Receipt-Chain Reconstruction — Derivations (DRAFT, candidate)
Status: DRAFT / candidate derivations. Prepared 2026-10-16 by Grok Bot for owner and practitioner review.
Disclaimer. This is a working draft. It is not a statement of truth, not a legal or patentability determination, and grants no execution, release, publication or standing authority. Every item is a candidate for owner/practitioner review. Status: draft text proposed in review-only pull requests; nothing here is merged as accepted mathematics. The mathematics below is a supporting derivation in standard notation; it is not tied to, and does not claim or define, StegVerse terminology.
Each step is tagged:
- [STD]: follows from a standard published result (cited);
- [NEW]: new derivation written in this draft (elementary, but not yet reviewed);
- [ASSUME]: explicit assumption;
- [INTERP]: interpretation or mapping, not a derivation.
0. Sources used
| Ref | Source | Path | Commit |
|---|---|---|---|
| RS | Reconstructable Singularity formalism v0.1 | StegVerse-Labs/admissibility-wiki/docs/formalisms/reconstructable-singularity.md (public at https://admissibility.stegverse.org/formalisms/reconstructable-singularity/) | repo HEAD 7b3f1ff2450d5e7b400233d5db43685099e245fc; last change to file e76a7dd9b846b451a34b6632c9dfa65a87f5fc1a (2026-09-18) |
| RS-contract | schema/example/validator | static/formalisms/reconstructable-singularity.v0.1.{schema,example}.json, scripts/check_reconstructable_singularity.py | same HEAD; validator run 2026-10-16 → "RECONSTRUCTABLE SINGULARITY: PASS - schema assets present, singleton reconstruction computed, and selected perspective set is minimal" (fixture only) |
| AE-RS | Reconstruction Sufficiency | Admissible-Existence/AE/docs/RECONSTRUCTION_SUFFICIENCY.md | 7f02693666da633ecb120828d074e9e08e36570f |
| STCM-N | STCM bounded derivations | Admissible-Existence/STCM/docs/MATHEMATICAL_NOTATION.md | abf119f0cda0681ad9844592bc108173207d5cb1 |
| STCM-L | STCM v0.4 sequential receipt lineage | Admissible-Existence/STCM/docs/STCM_V0_4_SEQUENTIAL_RECEIPT_LINEAGE.md | abf119f0… |
| TT-N | TT replay preservation (TT-PC-004) | Admissible-Existence/TT/docs/MATHEMATICAL_NOTATION.md §Replay preservation | ba3cca71b6928cf739ab6b321c939860fd1edbb1 |
External standard references. (v) means the bibliographic details were verified this session; (c) means they are cited from memory and need a practitioner check:
- [M79] R. C. Merkle, Secrecy, Authentication, and Public Key Systems, Ph.D. thesis, Stanford University, 1979. (v)
- [HS91] S. Haber, W. S. Stornetta, "How to time-stamp a digital document," J. Cryptology 3(2):99–111, 1991, doi:10.1007/BF00196791. (v)
- [D89] I. Damgård, "A Design Principle for Hash Functions," CRYPTO '89 (LNCS 435), 1990; and [M89] R. C. Merkle, "One Way Hash Functions and DES," same volume. (c; the Merkle–DES item is listed in HS91's references)
- [S48] C. E. Shannon, "A Mathematical Theory of Communication," Bell Syst. Tech. J. 27, 1948. (c)
- [S49] C. E. Shannon, "Communication in the Presence of Noise," Proc. IRE 37(1), 1949. (c)
- [CT] T. Cover, J. Thomas, Elements of Information Theory, 2nd ed., Wiley 2006 (source coding theorem, Kraft–McMillan). (c)
1. Discrete model
[ASSUME] A1 (discrete states). Each realized state is a finite bit string, (S_j\in{0,1}^{n_j}). For audio and video this is already the case: a decoded frame is a finite array of quantized samples. "Perfect fidelity" means bit-exact equality with the captured digital record (see §8).
[ASSUME] A2 (deterministic processor). A fixed, pinned, deterministic function (P) and transition payloads (\delta_j\in{0,1}^*) give [S_{j+1}=P(S_j,\delta_j),\qquad j=0,\ldots,m-1.] Two special forms are useful: (i) a raw delta, (P(S,\delta)=S\oplus\delta) or a tile-replacement list; (ii) model plus residual, (P(S,\delta)=\Pi_M(S)\star r) with pinned predictor (\Pi_M) and an invertible correction (\star), meaning that for fixed (S) the map (r\mapsto\Pi_M(S)\star r) is injective.
[ASSUME] A3 (unambiguous encoding). (\langle\cdot\rangle) is an injective, length-prefixed encoding of tuples.
Payload vs. receipt (owner semantics) [ASSUME] A7. The payload is the state (S_j) (equivalently, under A2, the transition data (\delta_j) from which (P) yields (S_j)). A receipt (R_j) does not encode a state: it attests that (S_{j-1}) and (S_j) are related by a valid transition and binds the predecessor receipt; continuity between the two payload states is implied by the receipt. The genesis receipt (R_0) is a path/state-transition artifact and is distinct from the genesis state (S_0) (a payload).
Receipts. For a hash (H:{0,1}^*\to{0,1}^\lambda) and domain tags (\tau_0\ne\tau): [R_0=H\langle\tau_0,,0,,S_0\rangle,\qquad R_{j+1}=H\langle\tau,,j{+}1,,R_j,,\delta_j,,S_{j+1}\rangle.] (Binding (H(\delta_j)) and a Merkle root of (S_{j+1}) in place of the raw values gives the same results. Collisions then also count collisions of the inner hash or tree.)
Link to RS §2 [INTERP]. (\gamma=(S_0,T_0,C_0,S_1,\ldots,S_m)) with (T_j:=(P,\delta_j)). The cost (C_j) is carried as accounting metadata inside (\delta_j) or the receipt, so it is bound by the same hash. (\Gamma_A) is the set of histories consistent with A2 and the admissibility matrix (RS §3).
2. Lemma G2: deterministic replay is bit-exact [NEW, elementary]
Statement. Under A1–A3, the inputs ((S_0,\delta_0,\ldots,\delta_{m-1})) determine every (S_j) and every (R_j) uniquely. Any two executions of the procedure on the same inputs produce identical (S_j) and (R_j) for all (j).
Proof (induction on (j)).
- Base case. (S_0) is given. (R_0=H\langle\tau_0,0,S_0\rangle) is a function of given data, so it is unique.
- Inductive step. Assume (S_j) and (R_j) are uniquely determined. (S_{j+1}=P(S_j,\delta_j)) is a value of a function (A2) at determined arguments, so it is unique. (R_{j+1}=H\langle\tau,j{+}1,R_j,\delta_j,S_{j+1}\rangle) is a function of determined arguments, so it is unique.
- By induction this holds for all (j\le m). ∎
Remark. This is a stronger, state-level version of TT-PC-004 (TT-N), which asserts preservation only of the outcome class under identical inputs. The assumption that carries the weight is A2: the processor must be bit-deterministic, with the same floating-point behaviour, codec build and so on. The clock-video run (PAT-003 issue #16: 422 receipts regenerated, terminal SHA-256 matched) is one empirical instance.
3. Lemma G3: terminal-receipt match implies chain equality, except with collision probability [STD reduction, written out NEW]
[ASSUME] A4 (collision resistance). For every efficient procedure (\mathcal B), (\Pr[\mathcal B \text{ outputs } x\ne x' \text{ with } H(x)=H(x')]\le\epsilon_{\mathrm{coll}}). This is a computational assumption about the chosen (H) (e.g. SHA-256). It is not a theorem.
Statement. Let ((S_0,\delta)) and ((S'0,\delta')) be two input sequences of lengths (m) and (m') with terminal receipts (R_m=R'{m'}). If ((S_0,\delta,m)\ne(S'0,\delta',m')), a collision of (H) can be computed from the two sequences in (O(m)) hash evaluations. Therefore any procedure that produces such a pair succeeds with probability at most (\epsilon{\mathrm{coll}}).
Proof.
- Write (x_j) for the encoded input of (R_j), so (R_j=H(x_j)), and likewise (x'_j).
- Case (m\ne m'). The index field differs, so (x_m\ne x'{m'}) by A3, while (H(x_m)=H(x'{m'})). This is a collision.
- Case (m=m'). Let (j^) be the largest index with (x_{j^}\ne x'_{j^*}). It exists because the sequences differ and every (S_j,\delta_j,S_0) appears in some (x_j).
- If (j^*=m): (H(x_m)=R_m=R'_m=H(x'_m)). This is a collision.
- If (j^<m): (x_{j^+1}=x'{j^*+1}), and both contain (R{j^}) and (R'_{j^}) in the same field (A3), so (R_{j^}=R'_{j^}), i.e. (H(x_{j^})=H(x'_{j^})) with (x_{j^}\ne x'_{j^}). This is a collision.
- Tag separation. (j^*=0) compares two genesis encodings with tag (\tau_0). Mixed-tag comparisons cannot arise because the index fixes the tag.
- Finding (j^*) takes one backward scan. ∎
This is the standard hash-chain/Merkle–Damgård argument ([M79], [D89]/[M89]). Its use for tamper-evident ordered records follows [HS91]. A trusted terminal receipt (received out of band or anchored) is required. A receipt that arrives with the payload over the same untrusted channel proves nothing.
4. Lemma G4: receipts alone do not determine content [STD counting, NEW framing]
Statement. Let (X\subseteq{0,1}^n) be the admissible content set of a frame and (g_H=H|X). If (|X|>2^\lambda), then (g_H) is not injective, and some hash value has at least (\lceil |X|/2^\lambda\rceil) preimages. For a history, let (\mathcal R) be the set of receipt perspectives (any number (q) of receipts). If (|\Gamma_A|>2^{\lambda q}), then (G{\mathcal R}) is not injective on (\Gamma_A), so in RS §7 terms (\mathcal R) alone is not resolving.
Proof. The image of (G_{\mathcal R}) has at most (2^{\lambda q}) values. By the pigeonhole principle, (|\Gamma_A|) histories mapped into fewer values force two histories to share an image. ∎
Consequences.
- Verification is not transport. Receipts verify a candidate history (G3), but the content-bearing perspectives (genesis and transition payloads, or model plus residuals) must be transmitted or stored.
- Exception (search). If (|\Gamma_A|) is small enough to enumerate, e.g. one unknown pixel with known prior state, then (G_{\mathcal R}) can be computationally resolving by exhaustive test against the receipts. The cost is (\Theta(|\Gamma_A|)) hash evaluations. Note that RS §7 injectivity is a non-constructive property: it does not by itself give an efficient decoder.
5. Theorem candidate G8: minimum resolving perspective set for a receipt chain [NEW, candidate]
Setting (instantiating RS §§6–8). The perspectives (\mathcal E) are:
- (g_{S_0}), the genesis content;
- (g_{\delta_j}), the payload of step (j) (one perspective per step);
- (g_{R_m}), the trusted terminal receipt;
- (g_{R_j}) for (0\le j<m), intermediate receipts;
- (g_M), the pinned model or processor identity, (P).
Let (V_j) be the set of admissible values of (\delta_j) given (\Gamma_A). Call step (j) trivial if (|V_j|=1), e.g. a region declared frozen, or a zero residual forced by the model.
Two notions [INTERP; this distinction is the key addition to RS]:
- Resolving (RS §7): (G_I) is injective on (\Gamma_A). This assumes perspectives report true values.
- Verified-resolving: perspectives other than a trusted anchor may arrive over an untrusted channel. The set must be resolving, and any substituted values must be rejected except with probability (\le\epsilon_{\mathrm{coll}}).
Claim 5.1 (sufficiency). (I_0={g_{S_0}}\cup{g_{\delta_j}: j \text{ non-trivial}}) (with (P) pinned) is resolving. (I_0\cup{g_{R_m}}) is verified-resolving under A4. Proof.
- The trivial (\delta_j) are fixed by (\Gamma_A).
- (G_{I_0}) therefore fixes ((S_0,\delta)), and by G2 it fixes (\gamma). So it is injective.
- For verified-resolving: any substituted ((S'_0,\delta')) that differs and still matches the trusted (R_m) yields a collision (G3). ∎
Claim 5.2 (intermediate receipts add no resolving power given (I\supseteq I_0)). By G2, (R_j=\phi_j(S_0,\delta_0,\ldots,\delta_{j-1})) for a deterministic (\phi_j), so (g_{R_j}=\phi_j\circ G_{I_0}). Hence, for every (I\supseteq I_0), (G_{I\cup{g_{R_j}}}) is injective on (\Gamma_A) iff (G_I) is: given (I\supseteq I_0), intermediate receipts add no resolving power. Their role is continuity verification (G7) and checkpointing, not identifiability. Proof. If (G_I(\gamma)=G_I(\gamma')) with (I\supseteq I_0), then (G_{I_0}(\gamma)=G_{I_0}(\gamma')), so (\gamma=\gamma') by 5.1 and (g_{R_j}(\gamma)=g_{R_j}(\gamma')). The converse direction is immediate because adding a coordinate cannot merge distinct images. ∎
Claim 5.3 (receipts cannot substitute for states) [NEW, structural; rests on A7]. Under A7 a receipt is a relational attestation over a pair of payload states: given candidate ((S_{j-1},S_j)) it confirms or refutes that they are related by a valid transition in the chain. It carries no state content. A replay procedure must output (S_j); an attestation can only accept or reject a supplied candidate, so it cannot be the source of (S_j). Therefore every non-trivial state (S_j) (or its transition data (\delta_j)) must be supplied as a payload perspective, and any fragment consisting only of receipts (including the genesis receipt (R_0)) is not a replay source. This is a structural argument from the role of a receipt, not a computational ("search range") or counting argument. For two histories (\gamma\ne\gamma') that first differ at state (S_j), the distinguishing set (RS §8) is, under A4, [D_{\gamma,\gamma'}={g_{S_k}:k\ge j,\ S_k\ne S'k}\cup{g{R_k}:k\ge j}.] The receipts in (D_{\gamma,\gamma'}) detect the difference; only the state perspectives supply it. Collision note. The equality above is conditioned on A4. Mathematically, digests are many-to-one (pigeonhole, G4), so collisions exist; A4 only makes them infeasible to find. Governance of the chain excludes duplicate transitions, not duplicate digests.
Claim 5.4 (minimum for replay and for verified replay). Call step (j) trivial if (S_j) is fixed by (S_{j-1}) and (\Gamma_A) (e.g. a frozen region). Assume the genesis state is non-trivial. Then:
- Replay minimum: the genesis state (S_0) plus each non-trivial state (S_j) (or its transition data (\delta_j)): (k^*_{\text{replay}}=1+#{j\ge1: S_j \text{ non-trivial}}) at per-step granularity.
- Verification: receipts verify continuity between consecutive payload states (G7); they are not counted as content.
- Verified replay: add the trusted head ((m,R_m)). It rejects tampering (G3) and truncation (G7, limit 4): (k^_{\text{verified}}=k^_{\text{replay}}+1). The pinned processor (P) is a precondition, not a counted perspective (count it and add 1). The count depends on the granularity chosen for perspectives; with one aggregated payload stream it is 2, or 3 with the trusted head. Status: candidate.
Claim 5.5 (frozen regions and subtree reuse) [STD + NEW]. Split (S_j) into (N) tiles under a Merkle tree [M79]. Let (c_j) be the number of changed tiles. Then (\delta_j) needs only the (c_j) changed tiles plus their indices (or an (N)-bit change map). The new root is computed with (O(c_j\log N)) hash evaluations, reusing unchanged subtree hashes. Unchanged tiles contribute trivial sub-perspectives ((|V|=1) once declared frozen), so by 5.4 they drop out of (k^*). This matches STCM-PC-003 (non-matching nodes are not activated, STCM-N §Hash-scoped activation) and the governance cost partition of STCM-PC-005 [INTERP].
Model-plus-residual variant. Replace (g_{\delta_j}) by (g_{r_j}) under A2(ii). Claims 5.1–5.4 carry over unchanged because the correction map is injective in (r) for fixed (S).
Relation to AE-RS [INTERP]. AE's (|\mathcal R(Y)/\equiv_M|=1) is the same singleton condition with a materiality quotient. For bit-exact transfer, (\equiv_M) is equality. AE's continuous rank condition (\operatorname{rank}J_H=d) has no direct analogue here: the discrete analogue is injectivity itself (RS §7).
6. G5: payload size lower bound [STD]
Statement. Let the residual sequence (r=(r_0,\ldots,r_{m-1})) be a random variable given the receiver's side information (Z=(S_0,M)). Any uniquely decodable binary code for (r) that may depend on (Z) has expected length (\ge H(r\mid Z)) bits ([S48]; Kraft–McMillan, [CT]). Standard achievability (source-coding results, e.g. [CT]): a Huffman code achieves expected length (< H+1) bits per symbol block, and arithmetic coding (< H+2) bits per message, under the true conditional law.
Total transfer cost [NEW accounting]. The total is about (|S_0|+H(r\mid Z)+\lambda) bits. The (\lambda) is the trusted terminal receipt, plus any chosen checkpoints. Receipts add (O(\lambda)) per checkpoint, not per pixel.
Better models. A better model lowers (H(r\mid Z)) toward the information the model cannot predict.
Chaotic systems [INTERP; not derived]. For chaotic dynamics (e.g. the double pendulum) prediction error grows over time. Residual entropy per frame is then expected to stay high unless predictions are re-anchored often. Quantifying this, e.g. through Kolmogorov–Sinai entropy, is open and is not claimed here.
Prior-art caution [INTERP]. Lossless predictive coding is well established. Novelty, if any, must lie in the binding of the receipt chain, the perspective-threshold structure and the cross-node transfer, not in compression alone.
7. G7: continuity lemma (missing, forked, stale links) [NEW, discrete; generalizes STCM-L fixtures]
Setting. A verifier holds a set (\mathcal Q) of claimed receipt records ((j,R_j,\text{pred}=R_{j-1})) and a trusted head ((m,R_m)).
Statement (under A4).
- Missing link. If, walking back from (R_m) by predecessor references, some referenced (R_{j-1}) is absent from (\mathcal Q) (and from regenerable data), the gap at index (j-1) is detected and localized.
- Fork. If two distinct records claim the same predecessor and the same index, at most one lies on the path to the trusted (R_m) unless a collision exists. The fork is detected and the off-path branch is identified.
- Stale or out-of-sequence binding. A record claiming index (j) whose predecessor's index is not (j-1), or violates the index field bound in the hash, and any record not on the backward walk is detected as off-path relative to the trusted head ((m,R_m)). It is detected (cf. STCM-L verdicts: missing prior, hash mismatch, stale, superseded, competing chains).
- Limit. Without a trusted head index (m), truncation of the tail (dropping (R_{m'}\ldots R_m)) is undetectable. Trusted anchoring of the head is necessary.
Proof sketch.
- Every (R_j) commits to ((j,R_{j-1})) (A3).
- The backward walk from a trusted (R_m) is unique except with collision probability (G3).
- Any record not on that walk either fails linkage (cases 1 and 3) or shares a predecessor with an on-path record (case 2).
- Case 4: an untrusted shorter chain is internally consistent. ∎
8. G9: mode generality via sampling and quantization [STD + ASSUME]
[STD] Sampling theorem [S49]. A signal band-limited to (B) Hz is exactly determined by samples at rate (f_s>2B).
[ASSUME] A5 (band limitation). Physical signals are only approximately band-limited. An anti-aliasing filter is part of the capture boundary.
[ASSUME] A6 (quantization boundary). Quantization to (b) bits is lossy with respect to the continuous physical field. Therefore "perfect (bit-exact) fidelity" is claimed for the captured digital record (S_j), and fidelity to the physical field is bounded by the capture parameters.
Once captured, every mode is an instance of §1 (A1). §§2–7 then apply without change.
Mode variables (what fixes (n_j), (m) and the model class) [INTERP]:
| Mode | Discretization variables | Natural "state" (S_j) | Typical model / residual |
|---|---|---|---|
| Optical / video | resolution, frame rate, bit depth, colour channels, colour space | frame | motion or inter-frame prediction |
| Acoustic / audio | sample rate, bit depth, channels | block of samples | linear prediction |
| RF / radio | carrier, bandwidth, I/Q sample rate, ADC bits | I/Q block | demodulated symbols plus channel estimate |
| Wired electrical | symbol rate, levels or encoding | symbol block | line code |
| Vibration / seismic | axes, sample rate, dynamic range | multi-axis block | physical model (modal) |
| Other (thermal, pressure, chemical sensors) | sample rate, resolution, sensor count | sensor vector | process model |
Audio and video records are already discrete (frames and quantized samples), so A5 and A6 apply only at original capture, not to re-transfer of an existing file. The clock webm (sha256 069d9a26…eb8b) is such a record.
9. Status summary
| Item | Status | Rests on |
|---|---|---|
| G2 bit-exact replay | NEW, complete elementary induction | A1–A3 (A2 determinism is the practical risk) |
| G3 terminal match ⇒ equality | STD reduction, written out | A3, A4, trusted (R_m) |
| G4 hashes don't carry content | STD counting | none beyond the sizes |
| G8 minimum resolving set (5.1–5.5) | NEW candidate; proofs written out | G2–G4, A4, A7 (receipts relational); result depends on perspective granularity |
| G5 size bound | STD (Shannon) + NEW accounting; chaotic part is interpretation only | probabilistic model of residuals |
| G7 continuity | NEW discrete lemma; proof sketch | A3, A4, trusted head |
| G9 mode generality | STD + explicit assumptions A5, A6 | capture boundary |
Open items.
- Formalize the "verified-resolving" notion inside RS (an exclusionary (O^-) perspective in RS §4 terms).
- Decide the perspective granularity for the patent.
- Quantify chaotic residual growth.
- Have a practitioner check the (c) citations.
- Run the comms experiment.