# Developer agent prompt — Sibyla v5.0, assignment P1-0e Hand this to the coding agent on the **development machine**, working in `D:\fileStorage\repos\GOTT.Sibyla`, branch `feature/p1-0d-follow-up`. It is written to be pasted as the opening message of a fresh session. > **P1-0d is complete and is not reopened.** Its rows, its provenance markers, its A3 measurement > and its governance text all stand and were independently re-verified on 2026-08-05: the 45 DOCEFL > rows re-parse to 45 unique EFCodes with no gap in `EF0000001`–`EF0000045`, all 19 DTCodes and both > ITMCLS `CLCode`s check out against their stated blobs, the A2 register is exhaustive, and the > commit touched documentation only. **This assignment closes what that verification found, and > nothing else.** --- > ## Read this first — state, and where to get it > > **`docs/PROJECT-STATE.md` is the only source of project state.** Read it before this prompt's > context and before any draft. Everything else in the repository is a record of a moment. > > - **C1–C19, A1–A7, C3, D1–D9, group-B scope, S1–S4, B1 and A3 are all decided.** Do not reopen one. > - **P1-0 is accepted and closed.** Nothing here reopens it; every item below is a defect in how a > closed decision is written down, not a new decision about the design. > - **Only O5 is open on the whole project.** Do not report anything else as a blocker, and never > write "pending", "unconfirmed" or "awaiting Luís" into a draft. --- ## Why this assignment exists P1-0d did the hard part correctly: it replaced an all-or-nothing source gate with per-field provenance, and it left **34 fields explicitly `absent`** rather than inventing plausible values. That was the right call and it is not being undone. But `absent` means *blank*, and **blank has to be legal somewhere**. Verification found that it is not. **This is the one item that matters, and it has a deadline: P1-1a is writing the migration train now.** If it emits the declarations as written, the three rows P1-0d authored cannot be inserted, and the failure surfaces at seed time in P1-1b — six weeks after the decision that caused it, which is exactly the failure mode this project keeps paying for. --- ## Deliverable 1 · Reconcile `absent` with nullability — the only urgent item `docs/p1-0-schema-mapping.md` declares DOCEFL with **only three nullable columns** (`DetectionPattern`, `AppliesToField`, `AutoActionCondition`) and BNKMAT with **one** (`FlagReviewNotes`). Against those declarations, the three authored rows are blank in columns that are not allowed to be blank: | Row | Fields `absent` | Of which the declaration forbids NULL | |---|---:|---:| | `EF0000000` (C11 sentinel) | 16 | **13** — `FlagCategory`, `RiskFactor`, `AutoReviewable`, `Summary`, `FlowDimension`, `PolicyReference`, `RequiredAction`, `AutoActionAllowed`, `ReviewOwnerRole`, `ReviewPriority`, `ResolutionEvidenceRequired`, `EffectiveFrom`, `EnforcementStartsAt` | | `EF0000053` (C12 Monthly-gap) | 12 | **11** — the same list minus `Summary`/`RequiredAction`/`ResolutionEvidenceRequired`, plus `BlockingLevel` | | `BT000012` (D6 BNKMAT) | 6 | **6** — `ReconciliationType`, `AutoProposeAllowed`, `AutoCloseAllowed`, `RequiresReview`, `MaxDateWindowDays`, `ToleranceAmount` | **Thirty columns. Decide each one, and record the decision in the schema mapping next to the column declaration — not only in a status report.** There are exactly three legitimate outcomes per column, and you pick one: 1. **The column becomes nullable**, because a rule that describes the *absence* of a rule genuinely has no value there and never will. Write `NULL` into the declaration and say in one line why the column is optional *for every row*, not just for the sentinel. 2. **The column stays non-nullable and the row supplies a value**, because the field is in fact structurally implied and P1-0d was over-cautious. Then it is `authored` with a same-cell justification, exactly as `Active` and `RuleVersion` already are — and `docs/p1-0-c20-authored-rows.md` is updated so the two documents cannot drift. 3. **The column stays non-nullable and the row is not seeded in P1-1a**, because the value depends on an activation decision nobody has taken. Then say *when* the row is created, by which command, and what supplies the field — and note it for P1-1b. **Do not resolve this by inventing a sentinel literal** — no `'N/A'`, no `'Unknown'`, no empty string standing in for NULL, no `0` risk factor, no epoch timestamp. A fake value that satisfies a constraint is worse than an honest nullable column, and it is the exact failure the `absent` marker was created to prevent. Two specific cases you must not skip: - **`DOCEFL` unique `(FlagReasonCategory, FlowDimension)`.** If `FlowDimension` becomes nullable, PostgreSQL's default `NULLS DISTINCT` means this constraint stops binding for every row that has no dimension. State `NULLS NOT DISTINCT`, or change the constraint, or explain why unenforced is correct here. Do not leave it implicit. - **`EF0000053` is `Active=Yes` with no `BlockingLevel`, `ReviewOwnerRole`, `ReviewPriority` or `FlagCategory`.** An active detector with no owner, no priority and no blocking level is governed but not runnable: DOCFLG instances it raises cannot be routed. Either those four are structurally implied by C12 and route 2 applies, or the row ships `Active=No` and is activated by the same governed command that supplies `EnforcementStartsAt`. **Say which, in the schema mapping.** --- ## Deliverable 2 · Two stale statements - **`docs/p1-0-schema-mapping.md`, "Group-B placement":** *"B1 remains Miguel's separate byte-scope confirmation and is not decided here."* **B1 was closed on 2026-08-05** by `docs/p1-0-b1-decision.md`, and `PROJECT-STATE.md` §2.2 records it. Replace the sentence with the decision: byte storage is per capture, `RetainedContentHash` is a duplicate-detection index and never a storage pointer, reference counting is not implemented, purge is capture-scoped storage hygiene and not erasure. **Then grep the whole `docs/` tree for any other sentence that describes a closed item as open, and report what you find even if you fix it.** - **Patch artifacts.** A literal `+` from a diff leaked into two markdown headings and renders as body text: `docs/p1-0-codes-taxonomy-archive.md` line 167 (`+### Extracted pinned DOCTYP rows`) and `docs/p1-0-schema-mapping.md` line 233 (`+#### Extracted 45-row DOCEFL base seed`). Remove both leading `+`. Check for others with a line-start scan; do not hand-inspect only these two. --- ## Deliverable 3 · Line-ending hygiene, before it contaminates a commit The repository has **no `.gitattributes`**. Files are committed with LF and checked out on Windows with CRLF, so a working clone reports **272 files modified and roughly 89,000 changed lines that contain no content change at all** — `git diff --ignore-all-space` is empty against them. The next person who commits from a Windows clone buries a real change inside that noise. Add a `.gitattributes` at the repository root that normalizes text to LF in the index (`* text=auto eol=lf`) with the obvious binary exclusions for `*.pdf`, `*.png`, `*.svg` if you treat SVG as text decide it explicitly, and `*.ps1`/`*.cmd` if the toolchain needs CRLF on disk. **Then run `git add --renormalize .` and commit that renormalization as its own commit, touching nothing else**, so the noise commit is separable in history from any content commit. This is the one place you may touch paths outside `docs/` — and only for renormalization. **You still may not change the content of any file under `src/`, `tests/` or `migrations/`.** Verify with `git diff --ignore-all-space HEAD~1` on your renormalization commit: it must come back empty. There is also a stale zero-byte `.git/index.lock` in the user's clone at `C:\WIP\GOTT.Sibyla`. That is not yours to fix from the dev machine; **name it in your status report** so the user deletes it. --- ## Deliverable 4 · Close the loop - Update **`docs/p1-0-c20-authored-rows.md`** for every field that moved from `absent` to `authored`, keeping the same-cell justification format. A field that stays `absent` keeps its evidence line unchanged. - Update **`docs/PROJECT-STATE.md`**: add a changelog row for P1-0e, and state in §3 that **P1-1a's DOCEFL and BNKMAT column nullability is now specified** and what it is. O5 stays the only open item unless deliverable 1 route 3 creates a genuinely new one — if it does, add it to the §2 table with who closes it. - Write **`docs/p1-0e-status.md`**: the thirty-column decision table with the route taken for each, the constraint decision, the `EF0000053` activation decision, the list of stale statements found by the grep, and the exact paths changed. **The line to read first is the thirty-column table.** --- ## What you may not do - Reopen P1-0d's extractions, its A3 measurement, or any C/A/S/D/B decision. - Invent a placeholder literal to satisfy a NOT NULL constraint. - Change content under `src/`, `tests/` or `migrations/` — renormalization only, and only in its own commit. - Write, apply or plan a migration. P1-1a owns the train; you are specifying its inputs. - Touch the prototype repository in any way. - Write "pending", "unconfirmed" or "awaiting Luís" anywhere. - Merge to `main`. ## When to stop and report — scoped to the deliverable, never global Stop **the affected deliverable only**, finish the others, and report: - A column's nullability cannot be decided without a decision nobody has taken. Name the column, say what the decision is, and route it — do not stall the other twenty-nine. - The renormalization commit shows a non-empty `--ignore-all-space` diff, meaning something other than line endings changed. Stop that deliverable immediately and report it. ## Definition of done - All thirty columns are decided, each recorded next to its declaration in `docs/p1-0-schema-mapping.md`, with `docs/p1-0-c20-authored-rows.md` in agreement. - The `(FlagReasonCategory, FlowDimension)` uniqueness question and the `EF0000053` activation question are answered in the mapping. - No document in `docs/` describes B1 — or any other closed item — as open. - No markdown heading begins with a stray `+`. - `.gitattributes` exists; the renormalization is its own commit; `git diff --ignore-all-space` on that commit is empty. - `docs/PROJECT-STATE.md` reflects reality and `docs/p1-0e-status.md` exists. - `dotnet build GOTT.Sibyla.slnx` still passes. If the sandbox cannot reach `api.nuget.org` (`NU1301`), say so and run it on the host; a network failure is not a build failure.