osrs-codex/tools/validate.py
Claude 671d025d9f fix(flip-candidates): release the physical GE slot after a completed flip via closed -> free (governor 10471)
A successful sell followed sell-filled-awaiting-collect -> closed, and
'closed' had no outgoing transition while the scheduler left closed
slots untouched — every GE slot could perform ONE flip and then became
permanently unusable (no rotations, breaks, or longer sessions).

Fix (finding-by-finding, governor comment 10471):
1. 'closed' is no longer a sink. The completed cycle's ledger is
   reconciled and ARCHIVED at close, and the physical slot is released
   to free via the explicit archival transition closed -> free
   (flip-candidates.yaml slot_transitions, whens, scheduler_iteration,
   a9 archive step, summary, risk, sources). The scheduler pass
   dispatches the release (SLOT_DISPATCH['closed'] = 'release'), so a
   closed slot is only ever transient.
2. A deliberately closed cancelled sell releases identically:
   sell-cancelled-awaiting-collect -> closed -> free (whens updated).
3. Committed two-rotation regression (validator, wired into
   ge_lifecycle_errors for every flipping entry): slot_reuse_violations
   walks one physical slot through free -> buy-open ->
   buy-filled-awaiting-collect -> ready-to-sell -> sell-open ->
   sell-filled-awaiting-collect -> closed -> free -> buy-open (second
   buy in the SAME slot), validating every ledger frame against both
   conservation gates; archived_ledger_preserved_after_second_buy
   proves the rotation-1 archived ledger survives rotation 2 bit-for-bit
   (no loss, duplication, or mutation). Negative fixtures prove a
   closed sink or a closed that does not release to free FAILS.
4. ge_lifecycle_errors now requires closed == ['free'] exactly and
   rejects any machine where a state has no transitions.
5. Worked example: t5/t8 walk the release, t8b runs rotation 2 in the
   same physical slots.

All prior fixtures preserved (tax x4, async-slot scheduler,
collection-box, limit-window, side-aware conservation x25) — now 35
selftest checks.

Evidence: python3 tools/validate.py -> 2 entries, 0 errors, 1
pre-existing warning; --selftest -> 35 PASS, 0 errors; git diff
--check clean.
2026-08-29 03:11:20 -05:00

909 lines
48 KiB
Python

#!/usr/bin/env python3
"""OSRS Codex validator — the gate for every codex/**.yaml entry.
Runs:
1. JSON-Schema validation (schemas/entry.schema.json)
2. Cross-field consistency checks the schema can't express:
- state_machine `next`/transition targets must be declared states
- `initial` must be a declared state
- duplicate action ids / step names
- status:verified entries with `verified:` older than --stale-days
=> WARNING (entry should be re-verified, not dropped)
- `best_practice` lines whose source starts with "researched"
=> WARNING (citation TODO — file a follow-up issue)
- GE-margin entries (tags include flipping and/or merchanting —
the unambiguous margin signals): the 2% GE convenience fee is
rounded DOWN to whole coins per item and capped at 5,000,000 GP
per item, so net_per_cycle MUST equal
income - min(floor(0.02*income), 5_000_000) - cost.
Fractional-tax arithmetic (0.98*sell - buy) and uncapped tax
(6M on a 300M sale) are errors.
NOTE: the generic `ge` tag alone does NOT gate this rule — a
training/processing/procurement entry may legitimately carry
`ge` while its income/cost model has nothing to do with flip tax.
- GE-flip lifecycle (flipping/merchanting entries): the 4h buy-limit
ledger (limit_used / limit_resets_at) is DISTINCT from offer
expiry — a limit-window reset never cancels an offer and cancelling
never renews the limit; every slot is an independent asynchronous
sub-state machine (canonical set: free / buy-open /
buy-filled-awaiting-collect / buy-cancelled-awaiting-collect /
ready-to-sell / sell-open / sell-filled-awaiting-collect /
sell-cancelled-awaiting-collect / ready-to-requote / closed) driven
by a global scheduler pass that never touches unrelated slots;
ge_cancel moves filled items, refunded CASH, unsold units and
proceeds to the GE collection box for a separate ge_collect;
per-slot conservation is SIDE-AWARE (buy: quoted units =
filled + still_open + cancelled_unfilled and acquired items =
filled ONLY — the unfilled/cancelled quantity is cash, never
refunded units; sell: quoted = sold + still_open + returned_unsold)
with phase gates (a full buy cannot sell before collection, a full
sell cannot close before proceeds collection);
closed has ONE outgoing transition — the archival release
``closed -> free`` — so a completed cycle's physical GE slot
returns to free for the next flip instead of stranding forever
(governor 10471); the two-rotation reuse regression proves a slot
completes buy/collect/sell/collect, archives the first ledger, and
places a second buy in the same slot without loss, duplication, or
mutation of the archived record;
economy.ceiling_gp_per_hour is the TYPED economic ceiling and
0 <= gp_per_hour_low <= gp_per_hour_high <= ceiling.
- Regression fixtures (--selftest): a fractional-tax flipping
entry must FAIL; an above-cap high-value flip with the
correctly capped-tax net must PASS; a non-margin entry carrying
only `ge` plus a non-flip tag must PASS; plus per-slot scheduler
dispatch, conservation, buy-limit and nesting fixtures.
3. Global id uniqueness across the whole codex tree.
Exit codes: 0 = clean (warnings ok), 1 = validation errors.
Usage: python3 tools/validate.py [--stale-days 30] [--selftest]
--selftest: run the GE floor-tax regression fixtures only (no tree scan)
"""
import datetime
import json
import math
import re
import sys
from pathlib import Path
import jsonschema
import yaml
ROOT = Path(__file__).resolve().parent.parent
SCHEMA = json.loads((ROOT / "schemas" / "entry.schema.json").read_text())
CODEX = ROOT / "codex"
STALE_DAYS = 30
def load_entries():
out = []
for p in CODEX.rglob("*.yaml"):
if p.name.startswith("_"): # dataset files (e.g. _f2p-alch-catalog.yaml)
continue
out.append(p)
return sorted(out)
MARGIN_TAGS = {"flipping", "merchanting"}
def ge_floor_tax_error(rel, data):
"""GE floor-tax gate for a single entry.
Applies ONLY to the unambiguous margin tags (flipping/merchanting).
The generic `ge` tag is not proof that income_per_cycle is a taxed
GE resale: a training/processing/procurement entry may carry `ge`
while modeling unrelated income/cost, so it must not be gated by
this rule. Returns an error message, or None when the entry is not
gated or its floored-tax arithmetic is correct.
Wiki: "The 2% tax rounds down to the nearest whole number, so items
sold for below 50 coins have a tax of less than one coin" — per-item
net MUST use the floored tax, never 0.98*sell - buy. The fee is also
capped at 5,000,000 GP per item (per-item cap, documented in the
flip-candidates best_practice/risk), so tax = min(floor(0.02*sell),
5_000_000). Owner decision 2026-08-28: one global formula for all
entries (option A), no per-entry tax-model discriminator.
"""
tags = {str(t).lower() for t in data.get("tags", [])}
if not tags & MARGIN_TAGS:
return None
eco = data.get("economy") or {}
if not all(k in eco for k in ("income_per_cycle", "cost_per_cycle", "net_per_cycle")):
return None
sell, buy = eco["income_per_cycle"], eco["cost_per_cycle"]
tax = min(math.floor(sell * 0.02), 5_000_000)
expected = sell - tax - buy
if eco["net_per_cycle"] != expected:
return (
f"{rel}: net_per_cycle {eco['net_per_cycle']} != capped floored-tax "
f"model {expected} (sell {sell} - min(floor({sell}*0.02), 5_000_000)"
f"={tax} - buy {buy}); "
f"GE tax rounds down per item and is capped at 5,000,000 GP per item "
f"— fractional 0.98*sell - buy and uncapped tax are invalid")
return None
# --- GE-flip lifecycle gates (owner guidance 2026-08-28, governor 10323 + 10390) ---
# Canonical per-slot sub-states: each GE slot is an INDEPENDENT asynchronous
# machine driven by a global scheduler pass; a pass never closes or advances
# a slot it did not touch. A COMPLETED offer leaves its assets in the GE
# collection box until ge_collect: fully filled buys and sells each get an
# explicit *-filled-awaiting-collect state (governor 10390 finding 2).
SLOT_SUB_STATES = [
"free", # no offer, nothing held
"buy-open", # resting buy offer (partial fills keep resting)
"buy-filled-awaiting-collect", # buy FULLY filled — units sit in the GE collection box until ge_collect
"buy-cancelled-awaiting-collect", # aborted buy: bought units + refunded CASH in the box
"ready-to-sell", # units in hand (ge_collect'ed), sell edge positive
"sell-open", # resting sell offer (partial sells tracked)
"sell-filled-awaiting-collect", # sell FULLY filled — proceeds sit in the box until ge_collect
"sell-cancelled-awaiting-collect", # aborted sell: proceeds + unsold units in the box
"ready-to-requote", # collected units/refunds; requote or abandon
"closed", # completed cycle: ledger reconciled + ARCHIVED; the physical slot is released to free via the closed -> free archival transition (governor 10471)
]
# Dispatch table: the pure action a slot in a given sub-state needs this pass.
SLOT_DISPATCH = {
"free": "quote",
"buy-open": None, # resting; fill/abort decided by policies
"buy-filled-awaiting-collect": "collect",
"buy-cancelled-awaiting-collect": "collect",
"ready-to-sell": "sell",
"sell-open": None, # resting; fill/abort decided by policies
"sell-filled-awaiting-collect": "collect",
"sell-cancelled-awaiting-collect": "collect",
"ready-to-requote": "requote",
"closed": "release", # archival release: closed -> free (governor 10471)
}
def scheduler_dispatch(sub_state):
"""The pending action for one slot in one scheduler pass, or None.
A None (resting) slot is left completely untouched — the pass must not
close, advance, or mutate it. This is the property the worked-example
regression proves. A 'closed' slot is NOT resting: its completed
cycle's ledger was archived at ge_collect time and the pass releases
the physical slot to free ('release' = the closed -> free archival
transition), so the slot can run another flip instead of stranding
forever (governor 10471). Resting states are buy-open / sell-open
only.
"""
return SLOT_DISPATCH.get(sub_state)
# Side-aware GE bucket names (governor 10390 finding 4).
# BUY: the player commits CASH; acquired items are `filled` ONLY — the
# unfilled/cancelled quantity is an unspent-CASH refund, never "refunded
# units". SELL: the player holds units; sold units yield proceeds.
BUY_UNIT_BUCKETS = ["filled", "still_open", "cancelled_unfilled"]
SELL_UNIT_BUCKETS = ["sold", "still_open", "returned_unsold"]
# Physically impossible co-occurrences per side (cancel is atomic: a live
# offer has no remainder once cancelled).
IMPOSSIBLE_UNIT_PAIRS = {
"buy": [("still_open", "cancelled_unfilled")],
"sell": [("still_open", "returned_unsold")],
}
def conservation_violations(record):
"""Side-aware per-slot GE conservation (governor 10390 finding 4).
record: {
'side': 'buy'|'sell', # REQUIRED — the ledger is
'quoted': {item: qty}, # side-aware
**unit buckets (side-specific): # {item: qty}
'cash_quoted': int, # BUY: cash committed
'spent'/'still_committed'/'refunded': int, # BUY cash partition
'proceeds_in_box'/'cash_collected': int, # SELL proceeds
'sub_state': str, # phase gate input
}
BUY invariants:
- unit partition: quoted == filled + still_open + cancelled_unfilled
per item;
- ACQUIRED ITEMS == filled ONLY: any sell-side unit bucket (sold /
returned_unsold) or a 'refunded' unit bucket on a buy record is a
violation — the unfilled/cancelled quantity is CASH (phantom
"refunded units" guard);
- cash partition: cash_quoted == spent + still_committed + refunded
(the unspent remainder of a buy is the cash refund);
- no item may be both still_open and cancelled_unfilled (cancel is
atomic).
SELL invariants:
- unit partition: quoted == sold + still_open + returned_unsold per
item;
- no item may be both still_open and returned_unsold (cancel is
atomic); sold + still_open IS legal (partial sell rest).
Shared: no negative quantities.
"""
out = []
side = record.get("side", "buy")
if side not in ("buy", "sell"):
out.append(f"record side {side!r} must be 'buy' or 'sell'")
return out
buckets = BUY_UNIT_BUCKETS if side == "buy" else SELL_UNIT_BUCKETS
foreign = {"filled", "still_open", "cancelled_unfilled", "sold",
"returned_unsold"} - set(buckets)
for b in sorted(foreign):
if record.get(b):
out.append(
f"{side} ledger uses {b} units — invalid for side {side}: "
f"buy-side unfilled/cancelled quantity is CASH "
f"(refunded), never {b} units; sell-side units partition "
f"into sold/still_open/returned_unsold")
if side == "buy":
# A unit map under a CASH field is the phantom-items guard: a buy's
# unfilled/cancelled quantity is an unspent-CASH refund, never
# "refunded units" (governor 10390 finding 4).
for c in ("cash_quoted", "spent", "still_committed", "refunded"):
if isinstance(record.get(c), dict):
out.append(
f"buy ledger carries {c} as units — the unfilled/"
f"cancelled quantity of a buy is CASH (refunded), "
f"never {c} units (phantom items)")
for b in buckets:
for item in record.get(b, {}):
q = record[b][item]
if q < 0:
out.append(f"item '{item}' bucket {b} is negative ({q})")
quoted = record.get("quoted", {})
for item, qq in quoted.items():
if qq < 0:
out.append(f"item '{item}' quoted is negative ({qq})")
total = sum(record.get(b, {}).get(item, 0) for b in buckets)
if total != qq:
out.append(
f"item '{item}' quoted {qq} != "
f"{' + '.join(buckets)} ({total})")
in_b = {b for b in buckets if item in record.get(b, {})}
for a, b in IMPOSSIBLE_UNIT_PAIRS[side]:
if (a in in_b and b in in_b
and record.get(a, {}).get(item)
and record.get(b, {}).get(item)):
out.append(
f"item '{item}' appears in both {a} and {b} — "
f"physically impossible (a cancelled offer has no "
f"live remainder)")
if side == "buy":
cash_fields = ["cash_quoted", "spent", "still_committed", "refunded"]
for c in cash_fields:
if isinstance(record.get(c), (int, float)) and record[c] < 0:
out.append(f"{c} is negative ({record[c]})")
cash_vals = {c: record.get(c) for c in cash_fields}
if all(isinstance(v, (int, float)) for v in cash_vals.values()):
if cash_vals["cash_quoted"] != (
cash_vals["spent"] + cash_vals["still_committed"]
+ cash_vals["refunded"]):
out.append(
f"buy cash quoted {cash_vals['cash_quoted']} != spent "
f"{cash_vals['spent']} + still_committed "
f"{cash_vals['still_committed']} + refunded "
f"{cash_vals['refunded']} — the unspent remainder of a "
f"buy is the CASH refund")
else:
for c in ("proceeds_in_box", "cash_collected"):
if record.get(c, 0) < 0:
out.append(f"{c} is negative ({record[c]})")
return out
def slot_phase_violations(record):
"""Phase gates (governor 10390 finding 4): assets may not be reused or
a slot closed before the GE collection box is withdrawn.
- 'buy-filled-awaiting-collect': filled must equal quoted (fully
filled) and NOTHING may have been sold yet — a full buy that sells
before ge_collect is a violation (owned units cannot leave the
collection box early).
- 'sell-filled-awaiting-collect': sold units require proceeds_in_box
> 0 — proceeds arrive in the box as units fill and only leave via
ge_collect.
- 'closed': nothing may remain uncollected — proceeds_in_box must be
0 and no cancelled/returned units may still be in the box (a full
sell may not close before proceeds collection).
"""
out = []
sub = record.get("sub_state")
side = record.get("side", "buy")
sold = sum((record.get("sold") or {}).values())
filled = sum((record.get("filled") or {}).values())
quoted = sum((record.get("quoted") or {}).values())
if side == "buy" and sub == "buy-filled-awaiting-collect":
if sold:
out.append(
"buy-filled-awaiting-collect slot has sold units — a full "
"buy cannot sell before ge_collect")
if filled != quoted:
out.append(
f"buy-filled-awaiting-collect slot is not fully filled "
f"(filled {filled} != quoted {quoted})")
if sub == "sell-filled-awaiting-collect" and sold and not record.get("proceeds_in_box"):
out.append(
"sell-filled-awaiting-collect slot has sold units but no "
"proceeds_in_box — proceeds arrive in the collection box on "
"fill and leave only via ge_collect")
if sub == "closed":
if record.get("proceeds_in_box"):
out.append(
"closed slot still has proceeds_in_box — a full sell may "
"not close before proceeds collection (ge_collect first)")
cancelled = sum((record.get("cancelled_unfilled") or {}).values())
returned = sum((record.get("returned_unsold") or {}).values())
if cancelled or returned:
out.append(
"closed slot still holds cancelled/returned units — "
"collect them before closing the slot")
return out
def buy_capacity_ok(limit_used, limit, qty_requested):
"""The 4h buy-limit ledger (governor 10323 finding 1): the limit resets
4h after the FIRST unit was bought; cancelling an offer does NOT renew
it, so a requote before the reset cannot buy beyond the remaining
capacity. Returns True when limit_used + qty_requested <= limit.
"""
return limit_used + qty_requested <= limit
# Deterministic ledger frames for ONE physical GE slot across TWO full
# flips (governor 10471): the fixture walks a complete buy/collect/
# sell/collect cycle, archives the first ledger at `closed`, releases the
# physical slot to `free`, and opens a SECOND buy in the SAME slot. Every
# frame must satisfy both conservation checks; the archived first-cycle
# record must survive rotation 2 untouched (no loss, duplication, or
# mutation). Prices: Rune arrowtips buy 223 / sell 244 / tax
# min(floor(244*0.02), 5M) = 4 / proceeds 2,440,000 (snapshot 2026-08-27,
# matches the worked example).
_CYCLE_ITEM = "Rune arrowtips"
_CYCLE_FRAMES = [
# (sub_state, ledger) — ledger is side-aware and conservation-clean.
("free", {"side": "buy", "quoted": {}, "filled": {}, "still_open": {},
"cancelled_unfilled": {}, "cash_quoted": 0, "spent": 0,
"still_committed": 0, "refunded": 0}),
("buy-open", {"side": "buy", "quoted": {_CYCLE_ITEM: 10_000},
"filled": {}, "still_open": {_CYCLE_ITEM: 10_000},
"cancelled_unfilled": {}, "cash_quoted": 2_230_000,
"spent": 0, "still_committed": 2_230_000, "refunded": 0}),
("buy-filled-awaiting-collect",
{"side": "buy", "quoted": {_CYCLE_ITEM: 10_000},
"filled": {_CYCLE_ITEM: 10_000}, "still_open": {},
"cancelled_unfilled": {}, "cash_quoted": 2_230_000,
"spent": 2_230_000, "still_committed": 0, "refunded": 0}),
("ready-to-sell",
{"side": "buy", "quoted": {_CYCLE_ITEM: 10_000},
"filled": {_CYCLE_ITEM: 10_000}, "still_open": {},
"cancelled_unfilled": {}, "cash_quoted": 2_230_000,
"spent": 2_230_000, "still_committed": 0, "refunded": 0}),
("sell-open", {"side": "sell", "quoted": {_CYCLE_ITEM: 10_000},
"sold": {}, "still_open": {_CYCLE_ITEM: 10_000},
"returned_unsold": {}, "proceeds_in_box": 0,
"cash_collected": 0}),
("sell-filled-awaiting-collect",
{"side": "sell", "quoted": {_CYCLE_ITEM: 10_000},
"sold": {_CYCLE_ITEM: 10_000}, "still_open": {},
"returned_unsold": {}, "proceeds_in_box": 2_440_000,
"cash_collected": 0}),
("closed", {"side": "sell", "quoted": {_CYCLE_ITEM: 10_000},
"sold": {_CYCLE_ITEM: 10_000}, "still_open": {},
"returned_unsold": {}, "proceeds_in_box": 0,
"cash_collected": 2_440_000}), # ARCHIVED here
("free", {"side": "buy", "quoted": {}, "filled": {}, "still_open": {},
"cancelled_unfilled": {}, "cash_quoted": 0, "spent": 0,
"still_committed": 0, "refunded": 0}), # released: physical slot reusable
# rotation 2: the SAME physical slot opens a second buy (ledger reset;
# the archived rotation-1 record above must remain untouched).
("buy-open", {"side": "buy", "quoted": {_CYCLE_ITEM: 10_000},
"filled": {}, "still_open": {_CYCLE_ITEM: 10_000},
"cancelled_unfilled": {}, "cash_quoted": 2_230_000,
"spent": 0, "still_committed": 2_230_000, "refunded": 0}),
]
# The canonical complete flip cycle: rotation 1 (buy/collect/sell/collect)
# plus the archival release and the start of rotation 2.
_CYCLE_PATH = [f[0] for f in _CYCLE_FRAMES]
# Index of the `closed` frame (archived ledger) and the frames around it.
_CLOSED_IDX = _CYCLE_PATH.index("closed")
def slot_reuse_violations(transitions):
"""Two-rotation slot-reuse regression (governor 10471).
Walks ``_CYCLE_PATH`` through the entry's ``slot_transitions`` and
validates every ledger frame against both conservation gates. A
``closed`` state with no outgoing transition (or one that does not
release to ``free``) strands the physical GE slot after one flip —
the exact defect this regression exists to forbid. Errors are
returned as a list of strings.
"""
out = []
for a, b in zip(_CYCLE_PATH, _CYCLE_PATH[1:]):
if b not in (transitions.get(a) or []):
out.append(
f"slot cycle: transition {a} -> {b} is missing — a completed "
f"flip must end by releasing the physical slot to free "
f"(archival release closed -> free), otherwise the slot "
f"performs one flip and then becomes permanently unusable")
for state, record in _CYCLE_FRAMES:
rec = dict(record, sub_state=state)
con = conservation_violations(rec)
if con:
out.append(f"slot cycle: conservation fails at '{state}': {con}")
ph = slot_phase_violations(rec)
if ph:
out.append(f"slot cycle: phase gate fails at '{state}': {ph}")
return out
def archived_ledger_preserved_after_second_buy():
"""The rotation-1 archived ledger survives rotation 2 intact.
``closed`` archives a deep copy of the completed cycle's ledger; when
the same physical slot is released to free and opens a second buy, the
archived record must be bit-for-bit unchanged (no loss, duplication,
or mutation) while the live ledger is a fresh empty record.
"""
archived = json.loads(json.dumps(_CYCLE_FRAMES[_CLOSED_IDX][1]))
archived_state = _CYCLE_FRAMES[_CLOSED_IDX][0]
live = {k: (dict(v) if isinstance(v, dict) else v)
for k, v in _CYCLE_FRAMES[-1][1].items()} # rotation-2 buy-open ledger
if archived != _CYCLE_FRAMES[_CLOSED_IDX][1]:
return ["slot cycle: archived rotation-1 ledger was MUTATED by rotation 2"]
if archived_state != "closed":
return [f"slot cycle: unexpected archived state {archived_state!r}"]
# The live ledger at rotation-2 buy-open must be a fresh record, not a
# copy of the archived one: quoted cash/units belong to rotation 2.
if live.get("cash_quoted") != 2_230_000 or live.get("proceeds_in_box"):
return ["slot cycle: rotation-2 live ledger is not a fresh record"]
return []
def ge_lifecycle_errors(rel, data):
"""GE-flip lifecycle gates (flipping/merchanting entries only):
- the 4h buy-limit ledger is distinct from offer expiry: the entry must
declare a limit_ledger + an independent quote_age_policy;
- the slot model is per-slot asynchronous: slot_state_machine must
declare exactly the canonical sub-states, valid transitions, a
scheduler_iteration, persisted_state, conservation and break_resume;
'closed' is NOT a sink: it must release the physical slot via the
archival transition closed -> free, and the committed two-rotation
reuse regression (slot_reuse_violations) must pass;
- rates: ceiling_gp_per_hour is the typed economic ceiling and
0 <= gp_per_hour_low <= gp_per_hour_high <= ceiling.
"""
errors = []
tags = {str(t).lower() for t in data.get("tags", [])}
if not tags & MARGIN_TAGS:
return errors
eco = data.get("economy") or {}
# --- typed ceiling vs realized bounds ---
ceil = eco.get("ceiling_gp_per_hour")
if not isinstance(ceil, (int, float)) or ceil <= 0:
errors.append(f"{rel}: flipping entries must declare "
f"economy.ceiling_gp_per_hour (the typed 4h buy-limit "
f"economic ceiling, distinct from realized rates)")
return errors
low, high = eco.get("gp_per_hour_low"), eco.get("gp_per_hour_high")
for name, val in (("gp_per_hour_low", low), ("gp_per_hour_high", high)):
if not isinstance(val, (int, float)) or val < 0:
errors.append(f"{rel}: {name} must be a non-negative realized "
f"scenario bound")
if isinstance(low, (int, float)) and isinstance(high, (int, float)) \
and isinstance(ceil, (int, float)):
if not (low <= high <= ceil):
errors.append(
f"{rel}: realized bounds must nest under the ceiling: "
f"0 <= gp_per_hour_low {low} <= gp_per_hour_high {high} "
f"<= ceiling_gp_per_hour {ceil}")
# --- per-slot asynchronous model ---
ssm = data.get("slot_state_machine") or {}
if not isinstance(ssm, dict):
errors.append(f"{rel}: flipping entries must declare a "
f"slot_state_machine block (per-slot asynchronous sub-states)")
return errors
declared = [str(s) for s in ssm.get("sub_states", [])]
if sorted(declared) != sorted(SLOT_SUB_STATES):
errors.append(
f"{rel}: slot_state_machine.sub_states must be exactly the canonical "
f"set: {', '.join(SLOT_SUB_STATES)} (got: {', '.join(declared) or 'none'})")
trans = ssm.get("slot_transitions") or {}
if not isinstance(trans, dict):
errors.append(f"{rel}: slot_state_machine.slot_transitions must be a mapping")
trans = {}
for from_s, tos in trans.items():
if from_s not in SLOT_SUB_STATES:
errors.append(f"{rel}: slot transition from unknown sub-state '{from_s}'")
for to_s in tos if isinstance(tos, list) else []:
if to_s not in SLOT_SUB_STATES:
errors.append(f"{rel}: slot transition '{from_s}' -> unknown "
f"sub-state '{to_s}'")
for state in SLOT_SUB_STATES:
if not trans.get(state):
errors.append(f"{rel}: slot sub-state '{state}' has no transitions "
f"— every state must be able to advance; a closed "
f"slot must release the physical slot via the "
f"archival closed -> free transition (governor 10471)")
if list(trans.get("closed") or []) != ["free"]:
errors.append(f"{rel}: 'closed' must have exactly the archival release "
f"['free'] — a closed slot with no outgoing transition "
f"performs one flip and then becomes permanently "
f"unusable (governor 10471)")
whens = [str(w) for w in ssm.get("slot_transition_whens", []) or []]
for w in whens:
arrow = w.split("->")
if len(arrow) != 2 or arrow[0].strip() not in SLOT_SUB_STATES \
or arrow[1].strip().split(":")[0].strip() not in SLOT_SUB_STATES:
errors.append(f"{rel}: slot_transition_whens entry '{w}' must be "
f"'from -> to: condition' with canonical sub-states")
for field in ("scheduler_iteration", "quote_age_policy", "persisted_state",
"limit_ledger", "conservation", "break_resume"):
if not str(ssm.get(field, "")).strip():
errors.append(f"{rel}: slot_state_machine.{field} must be declared")
# Two-rotation reuse regression (governor 10471): every flipping entry's
# encoded machine must complete buy/collect/sell/collect, release the
# physical slot to free, and support a second buy in the same slot.
errors += [f"{rel}: {e}" for e in slot_reuse_violations(trans)]
return errors
def run_selftest(verbose=False):
"""Regression fixtures for the GE floor-tax + lifecycle gates; returns
error strings."""
errors = []
def check(name, passed):
if verbose:
print(f"SELFTEST {'PASS' if passed else 'FAIL'} {name}")
if not passed:
errors.append(f"selftest: {name}")
fixtures = [
# name, tags, economy, expect_error
("fractional-tax flipping entry", ["flipping"],
{"income_per_cycle": 244, "cost_per_cycle": 223, "net_per_cycle": 16.12}, True),
# 5000 - floor(5000*0.02) - 3000 = 1900 != 2000: this fixture's
# economics deliberately do NOT satisfy the flip equation, so the
# old `ge`-tag scoping would false-positive on it (2000 != 1900).
("non-margin entry with only 'ge' + 'processing'", ["ge", "processing"],
{"income_per_cycle": 5000, "cost_per_cycle": 3000, "net_per_cycle": 2000}, False),
("correctly-floored flipping entry", ["flipping", "merchanting"],
{"income_per_cycle": 244, "cost_per_cycle": 223, "net_per_cycle": 17}, False),
# sell 300,000,000 -> uncapped floor tax is 6,000,000; the per-item
# cap makes it 5,000,000, so the correct net is 1,000,000. The old
# uncapped rule demanded net=0 and rejected this valid entry.
("high-value flip above the 5,000,000 GP per-item tax cap",
["flipping"],
{"income_per_cycle": 300_000_000, "cost_per_cycle": 294_000_000,
"net_per_cycle": 1_000_000}, False),
]
for name, tags, eco, expect_error in fixtures:
rel = f"selftest/{name}"
got_error = ge_floor_tax_error(rel, {"tags": tags, "economy": eco}) is not None
passed = got_error == expect_error
if verbose:
print(f"SELFTEST {'PASS' if passed else 'FAIL'} {name}")
if not passed:
errors.append(
f"selftest: '{name}' expected "
f"{'an error' if expect_error else 'no error'} but got "
f"{'an error' if got_error else 'no error'}")
# --- per-slot scheduler dispatch: the worked example as a regression ---
# Governor 10323 finding 3 + owner: reproduce the simultaneous
# partial-buy / full-buy-fill / no-fill-buy case and prove no
# unrelated slot is closed or advanced. Under governor 10390 finding 2
# a FULL buy fill goes to buy-filled-awaiting-collect (assets sit in
# the GE collection box) — the pass dispatches ge_collect, it never
# lets the slot sell directly.
slot_states = {"arrowtips": "buy-open", # partial fill 4,000/10,000 — resting
"iron": "buy-filled-awaiting-collect", # full fill — ge_collect this pass
"cosmic": "buy-open"} # no fill, edge positive — resting
dispatched = {name: scheduler_dispatch(s) for name, s in slot_states.items()}
ok = dispatched == {"arrowtips": None, "iron": "collect", "cosmic": None}
check("worked example: partial-buy + full-buy-fill + no-fill slots dispatch only the iron slot to ge_collect; unrelated slots stay untouched",
ok)
check("resting slots (buy-open) are never closed or advanced by a pass",
dispatched.get("arrowtips") is None and dispatched.get("cosmic") is None)
check("a fully filled buy must pass through buy-filled-awaiting-collect (ge_collect) before any sell",
scheduler_dispatch("buy-filled-awaiting-collect") == "collect"
and scheduler_dispatch("sell-filled-awaiting-collect") == "collect"
and scheduler_dispatch("ready-to-sell") == "sell")
# --- side-aware conservation invariants (finding 4) ---
# BUY balanced snapshot mirroring the worked example t1: 10,000 Rune
# arrowtips quoted @ 223; 4,000 filled (units in the GE collection
# box), 6,000 still open on the resting offer; cash 2,230,000 =
# spent 892,000 + still_committed 1,338,000 + refunded 0.
buy_partial = {
"side": "buy",
"quoted": {"Rune arrowtips": 10_000},
"filled": {"Rune arrowtips": 4_000},
"still_open": {"Rune arrowtips": 6_000},
"cancelled_unfilled": {},
"cash_quoted": 2_230_000, "spent": 892_000,
"still_committed": 1_338_000, "refunded": 0,
}
check("conservation: BUY side — partial fill splits units into filled + still_open and cash into spent + still_committed (partition sums hold, zero violations)",
conservation_violations(buy_partial) == [])
buy_cancelled = dict(buy_partial, still_open={},
cancelled_unfilled={"Rune arrowtips": 6_000},
still_committed=0, refunded=1_338_000)
check("conservation: BUY side — a cancelled partial buy carries filled units + refunded CASH (still_open -> cancelled_unfilled + refunded), zero violations",
conservation_violations(buy_cancelled) == [])
phantom_rec = dict(buy_cancelled, refunded={"Rune arrowtips": 6_000})
check("conservation: BUY side — 'refunded UNITS' on a buy is a phantom (unfilled/cancelled quantity is CASH): violation",
any("CASH" in v or "invalid for side" in v
for v in conservation_violations(phantom_rec)))
sold_on_buy = dict(buy_partial, sold={"Rune arrowtips": 4_000})
check("conservation: BUY side — a full buy that sells before ge_collect (sold units on a buy ledger) is a violation",
any("invalid for side" in v for v in conservation_violations(sold_on_buy)))
buy_cash_bad = dict(buy_partial, still_committed=1_380_000)
check("conservation: BUY side — cash partition mismatch (spent + still_committed + refunded != cash_quoted) is a violation",
any("cash quoted" in v for v in conservation_violations(buy_cash_bad)))
# SELL balanced: 10,000 sold @ 157, proceeds 1,570,000 in the box.
sell_full = {
"side": "sell",
"quoted": {"Iron bar": 10_000},
"sold": {"Iron bar": 10_000},
"still_open": {}, "returned_unsold": {},
"proceeds_in_box": 1_570_000, "cash_collected": 0,
}
check("conservation: SELL side — full sell partitions sold + still_open + returned_unsold, proceeds in the box (zero violations)",
conservation_violations(sell_full) == [])
sell_partial = dict(sell_full, sold={"Iron bar": 4_000},
still_open={"Iron bar": 6_000},
proceeds_in_box=628_000)
check("conservation: SELL side — a partial sell legitimately spans sold + still_open (zero violations)",
conservation_violations(sell_partial) == [])
sell_dup = dict(sell_full, sold={"Iron bar": 6_000},
still_open={"Iron bar": 3_000},
returned_unsold={"Iron bar": 1_000})
check("conservation: SELL side — units both in a live offer AND returned (impossible pair) is a violation",
any("both still_open and returned_unsold" in v
for v in conservation_violations(sell_dup)))
over_rec = dict(buy_partial, filled={"Rune arrowtips": 7_000})
check("conservation: over-reported buckets break the quoted partition (sum 13,000 != quoted 10,000)",
any("quoted 10000 !=" in v for v in conservation_violations(over_rec)))
neg_rec = dict(sell_full, sold={"Iron bar": -1})
check("conservation: a negative bucket is a violation",
any("negative" in v for v in conservation_violations(neg_rec)))
# --- phase gates (finding 4: assets cannot be reused/closed pre-collect) ---
check("phase gate: a full buy that sells before ge_collect (buy-filled-awaiting-collect with sold units) is rejected",
any("cannot sell before ge_collect" in v
for v in slot_phase_violations(dict(buy_partial, sub_state="buy-filled-awaiting-collect",
filled={"Rune arrowtips": 10_000}, still_open={},
sold={"Rune arrowtips": 10_000}))))
check("phase gate: a full sell that closes before proceeds collection (closed with proceeds_in_box > 0) is rejected",
any("before proceeds collection" in v
for v in slot_phase_violations(dict(sell_full, sub_state="closed"))))
check("phase gate: a sell-filled-awaiting-collect slot with sold units but no proceeds_in_box is rejected",
any("no proceeds_in_box" in v
for v in slot_phase_violations(dict(sell_full, sub_state="sell-filled-awaiting-collect",
proceeds_in_box=0))))
check("phase gate: a closed slot that has been fully collected (no proceeds_in_box, no cancelled/returned units) passes",
slot_phase_violations(dict(sell_full, sub_state="closed",
proceeds_in_box=0, cash_collected=1_570_000)) == [])
# --- 4h buy-limit ledger vs offer expiry (finding 1) ---
check("buy limit: a requote within the remaining capacity is allowed (4,000 used of 10,000 -> 6,000)",
buy_capacity_ok(limit_used=4000, limit=10000, qty_requested=6000))
check("buy limit: a requote beyond the remaining capacity is blocked (6,001 > 6,000 remaining)",
not buy_capacity_ok(limit_used=4000, limit=10000, qty_requested=6001))
check("buy limit: a cancelled offer does NOT renew the limit (10,000 used of 10,000 -> any requote blocked before reset)",
not buy_capacity_ok(limit_used=10000, limit=10000, qty_requested=1))
# --- structured lifecycle gates (findings 1-5 + governor 10390) ---
margin_tags = ["flipping"]
ssm_ok = {
"sub_states": SLOT_SUB_STATES,
"slot_transitions": {
"free": ["buy-open"],
"buy-open": ["buy-filled-awaiting-collect",
"buy-cancelled-awaiting-collect"],
"buy-filled-awaiting-collect": ["ready-to-sell", "free"],
"buy-cancelled-awaiting-collect": ["ready-to-sell", "ready-to-requote", "free"],
"ready-to-sell": ["sell-open"],
"sell-open": ["sell-filled-awaiting-collect",
"sell-cancelled-awaiting-collect"],
"sell-filled-awaiting-collect": ["closed"],
"sell-cancelled-awaiting-collect": ["ready-to-requote", "closed"],
"ready-to-requote": ["buy-open", "sell-open", "free"],
"closed": ["free"], # archival release: physical slot reusable (governor 10471)
},
"slot_transition_whens": [
"free -> buy-open: quote placed",
"buy-open -> buy-filled-awaiting-collect: offer FULLY filled — units sit in the GE collection box until ge_collect",
"buy-open -> buy-cancelled-awaiting-collect: abort on stale edge (margin no longer positive) or strategy reallocation — never on the 4h limit reset",
"buy-filled-awaiting-collect -> ready-to-sell: ge_collect withdrawn units; sell edge positive",
"buy-filled-awaiting-collect -> free: ge_collect withdrawn; margin gone, slot released",
"buy-cancelled-awaiting-collect -> ready-to-sell: collected, sell edge positive",
"buy-cancelled-awaiting-collect -> ready-to-requote: collected; margin positive and capacity remains",
"buy-cancelled-awaiting-collect -> free: collected, full cash refund, slot freed",
"ready-to-sell -> sell-open: sell offer placed",
"sell-open -> sell-filled-awaiting-collect: sell FULLY filled — proceeds sit in the box until ge_collect",
"sell-open -> sell-cancelled-awaiting-collect: abort on stale edge or strategy reallocation",
"sell-filled-awaiting-collect -> closed: proceeds collected and ledger reconciled + ARCHIVED",
"sell-cancelled-awaiting-collect -> ready-to-requote: collected, requote",
"sell-cancelled-awaiting-collect -> closed: collected, ledger archived, slot closed (exit taken)",
"closed -> free: archival release — completed cycle's ledger stays archived; the physical GE slot is reusable for the next flip",
"ready-to-requote -> buy-open: requoted buy",
"ready-to-requote -> sell-open: requoted sell",
"ready-to-requote -> free: abandoned",
],
"scheduler_iteration": "one pass over all slots; dispatch pending actions only",
"quote_age_policy": "offers rest while the scanned margin stays positive; aborted only on a stale edge or strategy reallocation — the 4h limit reset never cancels an offer",
"persisted_state": "per-slot record: sub-state + side-aware quoted/filled/sold/collected buckets",
"limit_ledger": "limit_used + first_fill_ts + limit_resets_at, independent of offer expiry; cancellation never renews it",
"conservation": "side-aware: buy quoted units = filled + still_open + cancelled_unfilled and acquired items = filled only; sell quoted = sold + still_open + returned_unsold",
"break_resume": "breaks between passes; state persists in the per-slot record",
}
flip_ok = {"tags": margin_tags, "economy": {
"income_per_cycle": 244, "cost_per_cycle": 223, "net_per_cycle": 17,
"ceiling_gp_per_hour": 149500, "gp_per_hour_low": 102600,
"gp_per_hour_high": 138800}, "slot_state_machine": ssm_ok}
check("flip lifecycle: ceiling + nested realized bounds + canonical 10-state slot machine passes",
ge_lifecycle_errors("ok.yaml", flip_ok) == [])
errs = ge_lifecycle_errors("bad.yaml", {"tags": margin_tags, "economy": {
"income_per_cycle": 244, "cost_per_cycle": 223, "net_per_cycle": 17,
"ceiling_gp_per_hour": 149500, "gp_per_hour_low": 160000,
"gp_per_hour_high": 138800}})
check("flip lifecycle: realized low above the ceiling is rejected (bounds must nest)",
any("nest" in e for e in errs))
errs = ge_lifecycle_errors("bad2.yaml", {"tags": margin_tags, "economy": {
"income_per_cycle": 244, "cost_per_cycle": 223, "net_per_cycle": 17,
"gp_per_hour_low": 0, "gp_per_hour_high": 138800}})
check("flip lifecycle: missing typed ceiling field is rejected",
any("ceiling_gp_per_hour" in e for e in errs))
errs = ge_lifecycle_errors("bad3.yaml", dict(flip_ok, slot_state_machine=dict(
ssm_ok, sub_states=["free", "buy-open"])))
check("flip lifecycle: non-canonical sub-state set is rejected",
any("canonical" in e for e in errs))
errs = ge_lifecycle_errors("bad4.yaml", dict(flip_ok, slot_state_machine=dict(
ssm_ok, slot_transitions={"free": ["buy-open"],
"buy-open": ["ready-to-sell",
"buy-cancelled-awaiting-collect"]},
slot_transition_whens=ssm_ok["slot_transition_whens"][:4])))
check("flip lifecycle: a sub-state without transitions (ready-to-sell missing) is rejected",
any("no transitions" in e for e in errs))
errs = ge_lifecycle_errors("bad5.yaml", dict(flip_ok, slot_state_machine=dict(
ssm_ok, slot_transitions=dict(ssm_ok["slot_transitions"], closed=[]))))
check("flip lifecycle: a 'closed' sink (no archival release) is rejected — the physical slot would perform one flip and strand",
any("archival release" in e for e in errs))
# --- slot reuse through rotations (governor 10471) ---
check("slot cycle: the archival release closed -> free is a dispatched scheduler action",
scheduler_dispatch("closed") == "release")
check("slot reuse: one full buy/collect/sell/collect lifecycle completes and the SAME physical slot opens a second buy (two-rotation walk passes)",
slot_reuse_violations(ssm_ok["slot_transitions"]) == [])
check("slot reuse: the rotation-1 archived ledger survives rotation 2 intact (no loss, duplication, or mutation)",
archived_ledger_preserved_after_second_buy() == [])
sink_trans = dict(ssm_ok["slot_transitions"], closed=[])
check("slot reuse: a machine whose closed has no outgoing transition FAILS the two-rotation regression (one flip, then permanently unusable)",
any("permanently unusable" in v
for v in slot_reuse_violations(sink_trans)))
closed_no_free = dict(ssm_ok["slot_transitions"], closed=["ready-to-sell"])
check("slot reuse: a closed that does not release to free FAILS the two-rotation regression",
any("closed -> free" in v
for v in slot_reuse_violations(closed_no_free)))
return errors
def main():
global STALE_DAYS
if "--stale-days" in sys.argv:
STALE_DAYS = int(sys.argv[sys.argv.index("--stale-days") + 1])
errors, warnings = [], []
if "--selftest" in sys.argv:
errors += run_selftest(verbose=True)
print(f"RESULT: selftest, {len(errors)} errors, 0 warnings")
return 1 if errors else 0
# The fix itself is gated: regression fixtures run on every invocation.
errors += run_selftest(verbose=False)
entries = load_entries()
if not entries:
print("ERROR: no codex entries found under codex/")
return 1
ids = {}
today = datetime.date.today()
for path in entries:
rel = str(path.relative_to(ROOT))
try:
data = yaml.safe_load(path.read_text())
except yaml.YAMLError as e:
errors.append(f"{rel}: YAML parse error: {e}")
continue
# 1. schema
v = jsonschema.Draft7Validator(SCHEMA)
for err in sorted(v.iter_errors(data), key=lambda e: list(e.path)):
loc = "/".join(str(p) for p in err.path) or "(root)"
errors.append(f"{rel} [{loc}]: {err.message}")
if errors and errors[-1].startswith(rel):
# schema already broke; cross-checks on a broken doc are noise
continue
eid = data.get("id", rel)
if eid in ids:
errors.append(f"{rel}: duplicate entry id '{eid}' (also in {ids[eid]})")
ids[eid] = rel
# 2. state machine consistency
sm = data.get("state_machine", {})
states = sm.get("states", {})
declared = set(states)
terminals = set(sm.get("terminal", []))
for sname, sdef in states.items():
if sname not in terminals and not sdef.get("next"):
errors.append(f"{rel}: non-terminal state '{sname}' has no 'next'")
for nxt in sdef.get("next", []):
if nxt not in declared:
errors.append(f"{rel}: state '{sname}' next->'{nxt}' not declared")
if sm.get("initial") not in declared:
errors.append(f"{rel}: initial state '{sm.get('initial')}' not declared")
for t in sm.get("transitions", []):
if t["from"] not in declared:
errors.append(f"{rel}: transition from '{t['from']}' not declared")
if t["to"] not in declared:
errors.append(f"{rel}: transition to '{t['to']}' not declared")
for term in sm.get("terminal", []):
if term not in declared:
errors.append(f"{rel}: terminal '{term}' not declared")
# action id/step uniqueness
a_ids = [a["id"] for a in data["actions"]]
a_steps = [a["step"] for a in data["actions"]]
if len(a_ids) != len(set(a_ids)):
errors.append(f"{rel}: duplicate action ids")
if len(a_steps) != len(set(a_steps)):
errors.append(f"{rel}: duplicate action step names")
# GE-margin floor-tax arithmetic (see ge_floor_tax_error): scoped to
# the unambiguous margin tags flipping/merchanting. The generic `ge`
# tag alone must not gate this rule — alchemy/processing/procurement
# entries can carry `ge` with income unrelated to a taxed GE resale.
err = ge_floor_tax_error(rel, data)
if err:
errors.append(err)
# GE-flip lifecycle gates (per-slot async model, 4h limit ledger vs
# offer expiry, typed ceiling vs realized scenario bounds).
errors += ge_lifecycle_errors(rel, data)
# 3. staleness / research TODOs
if data.get("status") == "verified":
try:
vdate = datetime.date.fromisoformat(data["verified"])
age = (today - vdate).days
if age > STALE_DAYS:
warnings.append(
f"{rel}: status=verified but facts are {age}d old "
f"(> {STALE_DAYS}d) — re-verify or drop to status:stale")
except ValueError:
pass
for line in data.get("best_practice", {}).get("equipment", []) + \
data.get("best_practice", {}).get("techniques", []):
if str(line.get("source", "")).startswith("researched"):
warnings.append(
f"{rel}: best_practice '{line.get('item') or line.get('name')}' "
f"carries a 'researched' source — citation TODO, file follow-up")
# report
for w in warnings:
print(f"WARN {w}")
for e in errors:
print(f"ERROR {e}")
print(f"RESULT: {len(entries)} entries, {len(errors)} errors, {len(warnings)} warnings")
return 1 if errors else 0
if __name__ == "__main__":
sys.exit(main())