Daily portfolio check — standalone runbook

CORRECTED 2026-09-09: nothing executes this. The line above said it was "executed by a scheduled task each morning in a fresh session with no memory". No Windows task and no cloud routine runs daily.py or references this file — the only three Covenant tasks are Distill (03:30), Guard (every 2 min) and Trader (09:00). It is reachable by hand through DAILY.bat, RUN.bat and TRADE.bat, which is a different thing from running by itself.

That matters here more than in most places: a runbook that claims to run daily invites the reader to treat its output as current. Everything needed IS in this document, and it still works when run — but somebody has to run it.

Owner holds crypto on Kraken and elsewhere. Nothing here trades. The output is a suggestion the owner acts on by hand, or ignores.

1. Holdings

SYMBOL   QUANTITY      AVG_BUY     NOTE
XLM      <qty>    <avg_buy>
SOL         <qty>  <avg_buy>
WLFI     <qty>      <avg_buy>      only ~110d of history; line is 110d not 200d
XRP       <qty>    <avg_buy>
ADA       <qty>   <avg_buy>
CC       <qty>      <avg_buy>      NOT tradable on Coinbase — price by hand, see §5
CRO      <qty>       <avg_buy>
ONDO      <qty>      <avg_buy>
HBAR     <qty>      <avg_buy>
PEPE   <qty>          <avg_buy>
CASH        <qty>          <avg_buy>

(The quantities and average buys above were the owner's real positions and were redacted from every commit on 2026-09-05; the live table is the gitignored holdings.txt, which daily.py reads. See docs/KNOWN_ISSUES.md issue 15.)

2. Regime lines — recomputed 2026-08-22, and a correction to how they age

daily.py computes the line itself from the window it just fetched, so this table is not an input to it. Keep it for one purpose only: rule 3 needs a previous state to say what changed. Compare against the "state 08-22" column.

symbol	line (200d)	state 08-22	price 08-22	state 08-19	price 08-19
XLM	0.172958	ABOVE	0.198716	BELOW	0.170252	flipped
SOL	81.15230	ABOVE	93.75	ABOVE	85.34
XRP	1.275320	ABOVE	1.4888	BELOW	1.1054	flipped
ADA	0.223836	ABOVE	0.22482	BELOW	0.18702	flipped (0.4% over)
ONDO	0.315602	ABOVE	0.36973	ABOVE	0.34702
CRO	0.066157	BELOW	0.06027	BELOW	0.0488
HBAR	0.084363	BELOW	0.07665	BELOW	0.07058
PEPE	0.00000333615	ABOVE	0.00000399	BELOW	0.00000288	flipped
WLFI	0.075889 (200d) / 0.059041 (110d)	BELOW	0.05879	ABOVE	0.06142	flipped

Computed two independent ways and reconciled: from the verified Kraken series in realdata/deep/ (hashes in IMPROVEMENT_LOG.md D1) and from Coinbase's own 200 settled closes. The two agree to ≤ 0.02% on every symbol — that agreement is the evidence the lines are right, not the fact that a script printed them.

THE CORRECTION — the old freshness rule measured the wrong thing

The previous version of this section said the lines drift "roughly 0.3–0.8% per week" and that the table could be trusted for 14 days. The drift figure is right. The conclusion drawn from it is not, and three days were enough to show why:

	measured over 08-19 → 08-22
line drift	−0.38% to +0.31% — as advertised, negligible
states flipped	5 of 9 (XLM, XRP, ADA, PEPE BELOW→ABOVE; WLFI ABOVE→BELOW)

The line is the slow number; the state is what the rules act on, and it is a comparison — it flips the moment price crosses, however still the line is. XRP moved +34.7% in three days. So: never read a stored state as current. Recompute the line from today's window every run (which daily.py does), and use this table only for the "what changed?" half of rule 3.

WLFI needs two lines, and the reason is not drift

WLFI's line reads +28.8% against the 08-19 table. That is not the average moving; it is the window changing — the old figure was a 110-bar mean and enough history now exists for 200. The mean is strongly window-dependent (110-bar 0.059041, 200-bar 0.075889, 355-bar 0.113809 — all today, same data), so quoting one number without its bar count is meaningless. Both are given above. WLFI is BELOW on every window length, so the flip itself is real — but on the 110-bar convention it is only 0.4% below, which is exactly the "sitting near its line" case rule 3 says is not a signal.

3. Fetching today's prices — RUN THE SCRIPT

python daily.py                 # Coinbase, verified against Kraken
python daily.py --push TOPIC    # and push the phone summary
python daily.py --source kraken # if Coinbase is unavailable

daily.py does everything this section used to describe by hand — the fetch, the four checks below, the forming-bar rule, the cross-venue check, the 20% cap and cash floor, and the circuit breakers — and it holds no key and cannot trade. Prefer it. Doing it by hand is for when the script itself is what you doubt.

CORRECTION 2026-08-22 — the network was never the problem

DAILY_CHECK_CLOUD_BLOCKER.md recorded two scheduled runs (08-20, 08-21) that reported no prices at all, and concluded the cloud sandbox cannot reach a price API: "proxy returns 403 Forbidden at the tunnel for api.exchange.coinbase.com, api.coinbase.com, and api.kraken.com."

Re-measured from the sandbox on 2026-08-22 with plain urllib:

api.kraken.com/0/public/Time                    HTTP 200
api.kraken.com/0/public/OHLC?pair=XXLMZUSD…     HTTP 200
api.exchange.coinbase.com/products/XLM-USD/…    HTTP 200   (23 004 bytes)
api.coinbase.com/v2/prices/XLM-USD/spot         HTTP 200

All four. What actually failed was WebFetch, which raises a one-time URL permission prompt that nobody is present to approve in an unattended run — PROVENANCE_REQUIRED. The runbook sent the reader to a URL, the reader reached for the summarising tool, and the tool asked for a human. daily.py has always used urllib and would have worked on both mornings. So: fetch with urllib from Bash. Never use WebFetch for prices.

The old "do not request the full series in one call" warning belongs to that same summariser (M1/M24): it dropped rows and altered values. A direct urllib call returns the series intact — verified byte-for-byte against Kraken, and the 200-bar means from the two venues agree to ≤ 0.02% (§2).

If you are doing it by hand

https://api.exchange.coinbase.com/products/{SYM}-USD/candles?granularity=86400
    element = [time, low, high, open, close, volume]      close is index 4
https://api.kraken.com/0/public/OHLC?pair={PAIR}&interval=1440&since={epoch}
    element = [time, open, high, low, close, vwap, volume, count]   close is index 4
    pairs: XLM=XXLMZUSD  XRP=XXRPZUSD  everything else is SYM+USD

Verify every window before using it — all four, every time:

consecutive timestamps differ by exactly 86400
no duplicate timestamps
no zero or negative prices
the newest timestamp is today or yesterday — if it is older, the read is broken; refetch, and if it fails again say so rather than reporting the numbers
if the newest bar is today's, drop it — it is still forming

And then the fifth check, which is the only one that catches a window that passes the other four. The 70-day-stale series of PRICE_DATA_INTEGRITY.md was internally perfect: contiguous, no duplicates, all positive. Every check above is internal to one response, so none of them could see it. Compare the last settled close against the other venue: they should agree to well under 1% (measured across all nine symbols on 2026-08-22: worst PEPE 0.292%). If they disagree by more than 1%, do not pick one — report the symbol as unpriced. There is no way to tell which venue is wrong, and a wrong price does not just misprice its own line, it mis-sizes the 20% trim on every other holding. Compare settled closes only; live prices differ between venues legitimately.

4. The rules

20% cap. No single coin above 20% of portfolio value. Over → trim the excess. Also hold a 10% cash floor.
Regime. Above its line: may hold and add. Below: hold, do not add.
Act on changes, not noise. Only flag a coin whose state changed versus §2. A coin sitting near its line will cross repeatedly — that is expected and is not a signal.
Never average down. Do not suggest buying more of anything below its line.
No claim of a profit edge. Ever. See §6.

5. CC (Canton)

Confirmed 2026-08-20 across four Coinbase endpoints: not tradable on Coinbase. Their price page carries the wording "Canton is not tradable on Coinbase" and showed $0.09105. Use $0.09105 as a hand-entered price, label it as hand-entered, and note it may be stale. Do not fetch CC-USD — it 404s.

There are impostor tokens using the Canton name, including a Solana one whose address ends in pump (a memecoin-launchpad marker). Never suggest adding to CC.

6. What to say, and what never to say

Push a short phone summary (under 200 characters): portfolio value, any rule action, any regime change. If nothing changed, say so plainly — "no action, all rules satisfied" is a complete and useful answer, and most days it is the true one.

Never state or imply that these rules predict price or produce profit. Tested properly, no timing edge survived out-of-sample — on 598 bars across ten held assets the bootstrap p-values run 0.47–0.92 and deflated Sharpe is 0.000 everywhere (TRADING_READINESS.md §1b).

CORRECTED 2026-08-22 — the "~3× drawdown reduction" claim in the previous version of this section is not supported. Measured on the deep data, drawdown reduction is 2.1× on SOL, 2.2× on ADA, 1.8× on XRP, 1.7× on PEPE/ATOM/AVAX, 1.4× on CRO — and none at all on XLM (0.9×), NEAR (0.9×) or ONDO (0.6×), where the rule whipsawed 9, 11 and 18 times and lost MORE than simply holding. Three of ten is not a tail. So the honest sentence is: the regime rule is risk control on most of this basket and actively harmful on some of it, and it is never a return edge. Do not assume drawdown protection for an asset it has not been measured on.

7. Hard limits — do not cross these

Never place, prepare, or offer to place a trade. The owner places every order by hand.
Never ask for, store, or use an exchange API key with trade or withdrawal permission. A "Query Funds"-only Kraken key is the maximum, and it lives at C:\Users\<user>\.kraken\credentials — outside the cloud-synced folder.
Never touch, request, or record the Ledger recovery phrase or passcode. There is no legitimate reason to ever type those anywhere.
The synced folder C:\Users\<user>\covenant leaves the machine. Nothing secret goes in it.