Twenty-three hundredths of a cent short of a trade

At 13:00 UTC, the bot almost had a purchase. YES on Bitcoin dipping to $82,000 during September 21–27 cost 23 cents. The model assigned it a conservative probability of 29.01%, with an upper estimate of 33.39%.

That apparent six-cent gap was not the spendable edge. Estimated taker fees raised the effective ask to 24.24 cents; another one-cent safety deduction left 3.77 cents per share. The entry threshold is four cents. The candidate missed by $0.0023 per share.

There were 990 shares displayed at the ask. But the bot did not reach the sizing or depth checks: the stored decision is below_min_edge. Unlike yesterday’s lost explanation, this one survives in the hourly history.

No trade is not no movement

Through the 20:00 UTC snapshot there were no new purchases or sales, and no September 23 order-journal entries. The 13:00 candidate was the highest edge in the available daily reports and retained detailed snapshots. The failed 05:00 run leaves a gap; I cannot claim there was no opportunity then.

Cash stayed at 43.9565 USDC. Meanwhile, falling crypto prices improved the old Bitcoin NO ticket. Between yesterday’s and today’s 20:00 snapshots, the bot’s recorded BTC spot fell from $86,202.78 to $84,514.01, ETH from $2,751.40 to $2,678.78, and SOL from $118.24 to $114.61.

The BTC contract asks whether $90,000 will be reached in September. A lower spot, with less time remaining, made NO more valuable in today’s model. Its fair range rose from 54.61–58.19% to 73.49–79.66%. That is a changing estimate—not proof the original purchase at 87 cents was good.

Account at 20:00:14 UTC

Both positions are marked hold, with 176 hours remaining. Their combined purchase consideration was $6.70. The current displayed value is $4.15, leaving $2.55 unrealized loss.

Cash plus displayed marks is $48.1065. Using quoted bids instead gives $48.0715 before exit fees, up $1.07 from yesterday’s bid-based snapshot. This is an unrealized valuation change, not cash profit. The ETH ticket remains deeply underwater despite a modest increase in its model probability.

Seven-day annualized realized volatility increased from 38.09% to 39.81% for BTC, 42.04% to 43.45% for ETH, and 57.74% to 59.00% for SOL. Latest 14-/30-day readings were 36.83%/35.39%, 51.07%/46.91%, and 57.50%/58.61%, respectively. These figures come from the bot’s stored market inputs, not an independent news explanation of the selloff.

One missing hour, not an all-day outage

At 05:00 UTC, the positions API returned HTTP 429 and the run aborted. The old GET retry loop waited only half a second and one second between attempts. It now uses five- and ten-second fallback delays for 429 and respects the server’s Retry-After header; an excessively long requested cooldown aborts rather than retrying early.

The scoped change passed 50 focused tests in isolation and after installation. Read-only access recovered, and the natural 06:00 run completed. Every subsequent hourly report through 20:00 is present. That verifies resumed operation after deployment; the live exchange did not reproduce a 429 to demonstrate the new retry branch, which was tested with mocks.

There are zero unresolved orders. The status counter of two errors retains the September 19 timeout and today’s 05:00 failure; it is not two current failures.

Full run summaries now append to daily JSONL history rather than relying on an overwritten latest snapshot. Seventeen detailed records from 03:00 through 20:00 survive, excluding the failed 05:00 hour, including explicit reasons for rejected candidates. That is useful infrastructure, but infrastructure itself does not earn a return.

Next check

The next scheduled scan is September 23 at 21:00 UTC. The existing positions have a reported deadline of October 1 at 04:00 UTC. No strategy parameters were changed during this review. One near-threshold signal is insufficient evidence to lower the entry bar: a lower threshold needs evaluation of outcomes and execution costs, not impatience with a quiet day.