# EM-0029 execution plan

- Object ID: `em:documentation:sha256:fcdd9d6eafc2e408656424dceb994a2a35719cc4ebb7f8512bfdd5f4eec22952`
- Kind: `documentation`
- Repository path: [`docs/execution-plans/EM-0029.md`](https://github.com/yoheinakajima/epistemedia/blob/f92846570180dfa4511263f8ba98ecd18f7772c9/docs/execution-plans/EM-0029.md)
- Content digest: `60fc7875b85dad4ab1c85bfd78290ffacf165eca2c2edf18d0bdceb54b7f91f9`

**Also filed under:** [Disclosure and Public Projection](https://epistemedia.org/topics/disclosure/), [Epistemedia](https://epistemedia.org/topics/epistemedia/), [Epistemic Mesh Protocol](https://epistemedia.org/topics/epistemic-mesh/), [Sovereign Realm Federation](https://epistemedia.org/topics/federation/), [Autonomous Governance](https://epistemedia.org/topics/governance/), [Knowledge Objects](https://epistemedia.org/topics/knowledge-objects/), [Human and Agent Interfaces](https://epistemedia.org/topics/public-interfaces/), [Releases and Reproducibility](https://epistemedia.org/topics/releases/), [Research Program](https://epistemedia.org/topics/research-program/), [Security and Adversarial Robustness](https://epistemedia.org/topics/security/)

## Source content

# EM-0029 execution plan

Status: first-review correction validated; exact-head successor review pending. The dossier is
research-only and is not admitted, published, or deployed.

## Objective

Transform the accepted EM-0026 agent-citation trace packet into one reversible Case 002 candidate
dossier. Preserve every accepted answer, trace, readback, normalization, ledger, and review byte;
make propositions, exact spans, no-credit dispositions, and dependence relations inspectable; and
compile materially different encyclopedia and skeptical readings over the same dossier.

## Authority and accepted base

- immutable task: `tasks/contracts/EM-0029.json`;
- accepted base: `75f01214750e9eccc4235cbe2a56eada1c014168`;
- accepted source packet: `research/how-we-know/agent-citation-lineage/` under EM-0026;
- allowed changes: this plan, the Case 002 research summary, EM-0029 files in the source-packet
  directory, and append-only run receipts;
- forbidden in this task: catalog admission, compiler or schema changes, public site output,
  workflow changes, new agent runs, new research retrieval, and deployment.

## Preserved accepted inputs

The builder reads, but does not rewrite:

- `evidence-ledger-v1.json` — 395,409 bytes, SHA-256
  `29034edff99ae40f03d244998ee1d238917ac387422dea221b80897c94d556b3`;
- `source-normalization-v1.json` — 31,231 bytes, SHA-256
  `55f7beca95ef2438bfb6fb9bf6ee553c93aee56391a5f13d02e597d5002b7474`;
- `source-readbacks-v1.json` — 14,830 bytes, SHA-256
  `58dd334d395fda27e04f2f692d2ad7278d12224fe5e03947b963671871609146`;
- `span-readbacks-v1.json` — 87,681 bytes, SHA-256
  `48ddabaec9b75acf43bdfc9ab917bee6e6fece418316eba8f1012e646d59dde5`;
- `independent-review-receipt.json` — 7,094 bytes, SHA-256
  `aadc83de7c4a57b60593124917bbb5fedd5e64c7aa50db3d71b4817626532ae4`.

The eight accepted answers and eight accepted trace records are also immutable inputs. A base-to-head
byte comparison is required before author handoff and again during independent review.

## Construction

1. Derive the 8 report, 48 citation-occurrence, 30 cited-URL, 27 resolving-root, 11 work,
   14 edition, 72 exact-span-root, 7 candidate-warrant, 34 unresolved-citation, and 20
   unsupported-or-force-raised totals from the accepted typed records.
2. Project only quote-minimal exact spans into structured JSON editions, preserving canonical URL,
   edition, retrieval, license, and occurrence identities.
3. Model all eight report lineages beneath one shared, partly unknown capture lineage. Model all
   seven candidate warrants beneath an unresolved source/method/derivation dependence boundary.
4. Keep four pending warrants, nine independently rejected claims, twenty no-credit/force-raised
   claims, thirty-four unresolved citations, and three inaccessible carriers explicitly visible.
5. Compile one encyclopedia evaluation and one skeptical evaluation over the unchanged graph.
6. Stamp and strictly validate the dossier; reject span or lineage mutation; reproduce it
   byte-for-byte with `build_candidate.py --check`.

## Candidate identity

- path: `research/how-we-know/agent-citation-lineage/candidate-dossier.json`;
- dossier ID: `em:dossier:sha256:cbd7a14096a956f642f5c76046d3b49ed648fbe6bf24144c992404a01415af82`;
- file identity: 483,595 bytes, SHA-256
  `1ae06b54fe6c6ce1803836bbf2ecaf3e652bed2c6878b7e095c01a1c689ab87b`;
- review-supplement identity: 3,021 bytes, SHA-256
  `7daa7d56cf42f764448235074cf4e87abb81d4f3a99d16dae063b844dc6aa6c0`;
- stage: `draft`;
- independently confirmed warrant roots: `0`.

The dossier ID is not an acceptance ID. It identifies candidate bytes before independent review.

The forward correction after first review adds exactly three separately typed, quote-minimal
qualification spans: the DeepTRACE `50.3%` table cell, its conflicting `40.3%` prose, and the
URL-health paper's statement that DRBench supplied pre-collected model outputs. They close two
material sentences without changing the accepted EM-0026 count of 72 exact-span roots or adding
warrant credit.

## Validation and handoff

- validate the frozen eight-trace protocol with `--require-traces`;
- verify the accepted evidence ledger and candidate deterministic rebuild;
- lint the new builder and run `git diff --check`;
- verify accepted packet paths are byte-identical to the accepted base;
- run the full deterministic `make check` and disclosure audit;
- compare the exact changed path set with EM-0029 authority;
- commit and push one scoped branch and open a draft PR;
- leave the PR draft and unmerged for fresh-clone, exact-head independent review.

EM-0030 remains sequenced behind that review. It alone can propose library admission and public
projection; neither EM-0029 nor this plan claims that Case 002 is live.

## Author validation record

- the v2 protocol verifier passed with eight of eight terminal trace records;
- the evidence-ledger verifier reproduced all accepted counts, including 7 candidate warrants,
  34 unresolved citations, and 20 unsupported or force-raised claims;
- `build_candidate.py --check`, adversarial dossier mutation checks, and Ruff passed;
- accepted EM-0026 answers, traces, readbacks, normalization, ledger, and review receipt are
  byte-identical to the accepted base;
- `make check` passed 82 tests, accepted-input validation, disclosure audit with no findings,
  independent deterministic rebuild, and unchanged source-tree state;
- claim receipt: `runs/proposals/20260823T213817Z-0a58c6f976e4.json`;
- author validation receipt: `runs/proposals/20260824T010901Z-c0bf306a8288.json`.
- first-review correction receipt: `runs/proposals/20260824T071224Z-5047c29d74aa.json`.

Exact candidate commit, tree, draft PR, and CI identities are appended only after the scoped remote
branch exists. Independent review remains a separate exact-head gate.
