The business caseThe problem this solves

Credentialing a provider file means reading the PSV finding to judge whether it describes any adverse action, however mildly worded, and checking the license and PSV check dates against the file's own effective date, before the file goes to a credentialing committee. Someone manually reading each credentialing file's PSV finding to judge whether it describes any adverse action, however mildly worded, and checking the license and PSV dates against the file's own effective date before the file goes to a credentialing committee.

The inputThe actual files

The corpus is 55 files, 0.03 MB (txt 55). Plain text, one format, invented rather than fetched — a real credentialing file cannot be published either (see SOURCES.md). Ten fields are chosen because the same ones matter to a credentialing review: who, license status, PSV source and date, and — the field this kit exists to test — whether the PSV finding describes any adverse action, however mildly worded.

Provider
--------
Casey D. Rossi

NPI
---
1985767922

Provider Type
-------------
nurse_practitioner

License Number
--------------
NP-083692

License Expiration Date
-----------------------
2027-06-15

Credentialing Effective Date
----------------------------
2026-08-06

PSV Check Date
--------------
2026-06-21

PSV Source
----------
OIG-LEIE exclusion database

PSV Finding
-----------
No adverse action found. License current and unrestricted.

The outcomeWhat a good result looks like

A ten-field extracted record per file, plus one pure-code computed flag folding three checks together — an expired license, a stale PSV check, or any adverse action found — informational only, never a credentialing or network-participation determination.

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 stack of credentialing files for which ones need a credentialing committee to review an adverse action, an expired license, or a stale PSV check before the file is approved — either tier — they tie at 1.00 recall and precision measured here
  100% review-flag recall on both tiers, against the free severe-keyword floor's 65.71% recall (12 missed review flags, all mild-worded adverse actions).
• Deciding the two tiers on cost or speed when accuracy is identical — the fast tier
  Lower p50 latency (2,500 ms vs 4,268 ms) and fewer output tokens per call (300 vs 315 avg) for identical extraction and flag scores — the smallest tier gap measured across this four-kit series so far.

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-credential-verify. 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 files segment into 495 sections in 0.009 seconds, and the assembled prompt for CR-0007 replays byte for byte — its three part sizes match what run r001 recorded.

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 file

Current processWhat it replaces

Someone manually reading each credentialing file's PSV finding to judge whether it describes any adverse action, however mildly worded, and checking the license and PSV dates against the file's own effective date before the file goes to a credentialing committee.

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 file with none falls back to one whole-document segment so a span still resolves

SetupWhat the setup figure measured

There is no index. Preparation is segmentation only — 55 files cut into 495 sections by src/segment.py, pure code, no model and no key.

LicenceLicence

MIT — this repository's own licence. Every provider, NPI, license number and PSV finding is invented; a real credentialing file cannot be published either — sanction and malpractice history is sensitive personal data tied to a real professional identity.

Bring your ownBring your own files

Replace data/corpus/*.txt, write data/fields.json, and supply a gold record per file. SECTION_HINTS in src/select.py maps fields to headings and will need editing for a different file layout; when it does not match, selection falls back to the whole document — slower, more expensive, always correct.

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 789 tokens assembled, 295 are instructions — 37% 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 subtraction against the provider's own tokenizer, not estimated from characters — see results/tokens-p001-credential-verify.json.

Full promptThe prompt in full

SYSTEM:
You extract structured fields from a provider credentialing file. You return JSON and nothing else.

RULES, in order of importance:
1. If the file does not state a field, return null for it. Do not infer it, do not compute it, and do not use what you know about the world.
2. `sanction_or_adverse_action_found` means the PSV finding describes ANY adverse action against the provider -- a reprimand, consent order, letter of concern, restriction, suspension, revocation or exclusion -- HOWEVER MILDLY IT IS WORDED. A finding that says the license is 'otherwise active' or imposes 'no practice restriction' is still 'yes' if any adverse action is named anywhere in it. Only answer 'no' when the finding describes no adverse action of any kind.
3. Copy values verbatim from the file wherever possible.
4. Use the exact allowed value for a field that lists them.
5. Return every field named in the schema, even when the answer is null.

USER:
Extract these fields:
- provider_name (string) -- the provider's name, verbatim
- npi (string) -- the provider's National Provider Identifier
- license_number (string) -- the provider's license number
- provider_type (enum) one of: physician, nurse_practitioner, physician_assistant -- the provider type stated on the file
- license_expiration_date (string) -- the license's expiration date, YYYY-MM-DD
- credentialing_effective_date (string) -- the date the credentialing file must be current as of, YYYY-MM-DD
- psv_check_date (string) -- the date the primary source verification check was actually performed, YYYY-MM-DD
- psv_source (string) -- the primary source the PSV check queried
- psv_raw_finding (string) -- the PSV finding, copied verbatim
- sanction_or_adverse_action_found (enum) one of: yes, no -- does the PSV finding describe ANY adverse action against the provider -- a reprimand, consent order, letter of concern, restriction, suspension, revocation or exclusion -- however mildly it is worded? A finding that says the license is 'otherwise active' or imposes no practice restriction is still 'yes' if any adverse action is named at all. Only a finding with NO adverse action of any kind is 'no'.

Return a JSON object with exactly these keys: provider_name, npi, license_number, provider_type, license_expiration_date, credentialing_effective_date, psv_check_date, psv_source, psv_raw_finding, sanction_or_adverse_action_found
Use null for any field the file does not state.

CREDENTIALING FILE
-------------------
Provider
--------
Jordan K. Haddad

NPI
---
1838505831

Provider Type
-------------
nurse_practitioner

License Number
--------------
PA-048188

License Expiration Date
-----------------------
2026-08-04

Credentialing Effective Date
----------------------------
2026-03-22

PSV Check Date
--------------
2026-03-19

PSV Source
----------
NPDB self-query report

PSV Finding
-----------
Public reprimand issued 2025-04-12 for recordkeeping violation; license otherwise active and in good standing.

Raw responseThe raw response

{
 "provider_name": "Jordan K. Haddad",
 "npi": "1838505831",
 "license_number": "PA-048188",
 "provider_type": "nurse_practitioner",
 "license_expiration_date": "2026-08-04",
 "credentialing_effective_date": "2026-03-22",
 "psv_check_date": "2026-03-19",
 "psv_source": "NPDB self-query report",
 "psv_raw_finding": "Public reprimand issued 2025-04-12 for recordkeeping violation; license otherwise active and in good standing.",
 "sanction_or_adverse_action_found": "yes"
}

Step 06 of 14Evals

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

In brief

Check a clinician's credentialing file before committee review — 55 files. Two tiers of one model family answered, and every answer was then graded Two 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 gradersTwo 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 severe-keyword 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 files 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 (586 of 789 tokens on the example call, 74%) outweigh the file sections sent (203 tokens) on this example — the floor every call pays before a single PSV fact is read.
• Output length: the model returns a full ten-key JSON record every call regardless of the file's own length.

Your volumeWhat it costs at your volume

Linear in files: each call is independent and self-contained, with no shared context or index to amortise. This run's 55 files cost about $0.071 projected onto Gemini 3 Flash's rate, so ten times the set is about $0.71 on the same rate and the same prompt — arithmetic on the measured per-call rate, not a second run.

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 credentialing files extracted, each scored by pure code against a mechanically-derived gold set. This run (r001-credential-verify, the fast tier) answered all 55 of 55 files with no truncation -- the row this table prices, the same one Cost.cost_by_model[0] uses.

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.

Four 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 file 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

You change it to: map fields to your own file'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

# Assemble the extraction prompt. One prompt per file, 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 computation downstream: three independent checks folded into one review flag

You change it to: this kit's own flat lookback window — replace with your actual accreditation standard's published requirement before trusting the computed flag for anything real

# Extract one file's fields: segment, select, prompt, one model call, then a pure-code

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
PSV_LOOKBACK_DAYS = 180
def load_fields():
def load_doc(stmt_id):
def documents():
def _parse_date(s):

evals/judge.pyjudge

score field accuracy and the flag's recall/precision 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 equal(field, got, want):
def score(fields, records, golds):
def score_flags(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: 781 input and 300 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 files are entirely synthetic (tools/build_corpus.py) -- no untrusted party wrote any file text. In a real deployment a credentialing file arrives from the provider's own application or a PSV vendor's report -- exactly the kind of externally-authored input this kit's architecture reads verbatim and trusts, with no verification step before its text reaches the model. No attack has been tried against this kit; see redteam.why.

Read from the shared .env or the real environment only, never written into the repo, never requested from a reader on any surface. src/app.py's /api/extract handler strips api_key and base_url out of any exception message before it reaches the browser, so a misconfigured key cannot leak into a UI error.

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

An indirect prompt injection needs a field an outside party controls that reaches the prompt. In THIS corpus every file is generated by tools/build_corpus.py, so there is no live untrusted text in the run this report measures -- but a real deployment's PSV findings arrive from a PSV vendor's report or the provider's own application, and that surface has not been attacked. The four gates below are boundaries confirmed by reading the code, not payloads run through it. 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, not by an attack trial. The fourth is the one this run has not tested: sanction_or_adverse_action_found is the field the whole guardrail depends on, and it is also the one field this kit's own span mechanism cannot check, because it is an enum, not a copied value -- no crafted input has ever been run against it.

The result0 attack trials, and three of four boundaries checked here hold in code: no write path exists, the PSV lookback window is a fixed constant the model cannot move, and a misconfigured key cannot leak into a UI error. The fourth -- whether a crafted PSV finding could talk the model into a fabricated 'no' on a real adverse action -- is unmeasured.

This run's corpus is entirely generated (tools/build_corpus.py) -- no file's text was authored by an outside party. A real deployment's PSV findings arrive from a PSV vendor's report or the provider's own credentialing application -- exactly the kind of externally-supplied text this kit's architecture reads verbatim and trusts, with no verification step. Whether a finding crafted to describe a real adverse action in routine-sounding language could talk the model into a false sanction_or_adverse_action_found='no' is unmeasured for this kit.

HonestyWhat this does not prove

• Whether a PSV finding crafted to describe a real adverse action in routine-sounding administrative language could talk the model into a false sanction_or_adverse_action_found='no' -- no red-team run exists for this kit.
• Whether the live app's own /api/extract behaves identically to the registered run under an adversarial file -- both use the same src/extract.py::extract(), but neither has been tested against one.
• Whether a code-level consistency check on sanction_or_adverse_action_found (re-deriving it from the finding text by pure code) would catch a model's fabricated 'no' in practice -- none has been built or 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.

A file is routed for review when any of three checks fires: the license is expired, the PSV check is stale, or an adverse action was found. All three are computed in pure code from the model's own extracted fields -- never asked of the model directly, never overridden by anything it returns.

src/extract.py::compute() -- PSV_LOOKBACK_DAYS and the flag arithmetic are a module-level constant and pure code, read once at import time. Unlike a prompt instruction the model could ignore, this guardrail cannot be talked out of firing by anything in the reply text; it only depends on the model's own extracted dates and its own sanction_or_adverse_action_found value being right in the first place -- a DIFFERENT, prompt-level dependency the flag inherits (see is_not and could_not_verify).

EvidenceDoes it hold?

The limitWhat a guardrail is not

• IT IS NOT A CHECK ON sanction_or_adverse_action_found ITSELF. The flag trusts the model's own answer for that field completely -- nothing re-derives it from the PSV finding text before computing the flag. This run found zero disagreements on that field across 55 files on both tiers (see Eval.scores), which is evidence the prompt's stated 'however mildly worded' rule held up on this corpus, not evidence the guardrail itself checked and cleared anything -- it never looks at the finding text at all.
• It does not guarantee a credentialing decision either way -- the flag is a routing signal for a credentialing committee, never itself a determination. See UI.shots and Data.breaks_on.
• It has not been attacked. Whether a PSV finding crafted to describe a real adverse action in routine-sounding language could evade the flag is unmeasured -- see Security.could_not_verify.

WatchedWhat is watched, and why that one

The kit measures 14 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 7 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, thinking True) — 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

2 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 2 runs measured the same system — they differ on 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 2 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.

Nothing fires yet. Not one of the 6 bands on this kit names a consequence, so there is no row to print. Which is the honest answer to the heading: a band that nothing is wired to is a number somebody has to notice, and this kit has only those. An empty table would have said the same thing while looking like a rendering failure.

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 PSV_LOOKBACK_DAYS in src/extract.py, or to the sanction_or_adverse_action_found rule in src/prompt.py -- either changes what the flag is computed FROM. Re-run evals/run.py (paid) on any change to MAX_TOKENS or the provider/model.

In one lineNo framework, deliberately

No framework and no dependencies beyond the standard library -- requirements.txt is empty, with a comment explaining that the emptiness is load-bearing. The whole extraction decision is three files: src/segment.py, src/select.py and src/adapters/__init__.py.

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 file -- segment, select, prompt, call, parse, compute -- with no branching and no state carried between files. A framework would add an orchestrator to a for-loop.

The other sideWhat a framework costs you

• A different field schema (data/fields.json) or a different review rule needs its own gold set and its own evals/check_labels.py pass -- a framework's own schema layer would not remove that work, only relocate it.
• SECTION_HINTS is a five-minute edit for a new file layout because it is a plain dict, not a configured retriever -- a framework's own chunking/retrieval abstraction would need its own re-tuning pass instead, with its own failure modes to learn.
• Swapping providers is one function and one entry in PROVIDERS (src/adapters/__init__.py) -- a framework's own model abstraction would add a dependency and a version to track for the same one-line change this file already gives away free.

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

Measured on r001-credential-verify 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 1 earlier run 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.

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, 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 provider, NPI, license number and PSV finding is invented; a real credentialing file cannot be published either — sanction and malpractice history is sensitive personal data tied to a real professional identity. 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?

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 generated dates and PSV finding the document states (never a target that seeded them); evals/check_labels.py asserts every field is populated on every row 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, 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 lineReview-flag confusion matrix

Does the run's own pure-code needs_review (folding together the expired-license, stale-PSV and adverse-action checks) match the same computation run over gold's own true values?

MethodHow the test was run

evals/judge.py::score_flags, in-process, no key and no model. Computed from the run's own pure-code needs_review (src/extract.py::compute), never from a second model call — the model never sees or states the flag directly.

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 needs_review, computed the identical way src/extract.py::compute() computes it — from the same generated dates and finding text the document states, using PSV_LOOKBACK_DAYS=180.

How tight can the band be? There is no tolerance band — the flag is a binary confusion matrix (flagged/not flagged against gold), never a continuous quantity to round.

Cadence: Re-score on any change to PSV_LOOKBACK_DAYS in src/extract.py or tools/build_corpus.py, or to data/fields.json's date or finding fields. Re-run evals/run.py (paid) on any change to MAX_TOKENS or the provider/model.

The decisionWhen to reach for it