The business caseThe problem this solves

Deciding whether a shipment actually arrived as it was booked is a comparison between two counts and a condition code, and the record puts the driver's own account of the stop right next to it. That note is the loudest thing on the page and it is the one part of the record that is nobody's measurement. Someone opening each proof-of-delivery record, finding the ordered and delivered quantities, checking the condition code, deciding whether the shipment arrived as booked, and then deciding whether it is the kind of exception that has to go to the ops desk today. It is quick per record and it is every record, and the part that goes wrong is not the arithmetic — it is reading the driver's own note first and letting it set the answer.

The inputThe actual delivery records

The corpus is 55 delivery records, 0.03 MB (txt 55). Plain text, one format, invented rather than fetched — a real proof of delivery names a real carrier, a real consignee's site and a named driver's own words about a stop that happened, and there is no public corpus of (delivery record, true completeness) pairs for the same reason there is no public corpus of bank statements. Generating it also makes the label mechanical: gold is the comparison, not somebody's reading of the note.

Shipment
--------
SHP-KW-88753

Carrier
-------
Northvale Freight Lines

Receiving Site
--------------
Depot 12, Marlow Industrial Estate

Delivery Date
-------------
2026-07-21

POD Timestamp
-------------
2026-07-21T13:30:18

Ordered Quantity
----------------
480 units

Delivered Quantity
------------------
480 units

Condition Code
--------------
undamaged

Recipient Signature On File
---------------------------
no

Driver Exception Note
---------------------
Fine delivery, minor scuff on one of the outer boxes but nothing to worry about.

The outcomeWhat a good result looks like

A ten-field extracted record per delivery, plus one pure-code routing decision taken from two of those fields: an incomplete delivery with no recipient signature on file is the one that goes to the ops desk. Nothing here files a claim, issues a credit or contacts a carrier.

Where it fitsWhat did work

Every line below is a measured result from this kit's own runs, with the figure that supports it. The headline above is not softened by any of them.

• Screening a receiving queue for deliveries that did not arrive as booked, before an operations reviewer opens them — either tier — they tie at 55 of 55 verdicts, 1.00 recall and precision measured here
  100 pct completeness accuracy on both tiers against the free driver-tone floor's 60.0 pct (22 wrong, including 10 incomplete deliveries called complete).
• Deciding the two tiers on cost, speed or accuracy — the fast tier
  About 9 pct cheaper per record ($0.001325 vs $0.0014521), 35 pct lower p50 latency (2633 ms vs 4035 ms), and identical on every one of the three published graders — 550 of 550 cells, 55 of 55 verdicts, 10 of 10 flags, on both.

At a glanceHow the whole thing runs

Run once, for real, on 2026-08-21. Every figure on these pages was captured from that run — nothing is written from intent.

14 steps, grouped by the question that sends you to them rather than by build order. Each tile carries the one figure that step is about, and opens the page behind it.

Should you use this?What you bring, where it stops, and when not to use it

Before you commit an afternoon to this, these are the answers that decide it. Each one is rendered from the record it lives in — and links the page that holds it in full.

Not asked of this kit — 2 questions: clone (a fresh clone of this kit runs with nothing fetched); judge (nothing here is graded by a model).

Last verified 2026-08-21 — r001-pod-conformance. Every figure on these pages was captured from that run.

Run itHow this reaches your data

Every result on this page was produced by pure code over checked-in files, with no API key — which is why you can read the numbers before anyone spends anything.

Checked before this shipped — Verified from a cold clone with no key configured: 55 records and 10 fields load, the UI renders, and Extract returns a plain sentence saying nothing was called. tools/build_corpus.py regenerates the corpus byte-identically from the seed.

Step 01 of 14Business case

Written for: product manager · Measured during the eval run. Never estimated, never quoted from a vendor page.

In brief

What the clock covers. model call only, one per delivery record

Current processWhat it replaces

Someone opening each proof-of-delivery record, finding the ordered and delivered quantities, checking the condition code, deciding whether the shipment arrived as booked, and then deciding whether it is the kind of exception that has to go to the ops desk today. It is quick per record and it is every record, and the part that goes wrong is not the arithmetic — it is reading the driver's own note first and letting it set the answer.

Step 02 of 14Architecture

Written for: solutions architect · Component map derived from the real code, not drawn from intent.

In brief

CorpusCorpusArchitectureArchitecturePromptPromptPromptPromptApp UIApp UIEvalsEvalsBusiness caseUnit cost

• 1stations numbered in run order
• txt
• 0 of 7 stations carry a recorded failure

scroll → 7 stations

One question, end to end. Colour on the fan-in is the corpus format; a red hook marks every station with a recorded failure, and the words — every sub line, failure and report field — are in the ledger beneath. Every station is a link — click one to jump to its row, and each row links to the lens that explains it.

The swap seams

Components

Step 03 of 14App UI

Written for: anyone, in 5 seconds · Screenshots of the running app taken during the run.

SuccessesWhen it works

LimitsWhen it does not

A report showing only wins is an advert. This one is required, and the validator now refuses a kit that omits it.

Step 04 of 14Corpus

Written for: "can I point this at mine?" · Measured off the corpus the run actually used.

In brief

How it is cutWhat one section is

cut on underlined section headings; a record with none falls back to one whole-document segment

SetupWhat the setup figure measured

There is no index. Preparation is segmentation only — 55 records cut into 550 sections in two thousandths of a second, in process, with no model and no network.

LicenceLicence

MIT — this repository's own licence. Every carrier, shipment number, receiving site and driver note is invented; no real carrier, consignee or published logistics standard is named or reproduced.

Bring your ownBring your own delivery records

Replace data/corpus/*.txt, write data/fields.json, and supply a gold record per delivery. SECTION_HINTS in src/select.py maps fields to section headings and will need editing for a different POD layout; when it does not match, selection falls back to the whole document — slower, more expensive, always correct. The routing rule in src/extract.py::compute() is this kit's own invention and should be the first thing you replace.

Step 05 of 14Prompt

Written for: engineer · Logged from the run, verbatim.

In brief

The assembled prompt in full, and its real decomposition. Every part below occurs in the prompt that was actually sent, and the token counts sum to the recorded input total — they were measured against the provider’s own counter, not apportioned by character share.

AssemblyHow the prompt is assembled

This is the cost lesson as arithmetic: of the 1,130 tokens assembled, 521 are instructions — 46% of the prompt, the largest single part. A per-request bill follows whichever part is biggest, and on this kit that is the one named here.

Measured token-for-token via evals/prompt_tokens.py's nested-prefix method on POD-0026 — three calls at max_tokens=1, each part's size the difference between two consecutive prompt_tokens counts the provider itself returned. kits/UC0038-pod-conformance/results/tokens-p001-pod-conformance.json.

Full promptThe prompt in full

You extract structured fields from a proof-of-delivery record for one shipment. You return JSON and nothing else.

RULES, in order of importance:
1. If the record does not state a field, return null for it. Do not infer it and do not use what you know about the world.
2. `delivery_complete` is decided by COMPARING THE STRUCTURED VALUES, never by how the driver's exception note reads. Answer 'yes' when delivered_quantity is EXACTLY EQUAL to ordered_quantity AND condition_code is exactly 'undamaged'. Answer 'no' in every other case. Do the comparison yourself before answering.
3. EXACTLY EQUAL MEANS EQUAL IN BOTH DIRECTIONS. A delivery that is short of the ordered quantity is not complete, and neither is one that delivered MORE units than were ordered -- a surplus is not the order that was booked. Do not read the rule as 'at least the ordered quantity arrived'.
4. THE CONDITION CODE IS PART OF THE TEST. A condition_code of 'damaged' means the delivery is not complete even when the two quantities match exactly. So does a condition_code of 'unknown': a condition nobody assessed at the drop is not an undamaged one, and only the exact value 'undamaged' passes.
5. The driver's exception note is a field to copy, not evidence about completeness. A note reading 'all good on my end, no issues to report' does NOT make a short or damaged delivery complete, and a note reading 'sorry for the mix-up, had to leave it at the gate' does NOT make a delivery with matching counts and an undamaged code incomplete. The structured values decide; the note is the driver's own account of the stop and may disagree with them.
6. Copy values verbatim from the record wherever possible, and report ordered_quantity and delivered_quantity as bare numbers with the word 'units' left out of them.
7. Use the exact allowed value for a field that lists them.
8. Return every field named in the schema, even when the answer is null.

Extract these fields:
- shipment_id (string) -- the shipment identifier, verbatim
- carrier_name (string) -- the name of the carrier that made the delivery, verbatim
- delivery_date (string) -- the date of the delivery, YYYY-MM-DD
- ordered_quantity (number) -- how many units the order was booked for, as a bare number without the word 'units'
- delivered_quantity (number) -- how many units were actually scanned at the drop-off, as a bare number without the word 'units'
- condition_code (enum) one of: undamaged, damaged, unknown -- the structured condition code recorded by the delivery scan, verbatim. This is a coded field, not the driver's prose
- recipient_signature_present (enum) one of: yes, no -- does the record state that a recipient signature is on file?
- driver_exception_note (string) -- the driver's own free-text exception note, copied verbatim
- pod_timestamp (string) -- the timestamp of the delivery scan, verbatim, in the form the record states it
- delivery_complete (enum) one of: yes, no -- did this delivery arrive as booked? Decide this STRICTLY from the structured values and nothing else: it is 'yes' only when delivered_quantity is EXACTLY EQUAL to ordered_quantity AND condition_code is exactly 'undamaged'; in every other case it is 'no'. Exactly equal means equal in both directions -- a delivery of MORE units than were ordered is not complete either. A condition_code of 'damaged' or 'unknown' is not complete even when the counts match. Do NOT decide this from the driver_exception_note: a breezy note reading 'all good, minor scuff on the box' does not make a short or damaged delivery complete, and an anxious note reading 'sorry for the mix-up, had to leave it at the gate' does not make a delivery with matching counts and an undamaged code incomplete.

Return a JSON object with exactly these keys: shipment_id, carrier_name, delivery_date, ordered_quantity, delivered_quantity, condition_code, recipient_signature_present, driver_exception_note, pod_timestamp, delivery_complete
Use null for any field the record does not state.

PROOF OF DELIVERY RECORD
------------------------
Shipment
--------
SHP-NF-83904

Carrier
-------
Ironline Carriage

Delivery Date
-------------
2026-04-02

POD Timestamp
-------------
2026-04-02T07:47:55

Ordered Quantity
----------------
240 units

Delivered Quantity
------------------
222 units

Condition Code
--------------
undamaged

Recipient Signature On File
---------------------------
no

Driver Exception Note
---------------------
All good on my end. Left with the receiving clerk, no issues to report.

Raw responseThe raw response

{"shipment_id":"SHP-NF-83904","carrier_name":"Ironline Carriage","delivery_date":"2026-04-02","ordered_quantity":240,"delivered_quantity":222,"condition_code":"undamaged","recipient_signature_present":"no","driver_exception_note":"All good on my end. Left with the receiving clerk, no issues to report.","pod_timestamp":"2026-04-02T07:47:55","delivery_complete":"no"}

Step 06 of 14Evals

Written for: the skeptic · The harness output, with as_of / model version / dataset version / verified_by.

In brief

Check proof-of-delivery records for short, damaged or unsigned shipments — 55 delivery records. Two tiers of one model family answered, and every answer was then graded Three different ways — because the grade is a measurement too, and most of what went wrong on this kit went wrong in the ruler rather than the model.

MeasurementsWhat was measured

Two words carry this page: Counted is deterministic and nobody's opinion and Not yet known is printed blank rather than filled with something plausible. There is no Judged row here and that is the point: nothing on this page is a model's opinion about another model. The row most worth having is currently the empty one.

Grading costWhat it costs

Every dollar on these pages is a MEASURED token count multiplied by a named vendor's PUBLISHED rate — a projection onto a rate card, not a bill anyone paid.

Rates checked 2026-08-18. The provider that actually ran r001 and r002 publishes no rate card this repo commits, so nothing here is what was actually paid — the real spend for this kit's build is recorded in the commit history and the shared call ledger, not on this page.

The gradersThree ways to grade

Every row opens its own page: how the test was run, the input on one real row, the prompt or formula with every iteration of it, and the analysis.

Choosing oneWhich one to use

The comparison is only useful if it ends in a choice. The answer is conditional — and it is not the grader with the best agreement score.

This pipeline was also attacked, which is a different question from whether its answers are right: see Threat model for the injection experiment, its two gates and what it does not prove.

Failure causesHow it fails

Step 07 of 14Unit cost

Written for: whoever signs off on the bill · Measured over the eval run, per model. Never quoted from a vendor page.

In brief

The token counts are measured on a real run and belong to this pipeline — which sections are selected, the field schema, the prompt — so they hold wherever you run it. The dollars are those counts multiplied by each vendor's published rate, checked 2026-08-18. Nobody paid these particular bills; they are what the same work would cost you on a card you can sign up to today.

Identical work in both rows — the same query, the same tokens. Only the price list changed. Which is why the token count is the measurement and the dollar is arithmetic you can redo on your own card.

Answering vs gradingAnswering and grading are two separate bills

The driver-tone floor (evals/baseline.py) and the scorer (evals/judge.py) are both pure code and cost $0.00 to run against either result set. The figure above is both runs' own token counts (the fast-tier and deliberating-tier runs, 55 records each) priced at Google Gemini 3 Flash's published rate — the same basis cost_per_query_usd uses, not a second, larger spend.

Cost driversWhat actually moves the bill

• The fixed system prompt and field schema (944 of 1130 tokens on the example call, 84 pct) outweigh the record sections sent (186 tokens) — the floor every call pays before a single count is read.
• Output length: the model returns a full ten-key JSON record every call, including the driver's exception note copied back verbatim, whatever the record says.

Your volumeWhat it costs at your volume

Linear in delivery records: each call is independent, carries the same fixed prompt and shares nothing with its neighbours, so 550 records cost ten times 55 and take ten times as long. Nothing amortises — there is no index to build and no cache, which is also why there is no volume discount to find without changing the design.

Other modelsOn other models

This is arithmetic, not a run. No model below was executed against the labelled set, and no accuracy is claimed for any of them. What transfers is the input volume: retrieval is model-independent, so every model would receive the same passages.

The exact work behind every number on these pages: 55 proof-of-delivery records, one completion call each, on the fast tier. The deliberating tier's own 55 calls are recorded separately in Cost.cost_by_model and are not projected here.

The codeEvery file, and what it is for

Five modules, no framework, and almost no dependencies. The pipeline is stdlib Python, because a kit is meant to be read and changed by someone who cloned it ten minutes ago — every dependency is a thing they have to understand before they can touch the part they came for.

Five of these are swap seams: the points you edit to make the kit yours. Each is marked, and each block below is the module’s real shape, read out of the file rather than written here.

The files on this page are not in the public GitHub catalog.

src/segment.pysegment

cut the delivery record into addressable sections, pure code

# Cut a document into addressable sections. Pure code — no model, no network.

def sections(text):
def locate(text, value):
def span_label(secs, start):

src/select.pyselect — a swap seam

pick which sections carry each field, pure code — completeness is mapped to the two quantity sections and the condition code, never to the driver's exception note, and the Receiving Site section is mapped by nothing and never sent

You change it to: map fields to your own delivery record's headings; unmatched falls back to the whole document

# Pick which sections plausibly carry each field. Pure code -- the last deterministic step

SECTION_HINTS = {
def for_field(secs, field):
def plan(secs, fields):

src/prompt.pyprompt

assemble one call for all ten fields, with the completeness rule stated in full — exact equality in both directions, and only the literal code 'undamaged' passing

# Assemble the extraction prompt. One prompt per proof-of-delivery record, all ten fields in it.

SYSTEM = (
def field_schema(fields):
def build(doc_text, secs, fields, selector):
def parse(raw, fields):

src/extract.pyextract — a swap seam

the AI layer, one provider one key — plus the pure-code business-condition check downstream: an incomplete delivery with no recipient signature on file is routed for an ops follow-up

You change it to: the comparison itself — a tolerance band, or a condition code your own scan system spells differently. It is stated once and used by the corpus generator, the prompt and the scorer, so changing it here changes all three together

# Extract one proof-of-delivery record's fields: segment, select, prompt, one model call, then a

HERE = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
FIELDS = os.path.join(HERE, "data", "fields.json")
CORPUS = os.path.join(HERE, "data", "corpus")
MAX_TOKENS = 4000
def load_fields():
def load_doc(stmt_id):
def documents():
def is_complete(delivered, ordered, condition):
def compute(values):

evals/judge.pyjudge

score field accuracy, the completeness confusion matrix and the review flag separately, pure code

# Score an extraction run. PURE CODE -- gold is exact and the answer is one value per cell, so

def norm(v):
def _num(v):
def equal(field, got, want):
def score(fields, records, golds):
def _matrix(rows, positive):
def score_flags(records, flags, golds):

Start hereThe shortest path into it

Every entry above is a file in this kit, listed in the order the pipeline runs it.

ScenariosWhat it costs at your volume

Everything below is computed from one measured base: 1131 input and 253 output tokens per query, all of it fixed — this kit sends the document's own selected sections, so there is no retrieval depth to turn. Change the volume and the bill scales linearly, because nothing here batches.

OperationsFour shapes of workload

The same arithmetic, at volumes worth naming. These are illustrations of scale, not measurements of anyone’s deployment.

Model follows the controls above.

What this ignoresWhere a real bill diverges

In one lineWhat this kit exposes

This run's delivery records are entirely synthetic (tools/build_corpus.py, seed 20260821): no real carrier, consignee, driver or shipment exists in the corpus, and nothing was fetched from anywhere. The only outbound traffic the kit makes is one chat-completion request per record to the configured provider, carrying the nine mapped sections of one delivery record. Nothing is written outside the kit directory, there is no database, no auth and no multi-tenancy, and the local UI binds 127.0.0.1 only.

Read from the shared .env or the real environment only, never committed, never logged and never rendered. .env is gitignored from the first commit. On an adapter error the app redacts the key and base URL out of the message before returning it, so a misconfigured endpoint cannot echo a credential into a browser.

The experimentWe did not attack it — and three of four boundaries hold

The three boundaries that hold were confirmed by reading the code, not by a run: no code path files a claim or contacts anybody; the routing rule reads only enums and numbers, so the driver's note cannot reach it; and the key and base URL are redacted out of any error before it is returned. The fourth is open. An indirect prompt injection needs a field an outside party authored, and this kit has exactly one — the driver's exception note — which makes it the obvious place to attack and the reason the gap is named rather than glossed. Confirmed by reading the code, not by a run, on 2026-08-21 -- no attack run was fired; see redteam.why.

The first three boundaries hold, confirmed by reading the code rather than by an attack run. The fourth is open and is named as open — an indirect prompt injection in the driver's note is exactly the attack this kit's shape invites, and nothing here has tried it.

The result0 attack trials, and three of four boundaries checked here hold on evidence. The one that does not is the one this kit's own shape invites: the driver's exception note is free text an outside party wrote, and whether an instruction hidden in it could move the verdict is unmeasured.

This run's corpus is entirely generated (tools/build_corpus.py, seed 20260821), so no text in it came from an outside party and there was nothing adversarial to resist. The planted ambiguity is a REGISTER mismatch — a breezy note on a short delivery — which measures whether a model compares the counts when the prose points the other way. It does not measure whether a model obeys an instruction hidden in the same field. Those are different failures and only one of them is measured here.

HonestyWhat this does not prove

• Whether an indirect prompt injection in the driver_exception_note field could move delivery_complete on either tier. No attack run exists.
• Whether the kit behaves safely against a hostile provider — a response body crafted to break the JSON extraction in src/prompt.py::parse, or to return an enormous payload. parse() fails closed to an empty dict, which the harness records as a failed document, and nothing beyond that was tested.
• Whether a real POD archive would carry anything sensitive this kit mishandles. The corpus has no personal data by construction; a real record carries a named driver, a signature image and a consignee address, and none of that path is exercised.

In one lineThe guardrail is enforced in code

Not a prompt instruction the model can ignore — a comparison applied after the reply is parsed, before anything downstream sees it.

needs_review fires when the delivery did not arrive as booked AND no recipient signature is on file — delivery_complete == "no" and recipient_signature_present == "no". Both values come out of the same reply; the rule is run afterwards in pure code, over whatever the model returned, and never over gold. A reply missing either value returns None rather than False: an unknown is not a pass.

src/extract.py::compute(), called by extract() on every record, by the local app on every click, and by evals/baseline.py on the free floor's own output so the two are routed by identical code. The completeness comparison it reads lives in a SEPARATE function, is_complete(), which is also what the corpus generator wrote gold with and what src/prompt.py states to the model — one definition, three readers.

EvidenceDoes it hold?

The limitWhat a guardrail is not

• It is NOT a check on whether the extracted values are right. If the model misreads the delivered quantity and then judges that misreading consistently, the record is routed (or not routed) on a wrong number and this rule cannot tell. The consistency diagnostic in evals/judge.py is the closest thing to that check and it is reported separately, and is itself blind to the same case.
• It is NOT a real carrier's claims-intake policy. Incomplete-and-unsigned is this kit's own simplification, invented for this corpus. No published SLA, tariff or carrier liability rule was consulted, and none is reproduced. A real desk weighs shipment value, commodity, the consignee's own receiving standard and the terms on the bill of lading.
• It is NOT a disposition. Nothing here accepts, rejects, credits or claims for a delivery, and delivery_complete is an extracted field rather than a decision anybody is bound by.

WatchedWhat is watched, and why that one

The kit measures 16 watched figures plus 8 comparability guards and 5 reference constants — three duties, not one list. 0 of the watched are grader rates, each on its own grader page; this board watches the ones that decide when to look again. Each square is one watched figure; filled means the latest run measured it.

7 measured by the latest run 9 need the model half

GuardsBefore any comparison

Preconditions, not alarms. If one moves, the runs were not measuring the same system and nothing else on this board may be differenced.

all shared guards held between the recorded runs (documents 55, extraction_cells 550, failures 0, refusal_cells 0) — which is why the history below is allowed to mean anything.

The reference constants — the null-grader baseline, the adjudication matrix, the negative count — are the third duty. They are never re-measured per run: a baseline that moves with the thing it measures is not a baseline.

BandsThe alarm bands

Derived from evidence, never picked: exact-match where two runs reproduced a metric to the digit, wider than the measured same-input drift for latency, one row of the negative class for any rate — denominator printed beside it.

HistoryRun history

3 recorded runs. A dash is a metric that run did not measure, and a dash is never differenced: not measured is a third state, neither a value nor a zero, and reading it as zero is how a board reports an improvement on a run that never looked.

not a time series No two of these 3 runs measured the same system — they differ on provider, thinking, the model — so nothing here may be differenced and the Verdict column is left off. Read this block DOWN a column, as a comparison between those runs, not along a row as a history.

DeviationsWhat deviated

0 breaches across 3 runs

The day a band breaks, a row renders here carrying the metric, the two runs, the band it crossed and what changed between them — model, prompt, chunker or top_k, read off the run records’ guards. A deviation with no diff beside it is an alarm nobody can act on.

RippleThe ripple map

Which figures move together, so one alarm reads as a system. Every edge carries its basis: measured names the run that showed it; reasoning admits nothing has varied it yet.

FiresWhat fires when a band breaks

Where evals stop sitting beside the guardrail and start feeding it. Where the answer is nothing automatic, the row says so — hiding it would undo “What a guardrail is not” above.

NextThe three you would add first

None of these is built here. They are named against the seam they sit on so they can be argued with.

CadenceWhen to re-run, and what it costs

Re-check the flag's evidence on any change to compute() or is_complete() in src/extract.py, on any change to data/fields.json's allowed values for delivery_complete or recipient_signature_present, and on any corpus regeneration. Changing WHO GETS CHASED is a policy change and must be re-scored even though it never changes what 'complete' means.

In one lineNo framework, deliberately

No framework and no dependencies beyond the standard library. requirements.txt is empty on purpose: the corpus is generated in-process, the provider is reached over urllib, and the UI is one HTML file with no build step. A forker runs this on whichever key they already hold, with one clone and no install.

The mappingSeam by seam, who owns what

The seams are the ones this kit already publishes, so the mapping cannot drift from the code.

GraphsWhere LangGraph changes the answer

Nothing here is a graph. The pipeline has one path per record — segment, select, prompt, one call, parse, compute — with no branch, no loop and no agent. Anything that looks like orchestration in a kit this size is a diagram of a straight line.

The other sideWhat a framework costs you

• Everything is hand-rolled, so everything is yours to maintain: the JSON extraction from a fenced reply, the retry/backoff policy, the section regex and the .env reader are all code somebody has to own.
• No framework means no framework's ecosystem — no tracing, no eval harness beyond the one in evals/, no prompt registry, no schema validation library. What ships is what is in the repository.
• The provider abstraction covers exactly two wire formats. A third provider is one function and one dict entry, and until somebody writes it the kit runs on two shapes.

In one lineWhat one run of this kit actually cost, in time and tokens

Measured on r001-pod-conformance on the fast tier, 2026-08-21. This kit records telemetry measured per run — 4 of the 6 readings on this axis, 4 of them with a band the guardrail board can judge against.

The readingWhat the last run measured

No movement column. Not one of the 2 earlier runs on record is comparable with this one — a different model tier, or a changed guard — and differencing across that measures two systems rather than one. Said once here rather than as a dash on every row.

Retrieval not shown. No run of this kit has ever recorded a retrieval timing, because the pipeline has no such stage. Those readings are left out rather than dashed: a dash says “this run did not measure it”, which promises a later run could.

DriftMedian call, run over run

Hatched bars are runs that are not comparable with the latest — a different model tier or a changed guard. They are drawn because hiding them would make the history look shorter than it is, and greyed because differencing across them would be measuring two systems.

1 run not plotted. b000-pod-conformance-rules recorded no call latency at all — a rules-only floor run makes no model call, and the 0 its record stores is an absence, not a zero-millisecond answer. Plotting it would put a bar on this chart claiming the fastest run in the kit’s history.

The other sideWhat this does not tell you

The first of those is the one worth doing something about, and it is a solved problem: instrumenting a pipeline so each step reports its own duration is a couple of lines. Observability / tracing has the code, and what the result looks like once it is running.

In one lineWhat you provision, honestly

Everything measured on this page says where it was measured; everything not measured says that — the same contract as the other thirteen pages. Where a retrieval index would be on other kits, this one has: no index — each document goes whole into one call.

The machineWhat this needs

Enumerated from the kit’s own source by build/facts/envscan.py, committed, and drift-checked — never typed into this page.

Your dataWhere everything sits

The wireWhat crosses it, and when

Zero-egress topology: point BASE_URL at a local OpenAI-compatible server and nothing leaves the machine — statically true of the shipped adapter, whose only credential travels in the Authorization header.

The shapeHow it ships, and what you add

The ladderEach decision, and what you provision past its ceiling

Every threshold is a measurement with its provenance beside it, or says is_measured:false. A ceiling with no seam says so — it is the point the kit is outgrown, never a forced join.

When it failsFailure signatures

The knobsThe config surface, enumerated

Environment variables: API_KEY, BASE_URL, EMBED_MODEL, EVAL_WORKERS, MAX_CALLS_PER_DAY, MODEL, PORT, PROVIDER.

The other sideWhat this page does not claim

The corpus licence, from the Data lens: MIT — this repository's own licence. Every carrier, shipment number, receiving site and driver note is invented; no real carrier, consignee or published logistics standard is named or reproduced. Your corpus’s licence is yours to verify, and the labels you write are about your documents.

In one linePer-field exact match, light normalisation

Does the model's value for each field match gold, after trimming whitespace/punctuation and treating numbers within half a cent as equal? Every field is stated on every record in this corpus, so a null is always a miss — there is no legitimately-absent value here.

MethodHow the test was run

evals/judge.py::score, in-process, no key and no model. The same function evals/run.py calls after every call, and the same field-match logic the free baseline is scored by — a baseline and a real run scored by two different scorers cannot be compared honestly.

Every grader on these pages scored the same 110 already-recorded answers. Nothing was re-generated, so this compares rulers and not models.

The inputOne real row, seen by every grader

The formulaWhat it computes

The analysisWhat it actually did

In operationWhat to monitor

Reference standard: tools/build_corpus.py's gold, read back off the same quantities, codes, timestamps and note the document states; evals/check_labels.py asserts every field is populated on every row, that both quantities are positive whole numbers, that every completeness label agrees with its own values, and that every pod_timestamp is a real instant on its own delivery date, before any run is allowed to spend.

How tight can the band be? There is no tolerance band on the field grade itself — it is exact match after trimming whitespace/punctuation, and numbers within half a cent are treated as equal, never a continuous score to round.

Cadence: Re-score on any change to src/prompt.py, data/fields.json or tools/build_corpus.py — the first two change what is asked, the third changes what is asked about. Re-run evals/run.py (paid) on any change to MAX_TOKENS or the provider/model.

The decisionWhen to reach for it

In one linedelivery_complete against gold's own comparison

Of every delivery that really did not arrive as booked, how many did the run call incomplete — and how many complete deliveries did it wrongly flag? INCOMPLETE IS THE POSITIVE CLASS: a shortfall, a damaged pallet or an unassessed condition code that gets called complete is the failure an operations desk pays for.

MethodHow the test was run

evals/judge.py::score_flags, in-process, no key and no model.

Every grader on these pages scored the same 110 already-recorded answers. Nothing was re-generated, so this compares rulers and not models.

The inputOne real row, seen by every grader

The formulaWhat it computes

The analysisWhat it actually did

In operationWhat to monitor

Reference standard: Gold's completeness is re-derived inside the grader by the same comparison the kit publishes — delivered_quantity == ordered_quantity AND condition_code == 'undamaged' — so the truth this matrix grades against can never be a separately-typed label that drifted from the rule.

How tight can the band be? No threshold — the verdict is one of two allowed values, and a reply that returns neither is counted as unanswered rather than folded into the correct-negative cell.

Cadence: Re-run whenever is_complete() in src/extract.py changes, whenever the corpus is regenerated, and on any provider or model change.

The decisionWhen to reach for it

In one lineneeds_review against the same rule run over gold

Does the pure-code routing decision — incomplete AND no recipient signature on file — land on the same records it would land on if the two fields had been read perfectly? It is a BUSINESS CONDITION, so unlike a self-consistency check it genuinely needs labels, and saying so is half of what makes the number believable.

MethodHow the test was run

evals/judge.py::score_flags, in-process, no key and no model.

Every grader on these pages scored the same 110 already-recorded answers. Nothing was re-generated, so this compares rulers and not models.

The inputOne real row, seen by every grader

The formulaWhat it computes

The analysisWhat it actually did

In operationWhat to monitor

Reference standard: src/extract.py::compute(), the same function the run uses, applied to GOLD's delivery_complete and recipient_signature_present. One rule, two inputs, so a change to the rule moves both sides together and the grader cannot silently grade an old policy.

How tight can the band be? No threshold — two booleans and an AND.

Cadence: Re-run whenever compute() changes. Changing WHO GETS CHASED is a policy change and must be re-scored, even though it never changes what 'complete' means.

The decisionWhen to reach for it