jjavi — Amsterdam 15 August: evidence package and reproducible ledger

Three immutable snapshots: the original audit contains public activity through 2026-08-15T09:00:49Z and positions received at 09:01:59Z..09:02:05Z. The tape addendum in section 13 was fetched around 11:40Z. The end-to-end addendum in section 14 was fetched around 14:15Z, after the last observed jjavi fill at 12:15:16Z. Mutable API calls were saved as separate raw files rather than overwriting earlier evidence.

Supersession notice: section 13 supersedes the original snapshot's claims that the maker/taker role and disposition of the final 151.8976 NO26 were unknown. Section 14 supersedes the open-portfolio boundary in section 13 using the known terminal result 23°C. Neither addendum rewrites an earlier snapshot as if later fills had already happened.

1. Direct answers

Is the behavior already described in the Yield Calculator PDF?

Yes, at the level supported by this event. NegRisk Coverage Band Yield Calculator v2.pdf now uses the final 149-fill ledger and describes the conditional 21–24°C payout floor, cheap maker ladders as branch completion, and NO26 as a hedge that was monetised to subsidise the YES basket. It also preserves the fee-aware EV boundary and the distinction between one profitable event and proven long-run +EV. It does not pretend that the private bucket-selection signal, target-share tolerance, or exit triggers have been recovered.

Can the complete market tape and price history recover more of the algorithm?

Yes. For this event, trades?eventId=842167&limit=10000&takerOnly=false returned the entire available snapshot in one page: 6,917 participant rows representing 2,708 unique transaction hashes. Comparing it with the default takerOnly=true tape makes the role observable: a jjavi row present in the taker-only result is a taker fill; a jjavi row present only in the all-participants result is a maker fill.

The exact URL with filterType=CASH&filterAmount=1 is not sufficient for this account. At the same snapshot it returned 2,473 taker rows and only 7 jjavi rows, while the unfiltered all-participants tape contained 132 jjavi rows. It removes the micro-fills that characterize the low-price passive strategy and, because takerOnly defaults to true, also removes maker fills.

Is this a successful completed event?

Yes, under the known terminal result 23°C. Total BUY cash was $933.842980, actual SELL cash was $486.069248, and the held YES23 inventory pays $2,018.145100. Including the reconciled 0.007687 NO26 dust gives terminal economic PnL +$1,570.379055, or +168.16% relative to total BUY cash.

Gamma had not yet marked the event resolved at the final collection time, so no REDEEM row was present. That is a settlement-status distinction, not payout uncertainty under the supplied 23°C result. One highly profitable event still does not establish the strategy's long-run +EV; that requires a terminal, fill-verified sample across many events without survivorship filtering.

What was the purpose of the cheap YES fills?

They are best explained as cheap completion of a selected payout band, not as independent lottery tickets. Across YES21..YES24, the bot ended with 1,695–2,041 shares per bucket even though the fee-aware entry VWAPs ranged from 0.0050 to 0.2497. If any one of these four mutually exclusive buckets wins, its payout alone exceeds the full $933.842980 gross BUY cash spent on the event. After monetising NO26, the minimum terminal profit within this band is $1,247.391913.

The guarantee is conditional, not market-wide: it holds only if the winner is in 21–24°C. The final portfolio still loses about $447.77 at 25°C, 26°C, 20°C/below, 29°C, or 30°C/higher; YES27 covers most but not all of the net cost, while YES28 is a separately covered cheap branch. The fills prove the resulting payout geometry. They do not expose the private signal that selected these particular buckets or prove that equalising shares was an explicit hard-coded target.

What exit behavior is now observed for NO 26°C?

Three price tiers and their execution roles are observed:

1. A block sale of exactly 300 shares (50.435% of the original NO inventory) at quoted price 0.765021; fee-aware cash price 0.756032833; return over fee-aware entry +22.501%. This was a taker exit.
2. Six fills totalling 142.926665 shares at quoted/cash price 0.90; return over fee-aware entry +45.828%. All six were maker fills.
3. Four fills totalling 151.89 shares at 0.86; return over fee-aware entry +39.346%. All four were maker fills and reduced the position to dust.

No fixed time step or fixed share decrement is observed. The 54 / 10 / 44 / 5.06 / 10 / 19.866665 fragments at 0.90, and 50 / 5.1 / 4.42 / 92.37 at 0.86, are fills, not evidence of ten independent decisions.

The observed policy is therefore closer to: take roughly half immediately at a profitable taker price, offer the rest passively at 0.90, then lower the passive offer to 0.86 and finish. The exact placement/cancellation times and the condition that caused the reprice from 0.90 to 0.86 are still not public; that sentence is an inferred policy, not recovered source code.

The first sale occurred while coarse NO price history moved from 0.80 (23:00) to 0.79 (23:10) and then 0.745 (23:20). A fill at 0.765021 at 23:13:15 is therefore also compatible with a protective/trailing trim during a decline, not only an absolute take-profit. The exact trigger remains not observable without a causal order-book snapshot and the unfilled/cancelled order stream.

What does the supplied prices-history URL add?

The supplied token is YES 26°C; its paired token is the traded NO 26°C. Both raw histories are included below. fidelity=15 produced points roughly 900 seconds apart, so it means a 15-minute sampling interval here, not 15 seconds. This is useful for the surrounding price path but is not a historical order book and cannot show spread, depth, queue position, or when a limit order was posted or cancelled.

2. Evidence manifest

Artifact	Contents	SHA-256
090237Z-activity.tsv	All 78 deduplicated event TRADE rows through the cutoff, including timestamp, condition/token, side, size, price, cash and tx hash	ad6f5078355a97f1969fb32f79657bae1b8d889b0baa57c48e7ae7ebfd1cdce3
090205Z-positions.json	Five unmodified Data API position records and per-response receipt times	ebe0e943b2e19eacf33d896d38f4252c28bf76126c2003673832ce58a422050a

Second snapshot and tape controls:

Artifact	Rows/points	Contents	SHA-256
event-trades-taker-false.json	6,917	Full event participant tape, one page, no cash filter, takerOnly=false	7fe92c62d957e198d75354d4711083570c7e4de0e80744ff960b128545317401
event-trades-taker-true.json	2,708	Same event with takerOnly=true; control used to classify jjavi's execution role	f58e25bc5b0bd143b81498517f77d6d498bcb69af5f48ecfb1c38bc93d8738f5
event-trades-taker-true-cash-gte-1.json	2,473	Same default taker tape with the proposed CASH >= 1 filter	c8bba676e6f104939077df9021645cc9241da08242f7e760b0e9383b2a6fff37
jjavi-trades-taker-false.json	133	Server-side eventId + user + takerOnly=false subject extract	49c18def6bee2b100728462c44e939b71169f50f2f8c03bb6cf2f83d5639e469
jjavi-activity.json	135	Later subject activity receipt with fee-aware usdcSize; two YES28 fills arrived after the tape files	a4c843348c6916c5fff0b83f3f35ef2ba4dc7066a580726dab9371ccb6c20605
current-positions.json	6	Open positions at the second snapshot	c0472ed38a4fdf839b6ec7e17c6ea5d8f118d8a8ce1e6a296154a44c8197e82d
closed-positions.json	1	Closed NO26 position and reported realised PnL	9bd1fafaa098db67794991a1b17b39bacdaa0f71f041f00ecf1f00354d14a1cd
yes26-price-history-fidelity-15.json	219	Supplied YES26 price-history query	7ddfcbd398d0532b69329a0d84f2517db724b31008940606aba0f80dc9dd40ba
no26-price-history-fidelity-15.json	219	Paired NO26 price history over the same range	6e0b1e6eaad5f59494330853b0c1b44208458d425bb4dcc1f20390ac6e143f16
event.json	1	Raw Gamma event and all 11 market/token mappings	dea0457835c0bd10360cd5bf073eca4800159881512158a3c572fddf9ec59eb4

The raw files intentionally remain compact API responses rather than reformatted or normalized copies. The 11:40 prefix identifies the collection batch, not a claim that every mutable API response was simultaneous. The activity response contains two trades at 11:39:33Z that were created after the event/user tape responses; analyses must use an explicit cutoff or treat those files as adjacent receipts, not an atomic snapshot.

End-to-end snapshot after the final observed fill:

Artifact	Rows	Contents	SHA-256
141456Z-jjavi-activity.json	149	Complete event activity through 12:15:16Z; all rows are TRADE	6e9eae19ce7a19473e4d08db8a70c733ae304917fb2b6780f1d142bb48d6ab30
141456Z-fill-ledger-with-role.tsv	149	Chronological fill ledger with maker/taker role reconstructed against the taker control	24b9187b672b0ccf08dceca9872071ed127391211f18d84e42e4b242a2b64873
141456Z-event-trades-taker-false-p0.json	8,754	Entire all-participant event tape in one page	bed2733e4337b2d75100986603b8870df93cec9f9baed0c2f4a221ae7a7ed346
141456Z-event-trades-taker-true.json	3,342	One-row-per-taker control used for execution-role classification	6d5bf2e83fc0d1cab1942f4d2a2715f61dc4f213a0cb4576f416be08eb3449ed
141456Z-jjavi-trades-taker-false.json	149	Subject-only all-participant trade response	6d315eedaa85024bc22c4d1aeb8cca2db0d86f69dfd809f42b5ad863b672266e
141456Z-current-positions.json	6	Final pre-resolution open YES inventory	c8b4702585582370f444decf086473c29d0334210f3dfe5f888ffce62901fe45
141456Z-closed-positions.json	1	Closed NO26 anchor	0b3e9e60e72096d0936a2b9da34e57b2606a35aecad688ed21dec93b537ff251
141456Z-event.json	1	Gamma rules, token mapping and pre-resolution prices	21654fd209b1911f0ceee63deda2a021e1e0caecd9e1c7ccab1236e40370f9ea
141456Z-event-low-price-015-window-30m.json	11 baskets	Deterministic all-participant takerOnly=true matched-flow report at price <= 0.15, with shares-weighted CLOB samples at ±30 minutes	0742cd5ca57d88b7c1e66a65a6b615cb2cec9c224e9116ff9dae4c131ba2cd52
141456Z-terminal-payoff-by-winner.tsv	11	Deterministic frozen-portfolio payout for every winning bucket	2205a6a0bc6632103c889ea8d62910696cf4ba807d96270c6dca72b848db4e1d
141456Z-coverage-and-hedge-summary.json	—	Derived coverage floor, per-branch cost coverage, and realised NO26 subsidy; calculated only from the frozen activity snapshot	5f72d8f52349de4d810f62add925ed1c6f85bf9af754f64b1bdb33efd5a44daa
141456Z-yes-price-history-5m/	11 files	Five-minute YES price histories for every event bucket	combined hash 7e78c8bfb88a5c7e49e55e4b4ce91e95f17f796f3a98adfdce3a5bcefcd190e4
141456Z-SHA256SUMS	—	Per-file integrity manifest for the final raw and derived package	44e7a96f7c1e94c2a3e658d6e539b66ba10c60f0e42e94f09d3f8687ffb73b5b

Tool identity used for the investigation:

./scripts/polymarket --version
polymarket 0.1.5

sha256sum ./scripts/polymarket
9b99d84d37e23f401ee33a3c4da26359be9dd343a871eee4ad564c470e83aaec

Wallet identity:

profile name: jjavi
proxy wallet: 0x6FF2CB14da8BE7Eb57541D250A0196C5F295f140

3. Sources and reproducible queries

Official public sources:

• Data activity: https://data-api.polymarket.com/activity?user=<wallet>&limit=500&offset=0
• Event participant tape: https://data-api.polymarket.com/trades?eventId=842167&limit=10000&offset=0&takerOnly=false
• Event taker control: https://data-api.polymarket.com/trades?eventId=842167&limit=10000&offset=0&takerOnly=true
• Cash-filter control: append &filterType=CASH&filterAmount=1 to the taker control
• Data positions: https://data-api.polymarket.com/positions?user=<wallet>&market=<condition_id>&limit=20&offset=0
• Data closed positions: https://data-api.polymarket.com/closed-positions?user=<wallet>&eventId=842167&limit=50&offset=0
• Gamma event: https://gamma-api.polymarket.com/events/842167
• CLOB book: https://clob.polymarket.com/book?token_id=<token_id>
• CLOB absolute price history: https://clob.polymarket.com/prices-history?startTs=<unix>&market=<token_id>&fidelity=15
• CLOB history through the local CLI: ./scripts/polymarket clob price-history <NO26-token> --interval max -o json

The Data API documents limit <= 10000 and offset <= 10000. Because the all-participant response contained only 6,917 rows, this particular event fit in one page at collection time. For an active event, offset pagination is not atomic: newly inserted trades can shift later pages. Use one maximum-size page when it fits; otherwise overlap pages, deduplicate by the fill key, and take the final archival snapshot after the event stops mutating.

Event activity coverage check:

offset 0:    total_rows=500, event rows present
offset 500:  total_rows=500, event_rows=0
offset 1000: total_rows=500, event_rows=0

The event extract applies this logical deduplication key:

(transactionHash, timestamp, asset, side, price, size)

The raw source had no duplicates under this key. The evidence TSV additionally applies the explicit cutoff timestamp <= 1786784449 so later wallet mutations cannot rewrite this audit.

4. Raw event metadata used

{
  "id": "842167",
  "slug": "highest-temperature-in-amsterdam-on-august-15-2026",
  "title": "Highest temperature in Amsterdam on August 15?",
  "active": true,
  "closed": false,
  "endDate": "2026-08-15T12:00:00Z",
  "negRisk": true,
  "enableNegRisk": true,
  "negRiskAugmented": false,
  "markets_count": 11
}

Held-market mapping at the snapshot:

Bucket/outcome	Condition ID	CLOB token used
YES 21°C	0xbbf3668e90fcb160ca1d5451a0493de6950870f4e857851082f7e7118c4b6af4	0x9ebd2c3d0b60fdd9b8ee517e57b3d47f1e74c867392f569a980c11b6d8363c03
YES 22°C	0x5fbadc95c3c2c217773bce9fcb0fcc19c9a58ca8d22c7c5c4148760e83ad3ec6	0x6b08162eaf129ca35785ce2db751a98b5fe12e66c0c6f024242860f93c0fa4ab
YES 23°C	0xe32bc4c6aa027e80482c252035416234e58ff2925555e83e3894d752b3db4ffb	0xa1d15ee022cdb26ddb7409a0ed9efdbe258140e81d033222e76317363f82323f
YES 24°C	0x4f1b9e2a40c5d130cdf966d9ed08beb00be83bd0720f77be8ff1153c675795bd	0x7423a456ed5e9b5ab88d413254ad630731cf6edc4525ae8c0cde9a3462abbdf8
NO 26°C	0x41b8cb46444cea91aafe2f4d089078b586c8d73ef298c1941efdb122e550c645	0x81cb71444d9093d2038ab5a03b09171d871fd1429ab6640e0d0edf1f5fff6f57

5. Raw-fill aggregation

The following is obtained by summing size and usdc_size in the TSV without substituting avgPrice or currentValue:

Bucket	Side	Fill rows	Shares	Cash (usdc_size)	Cash VWAP
21°C YES	BUY	28	2040.976000	10.281936	0.0050377545
22°C YES	BUY	32	1768.820085	35.648221	0.0201536727
23°C YES	BUY	4	2018.145100	81.572306	0.0404194456
24°C YES	BUY	6	1695.157958	423.230796	0.2496704180
26°C NO	BUY	1	594.824352	367.106018	0.6171670961
26°C NO	SELL	7	442.926665	355.443848	0.8024891615

Inventory reconciliation for NO26:

594.824352 BUY − 442.926665 SELL = 151.897687
Data API current size                  = 151.897600
rounding difference                    =   0.000087

This accounts for the entire NO balance through CLOB fills. There is no unexplained NO burn requiring convertPositions as an explanation.

6. Complete observed NO26 exit ladder

Fee-aware entry cash price:

367.106018 / 594.824352 = 0.6171670961

UTC	Shares	Quoted price	Cash proceeds	Effective cash price	Return vs entry	Cumulative original sold
2026-08-14 23:13:15	300.000000	0.765021000	226.809850	0.756032833	+22.501%	50.435%
2026-08-15 08:34:55	54.000000	0.900000000	48.600000	0.900000000	+45.828%	59.513%
2026-08-15 08:52:13	10.000000	0.900000000	9.000000	0.900000000	+45.828%	61.195%
2026-08-15 08:53:15	44.000000	0.900000000	39.600000	0.900000000	+45.828%	68.592%
2026-08-15 08:59:22	5.060000	0.900000000	4.554000	0.900000000	+45.828%	69.442%
2026-08-15 08:59:24	10.000000	0.900000000	9.000000	0.900000000	+45.828%	71.124%
2026-08-15 09:00:49	19.866665	0.899999975	17.879998	0.899999975	+45.828%	74.463%

Realised profit reconciliation:

NO sell proceeds                         = 355.443848
allocated fee-aware cost of sold NO      = 273.360048
realised PnL                             =  82.083800
Data API realisedPnl                     =  82.083800

7. Position snapshot and PnL reconstruction

The exact five Data API objects are in the positions JSON. Their API display totals are:

remaining initialValue = 627.270600
currentValue           = 1294.936100
cashPnl                =  667.665100
realizedPnl            =   82.083800
display combined PnL   =  749.748900

The fee-aware event ledger uses actual usdc_size instead:

all BUY cash outflow   = 917.839277
all SELL cash inflow   = 355.443848
net cash still out     = 562.395429
current marked value   = 1294.936100
fee-aware marked PnL   = 732.540671

The 17.208229 difference from display PnL is the difference between raw cash and the rounded/fee-excluding avgPrice × size basis used by open-position cashPnl.

Fee-aware contribution by leg:

Leg	Realised	Unrealised mark	Total contribution
YES 21°C	0	−9.261536	−9.261536
YES 22°C	0	+9.456679	+9.456679
YES 23°C	0	+261.512294	+261.512294
YES 24°C	0	+348.066004	+348.066004
NO 26°C	+82.083800	+40.683430	+122.767230
Total	+82.083800	+650.456871	+732.540671

Therefore most apparent success comes from unrealised repricing of YES23 and YES24, not from closed profit. The only realised result is the NO26 trim.

8. Terminal payoff if the snapshot is frozen

Using net cash outflow 562.395429 and remaining inventory:

Winning event bucket	Payout	Terminal PnL
21°C	2192.8736	+1630.4782
22°C	1920.7176	+1358.3222
23°C	2170.0427	+1607.6473
24°C	1847.0555	+1284.6601
26°C	0	−562.3954
Any other bucket	151.8976	−410.4978

These are counterfactual terminal branches for the frozen portfolio, not realised outcomes and not substitutes for the trader's actual future exit policy.

9. Retained CLOB evidence and its limitation

Sequential book reads were taken between 09:00:28.995Z and 09:01:08.956Z. The full raw book bodies were not persisted before the user requested this evidence package; only the exact calculated summaries below were retained. They must not be represented as a synchronized raw book capture.

Leg	Book time	Position size used	Total bid shares	Fillable shares	Full/partial bid VWAP
YES21	09:00:28.995	2040.9760	0	0	none
YES22	09:01:04.114	1768.8200	1822.35	1768.82	0.00140750
YES23	09:01:08.719	2018.1451	963.55	963.55	0.04736236
YES24	09:01:08.895	1695.1579	968.58	968.58	0.10389808
NO26	09:01:08.956	171.7643 (stale size)	11597.90	171.7643	0.77340533

Consequences supported by this retained evidence:

• the displayed mark PnL was not fully cash-realizable;
• YES21 had no bid;
• YES23 and YES24 lacked enough bid shares for a full exit;
• the NO26 row used a stale pre-final-fill size and is unsuitable for an exact synchronized liquidation total.

10. Historical price points around NO26 sales

clob price-history is sampled price history, not a historical order book. Retained points:

UTC minute	Price
2026-08-14 23:00	0.800
2026-08-14 23:10	0.790
2026-08-14 23:20	0.745
2026-08-15 08:30	0.865
2026-08-15 08:40	0.870
2026-08-15 08:50	0.890
2026-08-15 09:00	0.875

The first 300-share sale is bracketed by a declining sampled path. The later 0.90 fills occur around sampled prices 0.865..0.890, which is consistent with intermittent fills of a passive sell limit, but cannot prove placement time or desired total size.

11. Observer coverage boundary

Read-only VPS SQLite inspection found:

pm-trader-observer-jjavi.service = inactive
book_snapshots rows              = 628020
first observed_at_ms             = 1786721103365
last observed_at_ms              = 1786741300788
Amsterdam SELL activity rows     = 0

The last recorded book was 2026-08-14T21:01:40.788Z; the first NO26 SELL was 2026-08-14T23:13:15Z. Therefore the observer stopped more than two hours before the first exit and cannot establish the decision-time spread, depth, imbalance, or no-action controls.

12. Evidence labels at the initial 09:00 snapshot

This section records what was supportable before the event-wide tape was added. Statements about maker/taker role and the unsold NO balance are superseded by section 13; the other limitations remain.

Observed:

• exact BUY/SELL fills and cash fields in the activity TSV;
• partial NO26 exit at two price tiers;
• exact current inventory reconciliation without a NO burn;
• positive realised NO26 PnL +$82.0838;
• positive but mostly unrealised marked PnL at the position snapshot;
• incomplete contemporaneous bid liquidity in the retained sequential book summaries.

Inferred:

• the repeated 0.90 fills probably belong to a passive limit order or a small set of orders at the same target;
• the first 300-share sale may be a roughly 50% profit trim or a protective exit after a local NO price peak;
• NO sale proceeds may have financed later YES additions, but public data does not expose an internal capital-allocation label.

Not observable from retained evidence:

• order placement/cancellation time, maker/taker role and intended order size;
• whether the 0.765021 condition is absolute price, ROI, trailing drawdown, probability rank, book imbalance, weather observation, or a combination;
• whether the unsold 151.8976 NO26 was deliberately reserved or merely unfilled at 0.90;
• exact strategy profitability until fills are closed/redeemed and the event resolves without survivorship filtering.

The next acceptable test of an exit rule requires causal books at every SELL plus matched same-token no-action controls. The stopped observer did not collect that interval, so an exact causal threshold cannot be recovered retrospectively from price history alone.

13. Event-wide tape addendum (~11:40Z)

13.1 Completeness and filter experiment

The three raw event responses give a direct controlled comparison:

Query mode	Rows	Unique tx hashes	jjavi rows
takerOnly=false, no cash filter	6,917	2,708	132
takerOnly=true, no cash filter	2,708	2,708	11
default taker-only plus CASH >= 1	2,473	2,473	7

The all-participant response has more than two rows per transaction in aggregate because one taker execution can match multiple maker orders. Its 6,917 < 10,000 row count establishes that the event fit in the requested page at that receipt time. It does not make an active endpoint transactionally immutable.

For jjavi, the proposed URL exposes only 7 / 132 = 5.3% of the subject rows present in the unfiltered all-participant tape. It therefore cannot be the primary evidence source for this strategy. Use takerOnly=false and no server-side amount filter; apply analytical thresholds locally after the raw response is persisted.

The subject-only trade response has 133 rows, and the slightly later activity response has 135. The two activity-only rows are YES28 buys at 11:39:33Z. This is expected drift between adjacent calls to an active market, not a deduplication error.

13.2 Maker/taker reconstruction

Role is classified by the presence of the same wallet row, not merely the same transaction hash, in the takerOnly=true control. A transaction hash appears in the taker tape even when jjavi was maker, because the counterparty's taker row is present.

At the event-tape receipt, only 11 of jjavi's 132 rows were taker rows; 121 were maker rows. Thus 91.7% of the observed subject rows were passive-side executions. This directly supports a passive low-price bidding mechanism and explains the numerous fragmented fills at exact price levels such as 0.002, 0.005, 0.008, and 0.020. It does not reveal when those limits were placed or how far they sat from the contemporaneous best bid/ask.

For NO 26°C, execution roles are exact:

Action	UTC	Shares	Quoted price	Role
BUY	2026-08-14 19:38:43	594.824352	0.605220679	taker
SELL	2026-08-14 23:13:15	300.000000	0.765021000	taker
SELL tier	2026-08-15 08:34:55..09:00:49	142.926665 in 6 fills	0.900000000	maker
SELL tier	2026-08-15 09:21:19..10:11:24	151.890000 in 4 fills	0.860000000	maker

13.3 Complete NO26 ledger

Fee-aware entry basis from activity usdcSize:

BUY cash / BUY shares = 367.106018 / 594.824352 = 0.6171670961

Tier	Fills	Shares	Cash proceeds	Effective cash price	Return vs entry	Realised PnL
taker 0.765021	1	300.000000	226.809850	0.756032833	+22.501%	+41.659721
maker 0.90	6	142.926665	128.633998	0.899999997	+45.828%	+40.424363
maker 0.86	4	151.890000	130.625400	0.860000000	+39.346%	+36.883890
Total sold	11	594.816665	486.069248	0.817174899	—	+118.967974

Inventory and PnL reconciliation:

594.824352 bought - 594.816665 sold = 0.007687 NO dust
pro-rata fee-aware realised PnL       = 118.967974
Data API closed-position realisedPnl = 118.967700
rounding difference                  =   0.000274

Subtracting the entire $367.106018 entry cash from proceeds gives $118.963230; the $0.004744 difference is the fee-aware cost basis of the 0.007687 residual dust. This explains the two defensible totals instead of treating the discrepancy as an API error.

13.4 Price-history evidence

The supplied decimal token ID 101499751086362065576178614498558620908321047957444790295118593670413813098432 is YES 26°C. The NO token sold by jjavi is 58707809160173458979728895032303842312851256717980097247292434992996103974743.

Each fidelity=15 response has 219 points. Adjacent timestamps are approximately 900 seconds apart. Selected NO points surrounding the exit tiers are:

UTC	Sampled NO price
2026-08-14 23:00:14	0.800
2026-08-14 23:15:14	0.745
2026-08-15 08:30:13	0.865
2026-08-15 08:45:08	0.835
2026-08-15 09:00:14	0.875
2026-08-15 09:15:09	0.855
2026-08-15 09:30:11	0.825
2026-08-15 09:45:12	0.830
2026-08-15 10:00:13	0.835
2026-08-15 10:15:12	0.890

The 0.90 and 0.86 maker executions are consistent with passive asks being touched while the sampled price moved around them. The series cannot distinguish one persistent order from several cancel/replaces and must not be interpolated as a historical bid/ask book.

13.5 Updated algorithm hypothesis

The strongest evidence-backed formulation for this one anchor trade is:

1. Cross the spread to acquire a liquid NO anchor when the algorithm wants immediate exposure.
2. When profitable, cross again to remove about half: 300 / 594.824352 = 50.435%.
3. Offer the remainder passively at a high fixed price, observed here at 0.90.
4. If that level does not finish, cancel/reprice the residual lower, observed here at 0.86,
   and allow incoming takers to consume it in fragments.
5. Stop when only non-economic dust remains.

Steps 1, 2, the executed price levels, roles, quantities, and dust are observed. A single 0.90 order, a later cancel/reprice, and the motivation for the 0.86 level are inferred. No fixed time cadence or fixed child-order size is supported: fragment sizes are determined by counterparty matches.

Across the wider event, the 91.7% maker-row share supports the broader entry hypothesis that jjavi posts passive bids at discrete low prices and accumulates only when other traders sell into them. Exact entry gates based on spread, depth, rank, remaining time, or sibling-price sums remain unidentified because neither endpoint preserves historical books or unfilled/cancelled orders.

13.6 Current portfolio boundary

The second current-position response contains six open YES legs: 21, 22, 23, 24, 27, and 28. Their API display totals are:

initialValue =  544.3865
currentValue = 1573.9022
cashPnl      = 1029.5154

These are open marked values, not fill-verified exit proceeds. The completed NO26 anchor is a realised success, but a profitability verdict for the full construction still requires terminal or actual exit fills for every open YES leg and no survivorship filtering.

14. End-to-end reconstruction with terminal winner 23°C

14.1 Coverage

The final activity and subject-trades responses both contain exactly 149 rows. The activity page is below its 500-row cap, starts at the wallet's first Amsterdam fill, and all position inventories reconcile. The all-participant event tape contains 8,754 rows and 3,342 unique transaction hashes, below the requested 10,000 limit. This event therefore has no observed API pagination truncation.

All 149 wallet rows are TRADE. There are no SPLIT, MERGE, CONVERSION, REDEEM, reward, or rebate rows attributed to this event. The neg-risk construction was assembled and unwound with CLOB trades; public activity provides no evidence that convertPositions produced the profit.

Execution-role comparison yields 138 maker and 11 taker fills. Thus 92.6% of wallet fill rows are passive, but most dollars were not passive low-price bets: YES24 + NO26 consumed 84.63% of all BUY cash and were entered entirely as taker trades.

14.2 Whole-market movement

Five-minute CLOB histories show the rotation of the market's modal temperature. Prices below are the last sampled YES price at or before each milestone, not historical executable books:

Bucket	Initial entry 19:38	First NO trim 23:13	NO 0.90 starts 08:34	NO exit complete 10:11	Market end 12:00	Near-terminal 14:00
21°C	0.0030	0.0025	0.0005	0.0005	0.0005	0.0005
22°C	0.0070	0.0085	0.0285	0.0005	0.0005	0.0005
23°C	0.0500	0.1150	0.1650	0.0750	0.5000	0.9150
24°C	0.0950	0.2750	0.3950	0.3950	0.4200	0.0700
25°C	0.3200	0.3650	0.3000	0.4100	0.1050	0.0190
26°C	0.4150	0.2100	0.1350	0.1500	0.0250	0.0100
27°C	0.1350	0.0650	0.0150	0.0250	0.0100	0.0050
28°C	0.0350	0.0005	0.0015	0.0050	0.0050	0.0035

The market initially favoured 25–26°C, rotated toward 24–25°C, and only after the nominal market end moved decisively to 23°C. jjavi's largest winning fill — 2,000 YES23 @ 0.04 — occurred at 19:52:33Z, before that rotation was visible in the sampled market price.

14.3 Coverage-band sizing and execution

The important invariant is the resulting payout per covered branch, not equal dollars per order:

Leg	BUY cash	Cash VWAP	Final shares/payout if it wins	Payout / all $933.84 BUY cash	Terminal PnL after NO harvest
YES21	10.281936	0.005038	2,040.976000	2.186×	+1,593.209955
YES22	35.648221	0.020154	1,768.820085	1.894×	+1,321.054040
YES23	81.572306	0.040419	2,018.145100	2.161×	+1,570.379055
YES24	423.230796	0.249670	1,695.157958	1.815×	+1,247.391913

These four legs cost $550.733259 and their acquired VWAPs sum to 0.315281291. That sum is a descriptive reconstruction across asynchronous fills, not a claim that all four prices were simultaneously executable. Their share counts cluster around a common payout target despite very different prices: the cheapest missing branches required little cash, while the expensive 24°C branch required aggressive taker buys. Every individual 21–24°C payout exceeds not merely the four-leg basket cost, but the account's entire $933.842980 gross BUY cash for the event.

This supports a coverage-band interpretation: choose a set of plausible neighbouring buckets, then buy enough shares in each to raise the minimum payout if the winner lies in that set. It is not proof that the private code uses a literal equal-share target; observed holdings differ by up to 345.818042 shares and public data does not expose intended but unfilled quantities.

For an ideal equal-share basket with selected set S, target payout Q, and executable entry prices p_i:

basket_cost = Q * sum(p_i, i in S)
conditional_profit = Q - basket_cost
profitable for every covered branch iff sum(p_i, i in S) < 1

This is only a conditional guarantee because 21–24°C is not the entire event. For the actual unequal inventory, the correct post-harvest branch test is:

PnL(i) = YES_shares(i) + NO26_terminal_payout(i) - net_deployed

14.4 How the cheap maker ladders fit the construction

Every executed entry level is recoverable:

Bucket	Maker BUY levels	Taker BUY levels	Role in resulting portfolio
YES21	0.002, 0.005, 0.006	—	passively complete 2,040.976 shares for only $10.28
YES22	0.005, 0.008, 0.025, 0.030	0.012, 0.029333	raise a cheaper branch toward the common payout range
YES23	0.040, 0.060, 0.120	—	2,000-share maker block plus small later completion
YES24	—	0.110 through 0.279843, later 0.166981	actively build the expensive branch needed for the payout floor
NO26	—	0.605221	immediately acquire a broad non-26 payoff and monetisable reserve
YES27	0.020	—	small $8 extension; 399.981 shares do not fully cover final net cost
YES28	0.002	0.002 probe	cheap $8 extension whose 4,001-share payout does cover the portfolio

There is no single global 1–5¢ entry threshold. Cheap prices make it inexpensive to fill a missing payout branch: a maker fill of Q shares at price p adds $Q of conditional payout for only $Qp of cost. The economic control variable is therefore plausibly target branch payout versus incremental cost, while the observed price ladders are the execution mechanism. The private rule that decides which buckets belong to the coverage band remains unknown.

14.5 Reconstructed state machine

The revised highest-confidence algorithm for this event is:

SELECT_COVERAGE_BAND
  select the plausible/leading temperature region to cover
  observed resulting core set: YES21, YES22, YES23, YES24

SET_PAYOUT_FLOOR
  size each core branch toward roughly 1,700-2,000 terminal dollars
  exact target and tolerance are inferred, not observable

BUILD_COVERAGE
  taker BUY the expensive/urgent YES24 branch
  place passive bucket-specific ladders for cheap YES21/22/23
  keep cheap bids resting until fills lift branch payout toward the target range

ADD_MONETIZABLE_NO_ANCHOR
  taker BUY 594.824352 NO26 for $367.106018
  while held, it pays on every terminal bucket except exactly 26°C
  if probability rotates away from 26°C, its mark rises alongside the lower YES thesis

HARVEST_NO_ANCHOR
  taker SELL 300 shares at +22.501% fee-aware return
  maker SELL 142.926665 at 0.90
  maker SELL 151.89 at 0.86
  realised cash profit: $118.963230; retain only 0.007687 dust

EXTEND_COVERAGE_CHEAPLY
  add YES28 at 0.002 and YES27 at 0.02 for about $8 each

HOLD_YES_TO_SETTLEMENT
  no YES sells, no stop-losses, no conversion
  exactly one winning YES branch pays; the rest expire at zero

Chronology is consistent with this staged construction, but the labels are analytical states, not public order instructions. YES28 accumulation began at 09:03:04Z, the NO26 exit completed at 10:11:24Z, and YES27 accumulation began at 10:14:30Z. The final YES28 maker fill at 12:15:16Z may have consumed an older resting order; it does not prove a new post-12:00 decision.

14.6 NO26 as hedge, reserve, and cost subsidy

While held, NO26 is a structural hedge against every terminal result except exactly 26°C:

winner j in held YES set, j != 26: payout = NO26_shares + YES_shares(j)
winner j outside held YES set, j != 26: payout = NO26_shares
winner j = 26:                         payout = 0

It is not a conventional negatively correlated hedge. A rotation from 26°C toward 21–24°C raises NO26 at the same time as it strengthens the selected YES thesis, allowing the bot to turn the broad non-26 claim into cash before settlement. Here $367.106018 of NO26 was sold for $486.069248, a realised gain of $118.963230. That gain financed 20.990907% of the entire $566.736962 YES basket and reduced total net deployed cash to $447.773732.

This interpretation matches the user's proposed role: keep a reserve while the low-temperature YES positions develop, then harvest it as NO26 appreciates. It is still a risky construction: 26°C is the shared hole, and selling NO26 removes its terminal floor from all other uncovered outcomes. The executed exit tiers are observed; the causal signal for the first trim and the 0.90 -> 0.86 reprice remains unobservable.

14.7 Terminal cash-flow for winner 23°C

Using actual activity usdcSize, actual NO sells, and the deterministic 23°C payout:

all BUY cash outflow                 =   933.842980
all SELL cash inflow                 =   486.069248
net cash still deployed              =   447.773732
YES23 payout                          = 2,018.145100
NO26 reconciled dust payout           =       0.007687
terminal economic PnL                 = 1,570.379055
return / total BUY cash               =      168.162%

Contribution decomposition:

YES23 payout less its entry cash      = +1,936.572794
losing YES21/22/24/27/28 entry cash   =   -485.164656
NO26 sell cash less total entry cash  =   +118.963230
NO26 dust payout                      =     +0.007687
total                                 = +1,570.379055

No REDEEM row is required to know this terminal branch. At collection time Gamma was still open and YES23 traded near 0.92–0.95; the calculation treats 23°C as the known terminal result and the later redeem as mechanical settlement.

14.8 Risk shape and what actually produced the profit

The final portfolio is directional, not an arbitrage. After the NO anchor was sold, frozen-branch PnL ranged from approximately -$447.77 to +$3,553.28:

Winner	Terminal PnL
21°C	+1,593.21
22°C	+1,321.05
23°C	+1,570.38
24°C	+1,247.39
25°C	-447.77
26°C	-447.77
27°C	-47.78
28°C	+3,553.28
20°C/below, 29°C, 30°C/higher	-447.77 each

The realised winner was the early 2,000 YES23 @ 0.04 maker block, but that is not the whole algorithmic explanation. Before knowing the result, the bot had built four adjacent core branches whose individual payouts all covered total gross event spending. The 21°C and 22°C micro-price fills were therefore economically material coverage despite their tiny cash cost; YES24 was the expensive completion of the same floor. NO26 then lowered the effective cost of keeping that YES basket by $118.963230.

The construction was not risk-free and not a complete all-bucket arbitrage. Its strong floor was conditional on 21–24°C; the final post-harvest holes shown above remained. The right description is band-limited payout coverage plus a monetised non-26 hedge, not “one lucky cheap bet” and not “buy every market for less than one dollar.”

14.9 Confidence boundary

Observed with high confidence: all fills and cash, every executed level, maker/taker role, position inventory, absence of conversion/merge/split, NO exit ladder, no YES exits, market price trajectory, and the terminal payoff for winner 23°C.

Inferred with medium-to-high confidence: a targeted 21–24°C payout band with roughly aligned branch payouts; cheap maker ladders used to fill missing branches; and NO26 used as a monetisable non-26 hedge that subsidises the held YES inventory. Approximately $8 fixed-risk late-extension sizing and cancel/reprice from 0.90 to 0.86 remain medium-confidence details.

Not observable retrospectively: the private signal that selected Amsterdam or the 21–24 core; whether there was a literal target-share variable and its tolerance; the exact start time; unfilled order sizes and placement/cancellation timestamps; decision-time spread/depth/queue; weather/model inputs; and the precise conditional expression that caused each ladder change. Those cannot be made exact from trades plus sampled price history alone.

This event proves a profitable execution instance, not positive expected value. A +EV verdict still requires the same terminal reconstruction across an unbiased multi-event sample and the canonical Harvester/overlay EV gate with fill-verified realised inputs.

The calculator's editable source is NegRisk Coverage Band Yield Calculator v2.html; v1 remains preserved as the earlier partial-snapshot model.

15. All-participant low-price matched flow

Superseded analysis method: this original per-basket t±30m aggregation is retained as historical evidence but is not the current opportunity analysis. The wallet-independent episode model, SELL-taker capacity split and recovery lifecycle now live in amsterdam-2026-08-15-event-dip-lifecycle.md.

This section is event-wide and does not filter jjavi or any other wallet. The public Data API was queried by eventId=842167 with takerOnly=true, so every row is one matched taker transaction. This avoids counting the maker and taker participant rows of the same match twice. The source contains 3,342 rows, all unique under (transactionHash, timestamp, asset, side, price, size); 1,083 rows belong to the selected YES tokens. Historical unfilled and cancelled orders are not exposed by this endpoint.

For every match at an inclusive price <= 0.15, the analyzer selects the nearest saved CLOB price-history sample at t-30m and t+30m, accepts it only within 360 seconds, and computes a shares-weighted mean per basket. Percentages in parentheses are the share coverage of those history samples, not trading returns.

Basket	All matches	All shares	Matches <= 0.15	Shares <= 0.15	CLOB -30m	CLOB +30m
20°C or below	0	0.000000	0	0.000000	—	—
21°C	13	5131.470665	13	5131.470665	0.007144 (100.00%)	0.001194 (100.00%)
22°C	39	4356.871654	39	4356.871654	0.012374 (100.00%)	0.008095 (100.00%)
23°C	225	6797.385120	89	2785.347142	0.072514 (100.00%)	0.090493 (100.00%)
24°C	255	6858.214439	56	1358.831219	0.193936 (100.00%)	0.101073 (86.18%)
25°C	220	5466.899671	56	1747.422099	0.151705 (100.00%)	0.072908 (92.55%)
26°C	192	5143.815811	40	1183.105000	0.084574 (100.00%)	0.044635 (98.31%)
27°C	95	2686.360812	50	1776.688266	0.050052 (100.00%)	0.044339 (100.00%)
28°C	39	3106.038608	39	3106.038608	0.007009 (100.00%)	0.008899 (83.82%)
29°C	1	8.270000	1	8.270000	0.005500 (100.00%)	0.005500 (100.00%)
30°C or higher	4	151.117000	4	151.117000	0.000841 (100.00%)	0.002867 (100.00%)
Total	1083	39706.443780	387	21605.161653	—	—

Thus 35.73% of the observed YES matched transactions and 54.41% of their shares executed at <= 0.15. This is market-wide flow, not evidence that jjavi received all of those shares and not a profitability claim. The deterministic offline reproduction is:

scripts/analysis/forensics.sh event-low-price \
  --event 842167 --threshold 0.15 --window-minutes 30 \
  --max-price-gap-seconds 360 \
  --event-json docs/analysis/data/jjavi-amsterdam-2026-08-15-141456Z-event.json \
  --trades-json docs/analysis/data/jjavi-amsterdam-2026-08-15-141456Z-event-trades-taker-true.json \
  --offline-tape-complete \
  --prices-dir docs/analysis/data/jjavi-amsterdam-2026-08-15-141456Z-yes-price-history-5m