The conversion report
The conversion report
The report is where everything the YAML cannot hold goes, and it is what makes the YAML trustworthy. A candidate file on its own asserts a set of claims with no account of where they came from or what was left behind; a reviewer has no way to tell a faithful extraction from a confident one. Nothing is ever dropped silently.
Write it to <slug>.report.md beside the candidate. Fill every section. "None" is
a real answer and a useful one; an absent section is not.
# <Title> — conversion report
| | |
| ------------------ | ------------------------------------------------------------ |
| Source | `warehouse-study.pdf` |
| Digest | `shasum -a 256` of the source file, one line per file |
| Pages | 34 |
| Pages read | 1–34 |
| Reading conditions | text layer throughout / pages 12–18 are scans read as images |
| Converted | 2026-08-10 |
| Candidate | `warehouse-throughput.ltp.yaml` |
| `check:import` | passes, exit 0 — 16 units in 2 waves, nothing excluded |
## What the file claims
11 propositions, 11 roles, 4 relationships, 1 assumption, 0 assessments.
current_reality 4 future_reality 2 goal 3 prerequisite 2
No conflict view, no transition view. See **Not in this file**.
## Items
| id | view | role | page | confidence | from the document | note |
| ------------------- | --------------- | ------------------ | ---- | ---------- | --------------------------------------- | ------------------------------------------------ |
| r-e-orders-wait | current_reality | undesirable_effect | 12 | high | "Orders wait an average of three days" | |
| r-e-reported-weekly | current_reality | observation | 15 | high | "reported weekly rather than per shift" | role: the document never says what this is for |
| f-e-release-to-line | future_reality | injection | 18 | medium | "we recommend releasing each order" | a recommendation read as a change, not as a task |
Confidence: **high** — one sentence states it as fact. **medium** — assembled from
two or more sentences, or the view was a judgment call. **low** — paraphrased
across a section, or read from a scanned page.
## Relationships asserted
| id | kind | from → to | page | the sentence that asserts the link |
| --------------------------- | -------------- | ------------------------------------- | ---- | ----------------------------------------------------------------------- |
| r-rel-batch-causes-wait | causes | r-e-batch-of-200 → r-e-orders-wait | 12 | "This is because picking starts only once 200 orders have accumulated." |
| r-rel-reporting-contributes | contributes_to | r-e-reported-weekly → r-e-orders-wait | 15 | "Weekly reporting partly explains why the delay has gone unremarked." |
If the last column cannot be filled, the relationship should not exist.
## Deliberately omitted
| what | where | why |
| ----------------------------------------------------- | ----- | -------------------------------------- |
| "Automation may also contribute." | p.12 | hedged claim too weak to assert |
| Releasing orders would fix the three-day wait | p.18 | cross-view link the importer refuses |
| "Capacity and shift length together drive batch size" | p.14 | joint causation the format cannot hold |
| "This report is organised as follows…" | p.1 | boilerplate |
Categories, and use these words: hedged claim too weak to assert; cross-view link
the importer refuses; joint causation the format cannot hold; no identifiable
subject; outside the system boundary; duplicate of an earlier claim; boilerplate.
## Judgment calls for the reviewer
1. **Is "throughput is reported weekly" a cause of the delay, or an observation
about it?** The document states it and calls it a partial explanation. Read as
an `observation` with a `contributes_to` link. Alternative: `intermediate_cause`
with a `causes` link, which claims more than page 15 says.
2. **Is the recommendation a solution or a task?** Read as a `future_reality`
injection, because the document says what should be true and not who does it.
Alternative: a `transition` action, which would need an actor the document never
names.
## Validator output
<the tail of `bun run check:import`, verbatim>
## Not in this file
- **No conflict view.** The document never states a dilemma; it treats the delay
as a straightforward capacity problem.
- **No transition view.** No actions with actors, dates or sequence.
- **No assessments.** The document recommends; it never claims to have established
a critical root cause.
These are gaps in the source, not in the conversion.
Two notes on why the sections are shaped this way.
Judgment calls are questions with options, deliberately mirroring
ImportDecision's own question and choices fields. The reviewer answers in the
same form the importer would have asked in, and the answer is directly actionable.
"Not in this file" exists to stop a reviewer blaming the conversion for a gap in the document. Without it, a four-view candidate reads as a failed extraction rather than an accurate one.
