diff --git a/README.md b/README.md index 922a28f..12afd6f 100644 --- a/README.md +++ b/README.md @@ -50,6 +50,7 @@ Example output includes truth accuracy, minority-truth recovery, Brier score, ab - [`EVIDENCE-ALIGNMENT.md`](EVIDENCE-ALIGNMENT.md) — claim-to-record ledger, corrections, and remaining release blockers - [`RESEARCH-DIRECTION.md`](RESEARCH-DIRECTION.md) — proposed dual evidence/search ledger program and absence-claim boundary - [`research/knowledge-ledger/`](research/knowledge-ledger/) — public research method, experiment registry, interoperability fixtures, and results +- [`research/durable-history/`](research/durable-history/) — proposed experiments separating evidential support from durable historical anchoring - [`RESEARCH-HYPOTHESES.md`](RESEARCH-HYPOTHESES.md) — falsifiable hypotheses using the required template - [`contracts/authority-evidence-v0.1/`](contracts/authority-evidence-v0.1/) — vendor-neutral authorization/evidence draft and semantic conformance rules - [`benchmark/`](benchmark/) — synthetic worlds, evaluation metrics, and CLI diff --git a/research/durable-history/EXPERIMENTS.md b/research/durable-history/EXPERIMENTS.md new file mode 100644 index 0000000..152aec4 --- /dev/null +++ b/research/durable-history/EXPERIMENTS.md @@ -0,0 +1,86 @@ +# Falsification program + +Status: **preregistering candidates; not executed.** + +Each experiment must freeze its fixtures, success criteria, implementation, and +environment before confirmatory execution. Null, failed, and incomplete outcomes +remain visible. + +## DH-001 — Immutable falsehood + +**Question:** Does anchoring change evidential status? + +**Null:** A false or unsupported fixture has the same evidential status before +and after anchoring. + +**Test:** Commit a deliberately false fixture, verify its timestamp and inclusion +proof, and then evaluate it under the knowledge-ledger rules. + +**Pass condition:** Verification proves byte identity and historical inclusion +while the conclusion remains unsupported or false. + +## DH-002 — Reproduction without anchoring + +**Question:** Can evidence support a result without a blockchain? + +**Null:** Independent reproduction and evidential support do not require an +external anchor. + +**Test:** Independently reproduce a frozen, unanchored fixture in a clean +implementation and compare protected fields and conclusions. + +**Pass condition:** The result reproduces, while its durable-history status is +accurately reported as unanchored. + +## DH-003 — Mutation detection + +**Question:** Does the receipt expose later alteration? + +**Null:** Any one-byte mutation to a covered artifact invalidates its inclusion +proof or top-level commitment. + +**Test:** Mutate every covered file position across bounded synthetic packages. + +**Pass condition:** Every mutation is detected; unchanged packages verify. + +## DH-004 — Copies do not become roots + +**Question:** Can hashing or anchoring manufacture evidential independence? + +**Null:** Duplicating one ground source produces no additional independent root, +regardless of the number of file hashes, signatures, or timestamps. + +**Test:** Anchor increasing numbers of verbatim and paraphrased descendants of +one source and rerun root accounting. + +**Pass condition:** The independent-root count and bounded conclusion remain +unchanged. + +## DH-005 — Immutable uncertainty + +**Question:** Does durable preservation pressure a system to overclaim? + +**Null:** When a required search location is unavailable, the anchored receipt +still concludes `not_established`. + +**Test:** Preserve a package with declared incomplete coverage and verify it +through independent implementations. + +**Pass condition:** Every rendering and receipt preserves the unavailable +location, uncertainty, and bounded conclusion. + +## Comparative preservation study + +After DH-001 through DH-005 pass locally, compare Git commits, signed releases, +transparency logs, OpenTimestamps/Bitcoin, and an EVM L2 anchor. Measure: + +- independent verifiability without the original operator; +- resistance to deletion and silent history rewriting; +- proof size, latency, monetary cost, and operational complexity; +- privacy leakage and metadata exposure; +- dependence on vendors, administrators, sequencers, and key custody; +- long-term availability of artifacts and verification software. + +No provider or chain wins by definition. The preferred method is the least +expensive mechanism that satisfies the preregistered durability and independence +requirements for the record's consequence level. diff --git a/research/durable-history/README.md b/research/durable-history/README.md new file mode 100644 index 0000000..c977517 --- /dev/null +++ b/research/durable-history/README.md @@ -0,0 +1,58 @@ +# Epistemic support and durable history + +Status: **proposed research track; no experimental result is claimed.** + +## Research question + +Can a knowledge system preserve the exact historical form of a qualified result +without confusing persistence with truth? + +This track separates four layers that are often collapsed: + +1. **Ground source** — where a claim first touches observable reality. +2. **Evidence method** — how provenance, independence, search coverage, + counterexamples, uncertainty, and bounded conclusions are evaluated. +3. **Independent reproduction** — whether a separately implemented method + recovers the same protected meaning and result. +4. **Durable history** — how hashes, signatures, retained copies, transparency + logs, or public timestamps preserve which exact record existed when. + +The central hypothesis is that keeping these layers explicit prevents a durable +record from being mistaken for a true claim, while still making silent historical +revision detectable. + +## Candidate contribution + +> Reality is the source. Evidence is our contact with it. Reproduction tests +> that contact. The ledger preserves its lineage. A public anchor can preserve +> the moment, but it proves neither the evidence nor the truth. + +This formulation is a research direction, not a novelty claim. Any claim of +novelty requires a recorded primary-source literature review and comparison with +prior work in provenance, reproducible research, transparency logs, trusted +timestamping, content-addressed storage, and blockchain anchoring. + +## Relationship to the knowledge ledger + +The knowledge-ledger experiments determine whether a conclusion is supported. +This track begins only after a result has acquired a stable evidence package. It +tests preservation and verification; it cannot promote, repair, or strengthen an +unsupported result. + +Ordinary experiments do not require a blockchain. Many qualified records may be +batched into one Merkle root and anchored periodically. Canonical milestones may +justify an independently verifiable public timestamp when cost, authority, and +privacy permit. + +See [`EXPERIMENTS.md`](EXPERIMENTS.md) for the falsification program and +[`RECEIPT-SCHEMA.json`](RECEIPT-SCHEMA.json) for the proposed portable receipt. + +## Claim boundary + +An anchor may support only this statement: + +> A commitment to these exact bytes existed no later than the recorded point in +> the referenced public history. + +It does not establish that the committed statement is true, complete, +independent, safe, novel, or authorized. diff --git a/research/durable-history/RECEIPT-SCHEMA.json b/research/durable-history/RECEIPT-SCHEMA.json new file mode 100644 index 0000000..a4af596 --- /dev/null +++ b/research/durable-history/RECEIPT-SCHEMA.json @@ -0,0 +1,25 @@ +{ + "schema": "minority-prophet.durable-history-receipt.v0.1-draft", + "status": "proposed-not-validated", + "recordId": null, + "manifestDigest": null, + "merkleRoot": null, + "evidenceStatus": null, + "reproductionReceipts": [], + "signatures": [], + "anchor": { + "method": null, + "network": null, + "chainId": null, + "transactionHash": null, + "contractAddress": null, + "commitment": null, + "includedAt": null, + "finalityEvidence": null, + "publicVerificationInstructions": null + }, + "limits": [ + "Anchoring establishes historical commitment, not truth.", + "This draft schema has not passed the proposed experiments." + ] +}