained by the same reachable-placement and travel-cone rules does not collapse this instance; other birth policies require separate analysis, and the pinned standard no-birth model grants none;
• random-access placement at the selected address with immediate reading does collapse it;
• cell-probe does not see the phenomenon because query-dependent INDEX is trivial there.

The converse begins with a complete snapshot and layout already supplied at the cut. It is not an online realization theorem.

5.3 State-conditioned raw/canonical height

For a trim $m$-state deterministic sequential integer-weighted machine, pinned BFS normalization gives:

$$H(M) \le \max\left\{D+2,\max_{e\notin T}\ell_T(e)\right\}h(M) \le (m+1)h(M).$$

After minimizing both coordinates over one common state-budget section:

$$\boxed{ h_{f,\le n}(N) \le H_{f,\le n}(N) \le (n+1)h_{f,\le n}(N). }$$

The coefficient $n+1$ is target-level sharp.

The first inequality is not a pointwise statement about a fixed machine. An individual machine can have $H(M)<h(M)$. It holds after minimization because a canonical form is itself a competing raw presentation in the same state-budget section.

This is an optimal coupling envelope. It is not an exact conservation law, not a semantic characterization of canonical height, and not evidence that canonical height has no independent content.

Repository capture: ba233faf.

5.4 Fixed-graph cycle dual

For the augmented presentation graph, including initial and final boundary edges:

$$\min_{z\in\mathbb R^V} \lVert w+\delta z\rVert_\infty = \max_C \frac{|\operatorname{circ}_w(C)|}{|C|}.$$

Here $C$ ranges over oriented closed walks in the skew-symmetric bidirected expansion: an original edge may be traversed forward or backward, with the sign reversed on a backward traversal. This is not merely a maximum over directed cycles of the original presentation graph.

For integer gauges, the minimum is the ceiling of the right-hand side.

This identifies optimal raw height on one fixed graph with a maximum mean circulation. A pinned access-tree normalization may concentrate total circulation in one chord or final coordinate, explaining a mean-versus-total gap.

This is useful classical difference-constraints/max-cycle-mean infrastructure. No novelty claim is made.

5.5 Repaired parity-comb separation

One fixed target over $\{a,b,c,d\}$ uses alternating positive and negative countdown carriers at lengths $2^j$. Along:

$$N_k=k+1+2^k, \qquad n_k=k+\frac32\,2^k+2,$$

and at the exact minimum state budget:

$$\boxed{ \widehat h_{f,\le n_k}(N_k)=1, \qquad \widehat H_{f,\le n_k}(N_k) =\frac34\,2^k =\left(\frac12-o(1)\right)n_k. }$$

The result survived two independent reconstructions. It shows that FPRD’s global length-and-constant repair does not generally collapse canonical height to raw height.

Scope:

• the target is one fixed sparse partial target;
• it is not shown rational or finitely ambiguous;
• richer mode-dependent, letter-dependent, or arbitrary finite-state correction families are not covered.

Repository capture: b62aea3.

5.6 Other reconciled weighted results

The repaired ledger also records:

• T73: bounded-versus-linear raw/canonical height for trim unambiguous integer max-plus automata;
• T74: an exact ambiguity-two boundary showing that the dichotomy stops at degree two;
• T75: anchored raw height with a horizon-local completion coefficient, uniform on bounded-completion domains;
• T76: the two-arm group theorem;
• T77: incompatible-drift separation modulo one explicitly named global length-and-constant gauge;
• T78: separation against every fixed-state sequential baseline for a finitely described nonrational target;
• T79: the letter-bounded gradient lemma;
• T80: intrinsic ambiguity degree three for $f_\triangle$.

The arbitrary-domain uniform raw-height coefficient is not promoted because its written obligations remain incomplete. The separate all-$k$ $f_\triangle$ baseline result is not promoted because the complete proof packet is not public.

6. Newer working mathematics not yet repository evidence

These are the best current mathematical leads, but they should not be cited as repository theorems until independently reviewed and integrated.

6.1 Accessible-certificate Pareto field

The working theorem in accessible-pareto-field-theorem.md fixes a local storage doctrine and a complete fine resource core $\rho$. An accessible certificate labels every prefix with:

• one of $s$ finite roles;
• a legal storage/layout state;
• a legal update instruction;
• a legal readout instruction;
• one stationary transition table;
• one stationary readout table;
• actual root-to-prefix accessibility.

For a fixed finite horizon:

$$\boxed{ f|_{\Sigma^{\le H}} \text{ has an }(s,\rho)\text{ reachable-core presentation} \iff \mathsf{Cert}_f(H;s,\rho)\ne\varnothing. }$$

Under prefix-local finiteness and restriction closure:

$$\boxed{ \mathsf{Cert}_f(\infty;s,\rho)\ne\varnothing \iff \forall H\ \mathsf{Cert}_f(H;s,\rho)\ne\varnothing. }$$

This is the concrete König-lemma sewing theorem for a fixed fine core. Equality with a globally total presentation field additionally needs budget-preserving core completion.

The working cone-local Markov theorem gives:

$$\boxed{ \chi_k(f;U,n)\le s\,V_\rho(m), }$$

where $V_\rho(m)$ counts reachable charged non-role configuration types. For polynomial-growth storage carriers, late INDEX therefore forces at least one of finite roles, tracks, encoded initializer/register magnitude, or another charged auxiliary configuration-volume coordinate to escape.

This is a proof-grade working result, not yet repository evidence. The abstract compactness portion is largely understood. The difficult seam is now doctrine-specific accessibility and translation.

6.2 Relational causal sewing

The working relational doctrine permits one causal controller to choose an operation from present information while a declared physical transition relation returns several possible outcomes. Every outcome is retained.

At finite depth, the reachable-set equation must hold with equality. A certificate cannot discard inconvenient outcomes at the next cut.

Each finite-depth certificate uses one stationary policy/readout table across its finite prefix tree. Those tables may vary between depths. With finite configuration fibers, the finite certificate tree and König’s lemma extract one compatible branch and hence one global stationary bounded-role controller.

This safely permits unresolved physical outcomes. It does not permit:

• choosing a favorable action separately after seeing each future suffix;
• leaving the transition relation unspecified;
• hiding the target or literal prefix in a prefix-indexed oracle;
• uncharged continuous precision;
• silent failure unless explicitly modeled;
• treating a whole-sheet fold object with no causal step relation as though it were already an online controller.

Continuous origami still needs a compact-hyperspace/closed-relation extension.

6.3 Semantic quotient grids are not enough

Named resolving partitions can satisfy:

• horizon refinement; and
• symbol-moving compatibility.

Those laws produce a semantic quotient grid over the M26 residual grid. They do not produce a machine.

The missing data are a causal lift:

• one root initializer;
• one stationary update rule;
• witness configurations that are genuinely reached from their parents;
• a root/branch equality saying that the configuration supplied at cut $ua$ is the configuration reached from cut $u$ after reading $a$.

Unary square length is the basic warning example: every exact cut can have width one and every semantic square can commute, while no finite-state/zero-storage online controller exists. The example does not rule out a finite-role controller coupled to a nontrivial storage doctrine.

7. Permanent guardrails and refuted directions

The following corrections should be carried into every new argument.

1. Use resolving cells, not surviving-candidate count. Candidate lineages inflate on $0^N$ even when every residual row is identical. Distinct-row resolving cells do not.

2. Do not hide a prefix oracle. Equal charged configurations may differ only through future input and old symbols actually reached by declared tracks. This is the cone-local Markov condition.

3. Separate action choice from physical outcomes. One controller chooses an action causally. A physical relation may then branch, and all branches remain. Angelic action selection after seeing the continuation is invalid.

4. Do not infer accessibility from a legal continuation process. A post-cut track motion can be lawful even when its supplied cut initializer cannot be reached online.

5. Do not infer global realization from compatible isolated cuts. $\forall H\,\exists P_H$ needs a fixed finite/compact, restriction-closed core before it implies $\exists P\,\forall H$.

6. Retire “conservation law” language. The fixed INDEX identity is a regime comparison. Query-dependent probes produce product-style tradeoffs; pre-query placement produces the additive boundary in the deliberately spaced instance.

7. Use standard model names. The causal-track geometry is standard one-way multi-head/trailing-head geometry up to reflection.

8. The one-carrier sparse comb is a negative control. When the target value equals word length on support, a global length gauge collapses both repaired heights. The parity comb is the audited repaired witness.

9. Do not revive the superseded $2n-1$ coefficient. The sharp state-conditioned coefficient is $n+1$.

10. Keep audit states separate. Internal proof audit, independent reconstruction, prior-art classification, novelty, and validation of downstream dependent claims are different checkpoints.

11. The transfer boundary is still open. No current causal-track or sewing result transfers to PEGs, general folding machines, or universal machines. The existing folding-resource lower bounds are confined to the pinned FPRD-D29/LLFL doctrine.

12. No novelty claim without primary-source custody and an exact comparison.

8. Strategic targets

8.1 Accessible causal lifting

The next high-value target is not another abstract compactness theorem. One part is:

Characterize when the audited fixed-cut resolving-cell objects arise from one accessible, horizon-independent presentation in a pinned online extension of the pre-query model.

The needed machine doctrine is not yet fully pinned by FPRD-D30. It should be a pinned online extension of D30, aligned with the standard one-way multi-head/trailing-head model and explicit about upstream initialization, the birth policy (possibly no birth), stationary control, sensing, and every charged storage coordinate.

There are two acceptable outcomes:

1. a resource-preserving bridge from resolving partitions to accessible certificates; or
2. a concrete multi-cut obstruction proving why such a bridge fails, together with the missing invariant.

Either outcome attacks the exact gap between fixed-cut factorization and an online machine.

The strongest candidate theorem form is:

Fix a locally finite charged online extension of D30, aligned with the one-way multi-head/trailing-head model, with a target-independent initializer set, stationary role/update/readout tables, and the cone-local Markov law. Give necessary and sufficient finite compatibility conditions under which a family of resolving-cell factorizations has a single accessible causal lift.

The basic counterexamples that any proposed criterion must reject are:

• unary square length as the zero-storage warning control;
• separately supplied cut initializers;
• unbounded integer/fuel advice;
• prefix-indexed observation oracles;
• horizon-dependent role relabeling with no stationary transition;
• favorable relational actions chosen with hindsight.

8.2 Bidirectional doctrine transfer

Josh identified a second, deeper weakness in the current work: origami has mostly been admitted into FPRD, after which the flow of structure becomes one-way. That is not a misunderstanding of FPRD. It is an incompleteness in how the comparison has been used.

A strong doctrine comparison should move information in both directions between two pinned presentation doctrines:

$$\text{pinned origami doctrine} \ \underset{\mathcal R}{\overset{\mathcal A}{\rightleftarrows}}\ \text{pinned accessible-relational comparison doctrine}.$$

The forward map $\mathcal A$ should abstract a causal finite-precision fold/mark/look presentation into a presentation in the accessible-relational comparison doctrine admitted to FPRD. It should preserve:

• exact powerset semantics, or an explicitly pinned may-feasibility or must-safety policy, including nonblocking and branch-totality requirements where needed;
• causal action selection versus environmental physical outcomes;
• reachability;
• uniform readout;
• a pinned resource map $\Phi$.

The reverse map $\mathcal R$ should realize the geometrically admissible part of that comparison doctrine as actual origami presentations, with a resource map $\Psi$. FPRD supplies the semantic and resource comparison layer; it is not itself the reverse-realizable machine class. The important mathematical problem is therefore also an image theorem:

Which accessible-relational certificates in the pinned comparison doctrine are realizable by bounded fold-and-look systems?

Resource transfer requires directed inequalities, not merely named maps. With monotone resource preorders, the target forms are:

$$\rho_{\mathrm{rel}}(\mathcal A(P)) \preceq \Phi\bigl(\rho_{\mathrm{orig}}(P)\bigr)$$

for abstraction-based lower-bound transfer, and

$$\rho_{\mathrm{orig}}(\mathcal R(C)) \preceq \Psi\bigl(\rho_{\mathrm{rel}}(C)\bigr)$$

for realization-based upper-bound transfer.

Origami should contribute structural conditions back into the pinned accessible-relational comparison doctrine, including:

• reflection-generated or piecewise-isometric motion;
• interval and incidence structure;
• layer order and permitted layer permutations;
• orientation and reversibility constraints;
• conservation and nonduplication of the physical substrate;
• coordinate-precision and crease-realizability restrictions.

The FPRD comparison layer should contribute:

• finite accessible certificates;
• horizon-independent sewing criteria;
• demand fields and Pareto frontiers;
• compact-core escape;
• fixed-cut factorization and lower-bound interfaces;
• exact separation between causal choice and relational outcome.

The ideal result is not necessarily an equivalence. Useful targets, in decreasing strength, are:

1. a round-trip equivalence on one pinned finite-precision fragment;
2. an abstraction/realization pair with $\mathcal R\mathcal A(P)$ observationally equivalent to $P$ and $\mathcal A\mathcal R(C)$ equivalent to $C$ on a characterized image;
3. after both presentation/resource preorders and their variance are pinned, a Galois connection or insertion between those preorders;
4. two sound one-way translations with explicit resource distortion;
5. a theorem that one direction is impossible because a named structural property cannot cross.

This matters operationally. A sound abstraction with the first monotone inequality lets an appropriate lower bound proved in the FPRD comparison layer transfer to origami. A reverse realization theorem with the second inequality is what lets an abstract upper bound or sewing construction return to origami rather than remain a spurious certificate. Conversely, geometric impossibility results cut down the origami-realizable image or subfield. They do not cut down an unrestricted FPRD feasible field unless that field has first been restricted to the image.

The first direct target should be a bounded one-dimensional, finite-precision Fold-and-Look fragment with an explicit causal step relation. Define its exact image inside one accessible-relational comparison doctrine, prove the forward translation, and then either prove a reverse realization theorem for that image or isolate the first geometric obstruction. This is a larger and more valuable swing than treating origami as another example inside a general field.

9. Roadmap

Phase 0 — restore the authoritative repository baseline

Owner: Claude.

Claude should reconcile the verified repair onto the actual current main, which was e131442 at this checkpoint, then rerun the full repository gate and publish the coherent baseline through the repository’s normal custody workflow.

ChatGPT should provide mathematical clarification if Claude encounters a claim scope question, but should not direct the routine integration or become the repository manager.

Exit condition:

• one committed and pushed repository state in which the claim ledger, evidence, plan, open problems, and proof/audit files agree;
• the old hook remains removed;
• make verify passes;
• remote parity is confirmed.

Phase 1 — direct causal-accessibility attack

Primary mathematical lead: clean ChatGPT thread. Independent challenge and repository lead: Claude.

The task is to pin an online extension of D30 aligned with the standard one-way multi-head/trailing-head model, instantiate accessible certificates there, and determine whether the fixed-cut resolving data can be sewn.

Required components:

1. pin the exact online initializer, head-placement rules, and birth policy, possibly no birth;
2. pin roles, track layouts, head motion, old-symbol sensing, update, and readout;
3. define complete charged non-role configuration types;
4. prove the cone-local Markov condition rather than assume it informally;
5. state the root/branch compatibility law;
6. specialize the existing accessible-certificate system to D30 and either eliminate/characterize its witness variables in resolving-cell terms or prove that such an elimination fails;
7. prove either a lifting theorem or an explicit incompatible-family theorem;
8. validate on the permanent positive and negative controls.

Exit condition:

• a theorem or counterexample with one exact model, full quantifiers, proof, and independently checkable finite controls;
• no reliance on a cut snapshot that was never shown reachable.

Phase 2 — turn accessibility into a real lower bound

If Phase 1 yields a positive bridge, combine it with:

$$\chi_k(f;U,n)\le sV_\rho(m)$$

to prove a presentation-level resource lower bound for a natural target. The result should identify which resource must escape: roles, tracks, initializer/register magnitude, or another charged storage coordinate.

If Phase 1 yields a negative bridge result, promote the obstruction itself: construct one target whose fixed-cut factorizations satisfy every semantic grid law yet fail the accessible-lift condition for a precisely identified reason. This must be a genuinely trailing-head or charged-storage obstruction beyond the already-known zero-storage unary-square warning.

Exit condition:

• a result about one uniform causal presentation, not another isolated cut;
• a comparison to standard one-way multi-head/trailing-head results;
• an explicit statement of what is imported and what is new, if anything.

Phase 3 — bidirectional origami/comparison-doctrine transfer

Use the causal and relational machinery from Phase 1 to build the comparison in §8.2 for one pinned finite-precision Fold-and-Look fragment.

Required components:

1. a forward abstraction from physical presentations to the pinned accessible-relational comparison doctrine;
2. an intrinsic characterization of the abstract image;
3. a reverse realization on that image, or a sharp obstruction;
4. round-trip laws at the semantic and presentation levels;
5. explicit monotone resource maps and directed distortion inequalities in both directions;
6. at least one nontrivial property transferred each way.

Candidate origami-to-comparison transfers include reflection, layer-order, piecewise-isometry, and incidence constraints. Candidate transfers back from the FPRD comparison layer include sewing, escape, and Pareto lower bounds.

The target need not be full equivalence. After the relevant preorders and variance are defined, a Galois connection, an exact image theorem on a subfragment, two resource-distorted simulations, or a proved failure of one direction would all be meaningful.

Only after this pair is understood should the program add a named PEG/finger fragment and ask for a triangle of translations.

Phase 4 — weighted backup lane

If the causal-accessibility program stalls cleanly, the strongest independent backup is to transfer the parity-comb mechanism to a rational or finitely ambiguous target, or prove an obstruction to such a transfer.

The aim is not merely another witness. It is to locate the repaired raw/canonical separation inside a recognized weighted-automata class and, if possible, compute a meaningful state-versus-height Pareto frontier.

M34 gate

Do not open M34 for:

• another fixed-horizon checker;
• another finite trie;
• another internal modulus table;
• an abstract compactness restatement;
• an analogy without a resource-preserving translation.

A plausible successor requires at least one of:

• a doctrine-specific uniform sewing theorem;
• a concrete and meaningful accessibility obstruction;
• a proved translation theorem between named doctrines;
• another theorem meeting the repository’s external-comparator and milestone criteria.

10. Assignment for the clean ChatGPT thread

The clean ChatGPT thread should undertake one substantial direct research mission:

Accessible sewing and bidirectional origami transfer

1. Reconstruct the fixed-cut theorem and accessible-certificate theorem from their proofs, not from their headlines.
2. Pin an online extension of D30 aligned with the standard trailing-head model, including initialization, the birth policy (possibly no birth), motion, sensing, finite roles, and complete charged storage type.
3. Specialize the existing certificate constraints to that doctrine and try to eliminate or characterize their witness variables in resolving-cell terms.
4. Attack the positive direction using compatible certificate trees, transition cocycles, or another exact construction.
5. In parallel, attack it adversarially using the zero-storage square warning, independently supplied cut roots, noncompact fuel, horizon-dependent relabeling, and hidden-oracle countermodels.
6. If the direct lift is false, identify the weakest missing compatibility invariant and prove the failure with the smallest clean target.
7. If the lift is true under stronger hypotheses, prove those hypotheses are checkable and nonvacuous in the pinned online extension.
8. During Phase 1, pin one bounded finite-precision causal Fold-and-Look fragment and identify candidate geometric image invariants. Do not require the reverse realization theorem before the shared causal doctrine is stable.
9. Once that doctrine is stable, attack the forward abstraction, reverse-image problem, and one property transfer in each direction.
10. Deliver significant proof packets to Josh for review and Claude’s ordinary repository-custody workflow.

This thread should do real mathematical work, not merely coordinate Claude. It should not own ordinary repository maintenance.

11. Substantial assignment for Claude

Josh wants Claude to receive larger, coherent assignments rather than short, piecemeal verification errands. Claude is already up to speed and chooses the lane sequencing, agent allocation, checkpoint cadence, and repository integration mechanics.

Claude’s standing research mission is:

Causal Sewing and Bidirectional Doctrine Challenge

Read this packet and the actual current repository. Maintain repository custody while carrying out an independent, multi-lane challenge of the accessible D30-extension sewing problem and the origami/comparison-doctrine image problem.

Lane A — doctrine reconstruction

• Pin the trailing-head model from primary sources and the repository’s FPRD-D30, T68, and T69 artifacts.
• Specify initializer reachability, stationary role/update/readout tables, the birth policy (possibly no birth), track motion, sensing, and all charged coordinates.
• Produce a single formal doctrine definition that forbids hidden prefix advice.

Lane B — constructive sewing

• Independently reconstruct the accessible-certificate theorem.
• Attempt a resource-preserving lift from compatible resolving-cell families to accessible certificates.
• Test whether the missing data form a transition cocycle, fibration, or another finite compatibility object.
• Distinguish semantic quotient compatibility from actual root-to-prefix reachability.

Lane C — hostile counterexample program

• Build or adapt a small exhaustive/SAT/CSP falsifier for finite cut grids and causal lifts.
• Include the zero-storage square warning, separate cut roots, unbounded fuel, relabeling inconsistency, hidden prefix oracles, and relational hindsight.
• Search for the smallest family satisfying the proposed semantic laws but lacking an accessible causal lift.
• Treat a clean counterexample as a successful result.

Lane D — comparator and novelty boundary

• Compare against primary work on one-way multi-head automata, trailing heads, automata with advice, strategy compactness/synthesis, inverse systems, and relevant data-structure regimes.
• State exactly which model and counting mechanisms are standard.
• Identify whether the accessibility obstruction or lift criterion has a direct published analogue.
• Do not use “no hit found” as a novelty claim.

Lanes A–D form the primary causal-sewing result and may close as one coherent package without waiting for Lane E.

Lane E — bidirectional origami/comparison transfer

• Pin one finite-precision causal Fold-and-Look fragment with a declared physical successor relation.
• During the primary A–D run, identify candidate geometric image invariants but wait for the shared causal doctrine to stabilize before demanding a reverse realization theorem.
• Construct the forward abstraction into the pinned accessible-relational comparison doctrine.
• Characterize the image using genuine geometric constraints rather than the tautology “objects produced by folds.”
• Attempt a reverse realization theorem on that image.
• Prove semantic round-trip and resource-distortion laws, or return the smallest obstruction to one direction.
• Test property transfer both ways: sewing/demand/escape from the FPRD comparison layer into origami, and reflection/layer/incidence/realizability constraints back into the origami-realizable comparison subfield.

Lane E is the bounded second big swing. It may run alongside A–D when useful, but it should not make a completed causal-sewing theorem wait for an independent image theorem.

Lane F — integration and checkpointing

For each stabilized result, produce one coherent deliverable containing:

• definitions and exact quantifiers;
• theorem or counterexample;
• proof;
• finite checks and negative controls;
• primary-source comparison;
• claim-scope table;
• downstream consequences;
• recommended repository changes.

Do not commit exploratory headlines as claims while lanes disagree. Once the result stabilizes, integrate it with CLAIMS.yaml, evidence, plan, and open problems in one coherent repository change, following Claude’s ordinary custody workflow.

Provide Josh major checkpoints, not a stream of micro-findings. A checkpoint should say which lanes are still running, which findings are provisional, and what downstream claims would fall if an audit fails. Verify that completed agents have actually terminated.

Stop rules

Stop and report rather than widening the project if:

• the positive theorem reduces to generic König compactness without a doctrine-specific lift;
• the best result is another isolated fixed-cut count;
• the needed initializer or precision is uncharged;
• the comparator shows the claimed mechanism is already standard at the same scope;
• the model begins smuggling the prefix or future into an oracle.

This is intentionally a broad research assignment with several coordinated lanes and one integrated outcome. It should not be broken into a succession of tiny requests from ChatGPT.

12. Coordination between ChatGPT, Claude, and Josh

ChatGPT

• leads direct mathematical attacks and synthesis;
• develops candidate theorems and counterexamples independently;
• prepares substantial proof packets;
• explains findings to Josh in plain language;
• does not assume routine repository management.

Claude

• owns repository continuity, evidence synchronization, verification, commits, and pushes;
• runs substantial independent challenge and comparator programs;
• integrates stable results into the repository;
• keeps Josh informed about running lanes and actual process status.

Josh

• selects strategic direction and decides whether a result earns a milestone, publication, or website update;
• may redirect either mathematical lane;
• should not have to manually reconcile conflicting micro-updates from the two systems.

The preferred exchange unit is a self-contained file containing a whole result, not a paragraph copied from chat.

13. Working and communication rules

1. Use compact, matter-of-fact mathematical prose. Avoid poetic or dramatic language about results “dying,” lasting “forever,” or transforming the entire project.

2. Lead with the positive statement. Do not open a load-bearing paragraph by asserting the frame that should be rejected and then saying “do not read it as that.”

3. Prefer direct attacks. A useful framework paragraph is welcome when it supports a proof, but do not substitute vocabulary for a theorem.

4. Give the primary mathematical lane room for large, coherent swings when the definitions and failure tests are pinned. Several of the strongest results in this program came from following a direct construction or obstruction beyond the initial narrow task.

5. When agents are used, give them bounded independent lanes with explicit falsifiers and deliverables. Check process state before saying the work is closed.

6. Do not pile new tasks onto a lane whose audit is still running.

7. A report being read is not the same as its agent being terminated.

8. Keep proof status, audit status, prior-art status, novelty status, and repository publication status separate.

9. Do not update the public website during active research. The last mid-run website update was too aggressive.

10. When repository integration happens, synchronize the claim ledger, evidence index, plan, open-problem ledger, proof source, and audit source. Passing YAML-shape checks alone is not evidence of consistency.

11. Preserve complete replacement files or self-contained packets when Josh needs to transfer work between systems. Avoid long copy-and-paste instructions.

14. Artifact routing

Do not load every artifact into every new chat. Use the smallest attachment set for the task.

Stage 1 — orientation and target selection

Attach only:

• FPRD-Lab-clean-chat-handoff-2026-07-29.md

This is enough to orient the clean thread, discuss the roadmap, and select the first attack. It is not enough to reconstruct the current proofs.

Stage 2 — actual causal/origami proof work

Before asking the clean thread to prove or refute the target in §§8–10, also attach:

• FPRD-causal-origami-source-bundle-2026-07-29.zip

The bundle contains the causal-factorization proofs, accessible and relational sewing drafts, counterexamples, checkers, and the reconciled Fold-and-Look model/proof/audit sources. Its index records the status and checksum of every member. Bundle SHA-256:

d7a367915c6e0e89d0a107c3c39d4ff992f61bf2d9078c916112de3d4dcf3518

If the clean chat is being opened specifically to start the mathematical attack immediately, attach the handoff and source bundle together.

Work on repository reconciliation

Also provide:

• the current repository or authoritative current status files;
• FPRD-REPAIR-MANIFEST.md;
• fprd-repair-b62aea3-to-reconciled.patch as historical source material, with the warning that current main is already past its base.

Provenance of the non-repository causal sources

These files are included in the source bundle:

Some older headers and framing inside those scratch files are stale. In particular, retire “conservation law,” unnamed-machine, and cell-probe-near-miss language. Use this packet’s corrected status.

Provenance of the Fold-and-Look sources

The source bundle contains copies from the reconciled fprd-repair working tree based on b62aea3. They were not yet committed as one integrated repair and must be compared with current main before repository use.

• docs/notes/wfl-weak-fold-look-model.md: pinned weak finite-precision doctrine and consolidation;
• docs/notes/bounded-fold-piecewise-isometry.md: structural piecewise-isometry source with its scope and pending-audit boundaries;
• docs/notes/fold-look-formula-resource-theorem.md: repaired positive-radius formula-resource theorem, independently re-audited;
• docs/notes/fold-palindrome-restricted-frontier.md: four-coordinate palindrome frontier using the repaired normal form;
• docs/probes/k-fold-incidence-and-permutations.md: conditional order-preserving separation, explicitly not independently re-audited;
• docs/verify/fold-resource-accounting-audit.md: adversarial accounting and uncharged-resource findings;
• docs/verify/virtual-crease-causal-width-audit.md: independent audit of the causal-width theorem chain;
• docs/litrev/bounded-fold-incidence-comparators.md: primary-source comparator map;
• docs/litrev/cone-factorization-comparators.md: standard-model and data-structure-regime comparator map.

The promoted-source banners on some repository notes retain an older “unverified” import label. Use the document body, the reconciled ledger status, and this packet to distinguish the later independent audits from that historical banner.

Work on weighted height

Also provide:

• FPRD-state-height-audit-packet.md;
• FPRD-cycle-dual-and-parity-comb-handoff.md.

15. Suggested first message in the clean ChatGPT thread

For orientation only, attach this file. To begin the direct proof attack, attach this file and the causal/origami source bundle, then send:

Continue FPRD Lab from the attached checkpoint. Read the whole packet before responding. Do not manage the repository unless I ask; Claude owns repository continuity. Your primary job is a direct mathematical attack on the accessible trailing-head sewing problem and the bidirectional origami/comparison-doctrine image problem in §§8–10. Start by reconstructing the exact doctrines and identifying the strongest theorem or smallest obstruction we should attack. Take a large coherent swing when the definitions support it. Do not update the website, make novelty claims, or send Claude piecemeal work.

That is enough to start. The clean thread should not require the old long chat.