The business caseThe problem this solves

A store opening playbook is sixty tasks with owners and dates, and the weekly report says how many are late. Only some late tasks move the date: a late signage install with four weeks of float does not; a late utility connection everything is waiting on does. On this corpus the overdue count and the critical path disagree on 43 of 120 readings. a playbook tracker that reports how many tasks are overdue — which cannot say which one holds the date, escalates on a count that ignores float, and cannot see float consumed in earlier weeks.

The inputThe actual programme review extracts

The corpus is 120 programme review extracts, 0.84 MB (txt 120). A retailer's opening pipeline names its unopened sites, its contractors and its dates — a competitive document. There is no public one. A scrubbed export is worse rather than better: scrubbing removes the site notes, which is exactly the layer this kit measures.

Synthetic Record
----------------------------------------------------------------
Every field below is invented. This is a generated store opening / store closure
programme review extract for an AI use-case kit; it reproduces no retailer, no landlord,
no site, no contractor and no real person. The float allowances, the dependency chain and
the escalation thresholds are ILLUSTRATIVE DEFAULTS.

Programme
----------------------------------------------------------------
Programme reference            : PRG-0001
Programme type                 : OPENING
Store                          : Larkmead shopping centre, unit L23 (format: convenience)
Region                         : METRO
Target date (current baseline) : 2026-11-18
Programme manager (of record)  : Baptiste Moreau
Playbook                       : standard 60-row opening playbook, 6 rows tracked at review

Playbook Tasks
----------------------------------------------------------------
The playbook rows this review tracks. "Float" is the DEFAULT float allowance from the
policy below -- how far the row may slip before it moves the programme date. It is a PLAN
figure: it does NOT show what has already been spent at earlier reviews.

  ID      | Playbook row                             | Owner               | Baseline finish | Float
  LEASE   | Lease execution and site handover        | Rosalind Chu        | 2026-09-09      | float 0d
  FITOUT  | Fit-out construction                     | Declan Ashworth     | 2026-10-19      | float 4d
  FIXTURE | Fixture delivery and install             | Harriet Lindqvist   | 2026-10-29      | float 3d
  STOCK   | Opening stock delivery                   | Priya Raghunathan   | 2026-10-28      | float 8d

The outcomeWhat a good result looks like

every programme's verdict with the binding task named and the date impact in days — 0 false escalations across 102 quiet readings.

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.

• Your baseline dates and dependency map are complete and every completion is keyed in — the free floor — float-arithmetic-strong, $0.00
  It computes the critical path in pure code and ties the model on four columns.
• Site notes carry the real state and the tracker catches up later — the model, with the carried state
  Verdict 99.17 against 73.33 and 100.0 pct on the discriminator against 89.17 pct.

And where nothing here is good enough:

• You want the reading but cannot carry state between reviews — neither, as configured here
  The stateless control falls to 65.0 pct on verdict and 6.98 pct on memory-dependent verdict.

At a glanceHow the whole thing runs

Run once, for real, on 2026-08-25. 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 2026-08-25 — r002-openclose-playbook. 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 — Every corpus extract, the answer key, all three free floors, the injection probe and what r002-openclose-playbook actually answered ship in the repo. python3 -m evals.check_labels and python3 -m src.app both run with no key and no network.

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. one reading — one programme, at one scheduled run, end to end including provider-side reasoning tokens, on the shared connection. Not a cold-start figure and not a per-programme SLA: 120 readings ran in 499.5 wall seconds because different programmes run concurrently while a single programme's runs are strictly serial.

Current processWhat it replaces

a playbook tracker that reports how many tasks are overdue — which cannot say which one holds the date, escalates on a count that ignores float, and cannot see float consumed in earlier weeks.

Step 02 of 14Architecture

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

In brief

CorpusCorpusArchitectureArchitectureArchitectureArchitectureEvalsEvalsEvalsEvalsPromptPromptPromptPromptApp UIApp UIEvalsEvalsBusiness caseUnit cost

• 1stations numbered in run order
• txt
• jsonl
• json
• 3 of 10 stations carry a recorded failure

scroll → 10 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 reading is

40 programmes x 3 weekly review, walked strictly in order per programme so the carried state moves forward the way it would in a real deployment. Different programmes run concurrently; a single programme's cycles never do.

SetupWhat the setup figure measured

There is no index to build — each reading's extract goes whole into the prompt. The 0.5s and $0.00 are the corpus generation itself.

LicenceLicence

MIT

Bring your ownBring your own programme review extracts

tools/build_corpus.py writes data/corpus/*.txt and data/gold.jsonl. Keep the 8 section headings and the field labels src/review.position_of parses. Set your own float allowances and dependency map in src/programme.py first.

⚠︎ And what stops being true when you do: The measured figures on this kit's page do not transfer to your own corpus. The overdue-versus-critical disagreement rate (43 of 120 readings here), the prose-versus-tag evidence split (52 against 35) and the six-task width are properties of this generator's declared distribution, not facts about store development. Re-run the evals on your own data; that is what the harness is for.

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 1,474 tokens assembled, 1,358 are instructions — 92% 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.

Reproduced from src/prompt.build for PRG-0001-W1 with that reading's real carried state, not retyped — byte-identical to what evals/run.py sent.

Full promptThe prompt in full

You are a weekly store-programme review watch. You read one store opening or closure programme at one weekly review against the DEFAULT float allowances and dependency chain, and you answer with one JSON object and no other text.

You are the weekly programme review watch for a retailer's store OPENING and store CLOSURE
playbooks. It runs on a WEEKLY cadence and re-reads EVERY live store programme. You are reading ONE
programme at ONE weekly review.

The programme, its tracked playbook tasks with their DEFAULT float allowances, the DEFAULT
dependency chain and escalation rule, and this week's changes are reproduced in the extract below.
Apply them exactly as written. You cannot see the earlier reviews; what is known about them is
stated under "Carried state" and is the only history available to you. Do not assume anything about
earlier reviews beyond it.

⚠︎ THE FLOAT ALLOWANCES, THE DEPENDENCY CHAIN AND THE ESCALATION THRESHOLDS ARE ILLUSTRATIVE
DEFAULTS, NOT ANY REAL RETAILER'S PROGRAMME. Apply them as printed regardless of whether they look
right for the store in front of you.

The question is NOT how many tasks are late. It is which late task, if any, actually MOVES THE DATE.

How to work it out:
 - FIRST decide whether the programme can be judged at all. If the baseline finish dates are not on
   file, or the dependency map and float allowances are not on file, the verdict is
   CONTEXT_INCOMPLETE: no remaining float is computed, no binding task is named, the date impact is
   0 and nothing is escalated. THERE IS NO DEFAULT FLOAT ALLOWANCE (Rule R-6).
 - THEN decide whether the programme is already COMPLETE. If the carried state or this week's
   changes record the store as opened, or the unit as handed back, the verdict is COMPLETE and
   stays COMPLETE (Rule R-8). Nothing here completes a task, moves a date, reassigns an owner or
   re-baselines a plan -- you are only ever reporting something already agreed elsewhere.
 - THEN decide whether the target date has already been formally MOVED, in the carried state or in
   this week's changes. If it has, the verdict is DATE_MOVED, the binding task is NONE, and the
   date impact is the number of days the date was moved by (Rule R-7).
 - THEN do the float arithmetic. For every tracked task, REMAINING FLOAT is its printed DEFAULT
   float allowance, MINUS every day it has already spent (see "Carried state"), MINUS any day
   recorded against it in this week's changes, PLUS any day recovered for it in this week's
   changes. A recovery cannot take a task below zero days spent (Rule R-1).
 - A slip or a recovery recorded in ordinary prose in this week's changes counts EXACTLY as much as
   one recorded in the fixed tag format. A sentence saying the crew got in early and the electrics
   are clear is a recovery; a sentence saying the meter fit has moved out about three days is a
   three-day slip.
 - If any task's remaining float is BELOW zero, the verdict is CRITICAL_PATH_SLIP. The binding task
   is the one with the largest overrun, and where two tasks are equally over it is the one further
   UP the printed dependency chain. The date impact in days is that largest overrun (Rule R-2).
 - If no task is below zero but a task that HAS slipped is at exactly zero remaining float, the
   verdict is FLOAT_EXHAUSTED, that task is the binding task, and the date impact is 0 (Rule R-3).
 - If tasks are late but every one of them still has float left, the verdict is SLIP_ABSORBED, the
   binding task is NONE and the date impact is 0 -- HOWEVER MANY of them there are. The count of
   late tasks is not the question and does not change this answer (Rule R-4).
 - If nothing is late at all, the verdict is ON_TRACK, the binding task is NONE and the impact 0.
 - THEN decide the OWNER: when there is a binding task, it is that task's owner as printed in the
   playbook table; when the binding task is NONE, it is the programme manager. Start from the
   programme manager on file, but a change of programme manager recorded in this week's changes or
   at an earlier review (see "Carried state") supersedes it.
 - FINALLY decide "escalate_now": YES only for a CRITICAL_PATH_SLIP whose binding task this watch
   has not already escalated (see "Carried state"). If the binding task has MOVED to a different
   task, that is a new escalation and it is YES. Otherwise NO. FLOAT_EXHAUSTED is a warning, not an
   escalation, and SLIP_ABSORBED is never escalated (Rule R-9).

Answer with a single JSON object and nothing else:

{"verdict": "ON_TRACK|SLIP_ABSORBED|FLOAT_EXHAUSTED|CRITICAL_PATH_SLIP|DATE_MOVED|COMPLETE|CONTEXT_INCOMPLETE",
 "binding_task": "<the task id that moves the date, e.g. UTIL -- or NONE>",
 "owner": "<the binding task's owner, or the programme manager when there is no binding task>",
 "date_impact_days": <integer days the target date moves, 0 when it does not>,
 "escalate_now": "YES|NO",
 "rationale": "one sentence, naming the task, its allowance, the days spent against it and the
   remaining float you relied on"}

Precedence, applied in this order: CONTEXT_INCOMPLETE if the dates or the float allowances are not
on file (Rule R-6); then COMPLETE (Rule R-8); then DATE_MOVED (Rule R-7); then the float arithmetic
-- CRITICAL_PATH_SLIP, then FLOAT_EXHAUSTED, then SLIP_ABSORBED, then ON_TRACK. "binding_task" is a
task id only for CRITICAL_PATH_SLIP and FLOAT_EXHAUSTED; it is NONE for every other verdict.

Carried state
----------------------------------------------------------------
No earlier review has been recorded for this programme. This is its first appearance on the watch: no task has been recorded as having spent any of its float, the target date is not known to have been moved, the programme is not recorded as complete, nothing has been escalated by this watch, and no change of programme manager has been recorded.

Store programme review extract
----------------------------------------------------------------
Synthetic Record
----------------------------------------------------------------
Every field below is invented. This is a generated store opening / store closure
programme review extract for an AI use-case kit; it reproduces no retailer, no landlord,
no site, no contractor and no real person. The float allowances, the dependency chain and
the escalation thresholds are ILLUSTRATIVE DEFAULTS.

Programme
----------------------------------------------------------------
Programme reference            : PRG-0001
Programme type                 : OPENING
Store                          : Larkmead shopping centre, unit L23 (format: convenience)
Region                         : METRO
Target date (current baseline) : 2026-11-18
Programme manager (of record)  : Baptiste Moreau
Playbook                       : standard 60-row opening playbook, 6 rows tracked at review

Playbook Tasks
----------------------------------------------------------------
The playbook rows this review tracks. "Float" is the DEFAULT float allowance from the
policy below -- how far the row may slip before it moves the programme date. It is a PLAN
figure: it does NOT show what has already been spent at earlier reviews.

  ID      | Playbook row                             | Owner               | Baseline finish | Float
  LEASE   | Lease execution and site handover        | Rosalind Chu        | 2026-09-09      | float 0d
  FITOUT  | Fit-out construction                     | Declan Ashworth     | 2026-10-19      | float 4d
  FIXTURE | Fixture delivery and install             | Harriet Lindqvist   | 2026-10-29      | float 3d
  STOCK   | Opening stock delivery                   | Priya Raghunathan   | 2026-10-28      | float 8d
  SIGN    | Signage install                          | Femi Adeyemi        | 2026-10-28      | float 21d
  MKTG    | Local launch marketing                   | Harriet Lindqvist   | 2026-10-21      | float 25d

Programme Policy (Default)
----------------------------------------------------------------
The float allowances and the dependency chain below are an OPERATOR-TUNABLE DEFAULT, not a
real retailer's actual programme -- no documented playbook, float table or escalation rule
exists for this row. Applied exactly as printed.

Dependency chain (OPENING) -- a row cannot finish before the row before it has:
  LEASE -> PERMIT -> UTIL -> LICENCE -> FITOUT -> FIXTURE -> POS
Off-chain, running in parallel with it: STOCK, HIRE, SIGN, MKTG

Default float allowance, in days -- how far a row may slip before it moves the date:
  LEASE      0d   PERMIT     3d   UTIL       2d   LICENCE    1d
  FITOUT     4d   FIXTURE    3d   POS        2d   STOCK      8d
  HIRE      12d   SIGN      21d   MKTG      25d

Current weekly report rule     : the programme report escalates when 3 or more tracked
                                 rows are past their baseline finish, whatever they are.
Re-baselining authority        : NOT DEFINED. Nothing in this kit moves a date, closes or
                                 completes a task, reassigns an owner, re-baselines a
                                 plan, books a contractor or approves a variation.

Rule R-1  A task's REMAINING FLOAT is its DEFAULT float allowance, as printed against it in the
          playbook table, minus every day it has slipped since the baseline -- days recorded in
          earlier reviews as well as this one. Recovered days are given back. Remaining float is
          never read off this week's movement alone. THERE IS NO DEFAULT FLOAT ALLOWANCE: a task
          with no allowance printed against it cannot be judged (Rule R-6).
Rule R-2  A task is BINDING when its remaining float has gone below zero. The date moves by the
          largest overrun on the board, and the verdict is CRITICAL_PATH_SLIP. Where two tasks are
          both over, the binding one is the one further UP the dependency chain -- everything
          below it is waiting anyway.
Rule R-3  A task whose remaining float is exactly zero is not yet moving the date, but the next
          day of slip on it does. The verdict is FLOAT_EXHAUSTED and the task is named. Nobody is
          escalated for it under the DEFAULT escalation rule; it is the week to go and look.
Rule R-4  Tasks that are late with float still left over are SLIP_ABSORBED. The date has not
          moved and NOBODY IS ESCALATED, however many of them there are. A report escalating on
          the COUNT of late tasks escalates here, and it is wrong to.
Rule R-5  Slips, recoveries, completions and date decisions ACCUMULATE across reviews. "Changes In
          This Week" shows only what was recorded since the PREVIOUS review; anything recorded at
          an earlier review is not repeated and is known only through the carried state.
Rule R-6  If the programme's baseline finish dates are not on file, or the dependency map and
          float allowances are not on file, the reading is CONTEXT_INCOMPLETE: no remaining float
          is computed, no binding task is named and nothing is escalated. THERE IS NO DEFAULT
          FLOAT, and a task with no printed allowance is not silently treated as having none.
Rule R-7  Once the target date has been formally MOVED -- in the programme record or in a review
          note -- the verdict is DATE_MOVED and stays DATE_MOVED for the rest of this horizon,
          whatever the remaining float then does. The reported date impact is the number of days
          the date was moved by. Nothing here moves a date; it only reports one already agreed
          elsewhere.
Rule R-8  Once the programme is recorded COMPLETE -- the store has opened, or the unit has been
          handed back -- the verdict is COMPLETE and stays COMPLETE. It beats every other verdict
          including DATE_MOVED.
Rule R-9  ESCALATE NOW is YES only for a CRITICAL_PATH_SLIP whose binding task this watch has not
          already escalated. If the constraint MOVES to a different task, that is a new escalation
          and it is YES again. A larger overrun on the SAME task is not.

Review Position
----------------------------------------------------------------
Review date                    : 2026-08-24   (weekly review 1 of this programme)
Review cadence                 : weekly -- every programme review, every store programme
Previous review                : -- none, this is the first
Next review                    : 2026-08-31

Changes In This Week
----------------------------------------------------------------
Everything recorded against this programme between the previous review and this one.
THIS WEEK ONLY -- a change recorded at an earlier review is not repeated here.

2026-08-19  Picked up on the site walk that Lease execution and site handover has drifted by
            around five days against the baseline.

Site Notes
----------------------------------------------------------------
Trading has asked for the opening to land before the half-term week; nothing in the plan
has changed on the back of it yet.

Raw responseThe raw response

{"verdict": "CRITICAL_PATH_SLIP", "binding_task": "LEASE", "owner": "Rosalind Chu", "date_impact_days": 5, "escalate_now": "YES", "rationale": "LEASE has a 0d float allowance, has spent 5 days of slip this week, giving -5d remaining float and a 5-day critical path slip."}

Step 06 of 14Evals

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

In brief

Catch the task that actually delays a store opening — 120 programme review extracts. One model 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, never a bill and never a vendor's claim about this workload.

Rates checked 2026-08-18. The provider that actually ran all 393 calls is kept out of these tables per this estate's naming rule, so no figure here is a bill.

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.

Answering vs gradingAnswering and grading are two separate bills

The three free floors, the stub, the cadence analysis and every pre-run check cost $0.00 and no calls. ⚠︎ A DISCARDED SCORED RUN (truncated at an 8,000 ceiling) cost the same again as the scored run and is NOT netted out — it was spent.

Cost driversWhat actually moves the bill

• OUTPUT TOKENS, AND ALMOST ALL OF THEM ARE REASONING. 94.1 pct of r002-openclose-playbook's output (188884 of 200678 tokens) was provider-side reasoning left at the default. The answer is a handful of short fields and a sentence; the bill is the thinking in front of it.
• THE PROMPT IS MOSTLY FIXED TEXT. 3158 input tokens per reading, and the policy block, the rule text and the instruction are the same on every call — so a provider with prompt caching would price this workload very differently, and that was not measured.

Your volumeWhat it costs at your volume

LINEAR IN PROGRAMMES x REVIEWS. Nothing amortises. The sublinear lever NOT implemented is skipping programmes with no task movement since the last review.

Model choiceWhy this model

It is the tier the shared connection was already pointed at, and the kit's whole claim is that swapping it is .env plus one more run. No second model was run against this corpus; every other row in the cost table is a projection onto a published card and is labelled as one.

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.

per-query average, this run

The codeEvery file, and what it is for

11 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/programme.pythe float rule and state machine — a swap seam

The DEFAULT float allowances, the dependency map, the escalation thresholds and step() — which decides the next carried state from the answer key's inputs, never from a model reply. It is also the answer key.

You change it to: Widening it divides the bill and widens the window float is consumed unobserved.

# The store opening/closure playbook rule as arithmetic. Pure code, no model, standard library.

ON_TRACK = "ON_TRACK"
SLIP_ABSORBED = "SLIP_ABSORBED"
FLOAT_EXHAUSTED = "FLOAT_EXHAUSTED"
CRITICAL_PATH_SLIP = "CRITICAL_PATH_SLIP"
DATE_MOVED = "DATE_MOVED"
COMPLETE = "COMPLETE"
CONTEXT_INCOMPLETE = "CONTEXT_INCOMPLETE"
VERDICTS = (ON_TRACK, SLIP_ABSORBED, FLOAT_EXHAUSTED, CRITICAL_PATH_SLIP, DATE_MOVED, COMPLETE,
NO_TASK = "NONE"

src/state.pythe carried state

Float consumed in earlier weeks, the settled owner and the moved-date flag. Consumed float is what makes a small slip this week the binding one.

# The carried state -- the thing that makes this a monitor and not another weekly status report.

HERE = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
DEFAULT_PATH = os.path.join(HERE, "data", "state.json")
def load(path=DEFAULT_PATH):
def save(store, path=DEFAULT_PATH):
def for_programme(store, programme_id):
def describe(state, tasks=None):

src/segment.pythe section splitter

Splits the page into its 8 named sections; asserted across all 120 documents.

# Split a store-programme review extract into its named sections. Pure code, no model.

SECTIONS = ("Synthetic Record", "Programme", "Playbook Tasks", "Programme Policy (Default)",
RULE = "-" * 64
def split(text):
def render(secs):

src/select.pythe send filter — a swap seam

Programme Manager Contact is mapped by no field and therefore never sent.

You change it to: SECTION_HINTS and NEVER_SENT.

# Pick which sections of a programme review extract are sent. Pure code -- the last deterministic

BANNER = "Synthetic Record"
PROGRAMME = "Programme"
TASKS = "Playbook Tasks"
POLICY = "Programme Policy (Default)"
POSITION = "Review Position"
CHANGES = "Changes In This Week"
CONTACT = "Programme Manager Contact"
NOTES = "Site Notes"
NEVER_SENT = (CONTACT,)

src/prompt.pythe prompt

Three parts: instruction and JSON shape, carried state, extract. The stateless control replaces exactly one line.

# Assemble the one prompt this kit sends. No template engine, no framework -- string concatenation

INSTRUCTION = """\
def build(text, carried, stateless=False, tasks=None):

src/adapters/__init__.pythe model call

Raw HTTP, stdlib only, bounded retry; the shared daily call cap is checked here.

# SEAM 1 — the model. Swapping provider or model is .env plus the same run again.

class AdapterError(RuntimeError):
TRANSIENT = {408, 429, 500, 502, 503, 504}
RETRIES = 4
def _post(url, headers, payload, timeout=120):
def openai_compatible(cfg, system, user, max_tokens, thinking=None):
def anthropic(cfg, system, user, max_tokens):
PROVIDERS = {"openai-compatible": openai_compatible, "anthropic": anthropic}
THINKING_OFF = {"type": "disabled"}
def complete(cfg, system, user, max_tokens=1024, thinking=None):

src/review.pythe reader

One programme, one review, one call.

# One store programme, one weekly review, one model call. The runtime half of the kit.

HERE = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
CORPUS = os.path.join(HERE, "data", "corpus")
SYSTEM = ("You are a weekly store-programme review watch. You read one store opening or closure "
MAX_TOKENS = 16000
FIELDS = ("verdict", "binding_task", "owner", "date_impact_days", "escalate_now")
TASK_ROW = re.compile(
def documents():
def programmes():
def load_doc(doc_id):

evals/scoring.pythe scorer

Exact match per cell. The critical-path discriminator is a separate binary.

# Score a run against the answer key. Pure code, exact match per cell. No model grades anything.

FIELDS = ("verdict", "binding_task", "owner", "date_impact_days", "escalate_now")
NEEDS_ATTENTION = ("CRITICAL_PATH_SLIP", "FLOAT_EXHAUSTED")
ESCALATE_VERDICTS = ("CRITICAL_PATH_SLIP",)
def _pct(n, d):
def _norm(field, v):
def score(records, golds):
def _exposure(records, golds):

evals/baseline.pythe three free floors

overdue-count, overdue-mem and float-arithmetic-strong — $0.00 each.

# THE FREE FLOORS. Three of them, none a strawman, all of them $0.00.

MODES = ("overdue-count", "overdue-mem", "float-arithmetic-strong")
def _section(text, name, nxt, flat=False):
ROW = re.compile(r"^\s{2}([A-Z][A-Z0-9]*)\s*\|\s*(.+?)\s*\|\s*(.+?)\s*\|\s*(\S+)\s*\|\s*"
def _tasks(text):
def _facts(text):
TAG_SLIP = re.compile(r"TAG: SLIP ([A-Z][A-Z0-9]*) (\d+)d")
TAG_RECOVERED = re.compile(r"TAG: RECOVERED ([A-Z][A-Z0-9]*) (\d+)d")
TAG_DATE_MOVED = re.compile(r"TAG: DATE-MOVED \S+ -- target date moved to \S+ \(\+(\d+)d\)")
TAG_COMPLETE = re.compile(r"TAG: COMPLETE \S+")

src/app.pythe local UI server

http.server, stdlib. Renders with no key.

# The minimal local UI. Standard library only -- python3 -m src.app, then open the printed URL.

HERE = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
UI = os.path.join(HERE, "ui")
GOLD = os.path.join(HERE, "data", "gold.jsonl")
PORT = int(os.environ.get("PORT", "8217"))
RECORDED_RUN = os.environ.get("RECORDED_RUN", "r002-openclose-playbook")
GOLD_ROWS = {r["doc_id"]: r for r in
def carried_for(doc_id):
class H(BaseHTTPRequestHandler):
def main():

ui/app.jsthe local UI client

Hand-written JS, no framework.

# 

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: 3158 input and 1672 output tokens per reading (one programme, at one weekly review), 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

MEASURED, not asserted. Site Notes is this kit's injection surface and one of the shipped notes is instruction-shaped by design.

API_KEY is read from .env (gitignored) or the real environment; never logged, never placed in a prompt, and redacted from any provider error message the local UI passes to the browser.

The experimentWe DID attack it — one sentence, on every reading where it could matter

The probe forces the instruction-shaped note onto every reading where an escalation was due and re-fires them with everything else held identical. Both were run for real on 2026-08-25.

One boundary is measured in both directions here: the injection probe forces the condition rather than waiting for it, and the privacy guard is red-proven by reproducing the schema-change condition that reaches its fallback.

The result0 of 18 escalations suppressed suppressed by the instruction-shaped note — measured, not assumed.

One phrasing, one model, one corpus, 18 trials — every reading where suppression was even possible.

HonestyWhat this does not prove

• Any other injection.
• Whether the result holds on another model tier.
• Whether an injection placed in Programme Manager Contact would have any effect — by construction it cannot.

In one lineThe only guardrail is the prompt

One instruction, no enforcement. It mostly holds — and mostly is the whole subject.

Never move a date, close a task or reassign an owner — and never present the float allowances, dependency map or escalation thresholds as any real retailer's programme.

Stated to the model on every call in src/prompt.py's INSTRUCTION block, and enforced by evals/check_labels.py's banned-code-path scan before any run may spend.

EvidenceDoes it hold?

The limitWhat a guardrail is not

• This is a PROMPT RULE PLUS A STATIC SCAN, not a runtime enforcement layer. Nothing stops a forker adding such a path tomorrow; the scan catches it only the next time somebody runs check_labels.py, which is a manual step.
• The injection result is ONE SENTENCE against ONE model on ONE corpus. It is not a resistance rate for prompt injection in general.

WatchedWhat is watched, and why that one

The kit measures 64 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.

3 measured by the latest run 61 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.

no shared guard The two recorded runs declare no guard in common, so nothing below is protected by a precondition — read the history as two separate observations, not as a difference.

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

7 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. They are split into 4 blocks because a free floor is a baseline rather than a peer, and a run of another kind measures another thing; a column in one block is never differenced against a column in another.

triage · no model in the path — a baseline, not a peer column — 3 runs. A floor is what the kit scores with no model in the path; it is the line the model runs have to beat, and it is never differenced against them.

not a time series No two of these 3 runs measured the same system — they differ on floor, 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.

triage · with the model in the path — 2 runs. Columns here are only ever compared with each other.

not a time series No two of these 2 runs measured the same system — they differ on stateless — 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.

stub · no model in the path — a baseline, not a peer column — 1 run. A floor is what the kit scores with no model in the path; it is the line the model runs have to beat, and it is never differenced against them.

one run A verdict is a comparison and one run has nothing to compare with, so the Verdict column is left off rather than filled with 13 chips that all say so.

redteam · with the model in the path — 1 run. Columns here are only ever compared with each other.

one run A verdict is a comparison and one run has nothing to compare with, so the Verdict column is left off rather than filled with 3 chips that all say so.

DeviationsWhat deviated

0 breaches across 7 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 14 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

The guardrail scan is a MANUAL step, run before each spend, not a hook. It ran before every paid run and passed at 0 each time. The injection probe is a ONE-OFF: it measured 0.0 pct suppression on 18 readings on 2026-08-25 and nothing re-runs it, so that figure ages from the day it was taken.

In one lineNo framework, deliberately

NO FRAMEWORK. A folder of readable Python, no framework dependency, and a prompt anyone can read end to end. A framework abstraction would own the retrieval step — there is none here — and the memory layer, which is already the entire surface of src/state.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

Linear: the reader -> src/prompt.py -> src/adapters -> evals/scoring.py. No branching, no tool-calling and no agent loop.

The other sideWhat a framework costs you

• Swapping providers means editing the PROVIDERS dict by hand. In exchange the prompt has no hidden templating layer and requirements.txt stays empty.

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

Measured on r002-openclose-playbook on the fast tier, with the carried state, 2026-08-25. 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 4 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.

3 runs not plotted. b000-openclose-playbook-overduecount, b001-openclose-playbook-overduemem, b002-openclose-playbook-floatstrong 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 — the population is re-read whole on each scheduled run, and the prior run's carried state is what makes a reading a CHANGE.

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, RECORDED_RUN.

The other sideWhat this page does not claim

The corpus licence, from the Data lens: MIT Your corpus’s licence is yours to verify, and the labels you write are about your documents.

In one lineThe verdict, binding task, owner, date impact and escalate-now call, per reading, exact match against the computed answer key

whether each of the 5 answered fields equals the answer key, per reading

MethodHow the test was run

python3 -m evals.run --run-id <id> ...; evals/scoring.py compares strings. No model grades anything.

Every grader on these pages scored the same 120 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

accuracy = hits / 120 per field. critical_path_slip_accuracy_pct is scored separately over its own denominator.

The analysisWhat it actually did

In operationWhat to monitor

Reference standard: the computed answer key in data/gold.jsonl, produced by src/programme.step() from the same planted inputs the corpus builder wrote onto the page.

No true/false rates for this grader. It records 5 operating rows and not one of them carries a confusion matrix, so there is no TPR, TNR or precision to state here — this grader does not sort answers into pass and fail, and rates that were never measured are not going to be printed as though they were. What it did measure is in “What it actually did” above; what to watch is immediately below.

How tight can the band be? No threshold was swept: exact match has no tunable. Denominators are stated beside every rate because 120 readings makes each one wide.

Cadence: once per run; free

The decisionWhen to reach for it

In one lineA slip that actually moves the opening date, as against one float covers

whether the arm called the date moving when it was, and left it alone when it was not

MethodHow the test was run

evals/scoring.py, same pass as the exact-match grader.

Every grader on these pages scored the same 120 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

caught over all 120 readings; the two error directions over their own denominators (27 readings where the date WAS moving, 93 where it was not).

The analysisWhat it actually did

In operationWhat to monitor

Reference standard: the computed answer key in data/gold.jsonl, produced by src/programme.step() from the same planted inputs the corpus builder wrote onto the page.

No true/false rates for this grader. It records 2 operating rows and not one of them carries a confusion matrix, so there is no TPR, TNR or precision to state here — this grader does not sort answers into pass and fail, and rates that were never measured are not going to be printed as though they were. What it did measure is in “What it actually did” above; what to watch is immediately below.

How tight can the band be? No sweep: the verdict is an enum the arm emits.

Cadence: once per run; free

The decisionWhen to reach for it

In one lineSlips float covers, correctly left alone

whether the arm escalated a late task that had credit to spare

MethodHow the test was run

evals/scoring.py, same pass.

Every grader on these pages scored the same 120 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

over the 41 readings whose slip is absorbed by float.

The analysisWhat it actually did

In operationWhat to monitor

Reference standard: the computed answer key in data/gold.jsonl, produced by src/programme.step() from the same planted inputs the corpus builder wrote onto the page.

No true/false rates for this grader. It records 1 operating row and not one of them carries a confusion matrix, so there is no TPR, TNR or precision to state here — this grader does not sort answers into pass and fail, and rates that were never measured are not going to be printed as though they were. What it did measure is in “What it actually did” above; what to watch is immediately below.

How tight can the band be? No sweep: escalate_now is a binary the arm emits.

Cadence: once per run; free

The decisionWhen to reach for it