2 · Follow the orders and the capital
Before this run, copy the files from for-later/reveal in the ZIP into cli-practice: OUTCOME_SOURCES.csv, order.txt, outcomes.txt.
Quit the agent, start it again in cli-practice, and paste only this:
Work only inside this practice folder. Read the original frozen packet and saved analysis first; then read outcomes.txt, order.txt and OUTCOME_SOURCES.csv. Never modify any source or original output. No network, installs, reads outside the folder, or questions. Use Python 3 standard library only. Use American English spelling in all page copy, such as "collateralized," never British variants. Do not use em dashes or en dashes. This is a fresh session: the new records were not available on August 3, 2012. Write only reveal.json, reveal.html and reveal-check.py. Build a dated, source-linked order-flow timeline directly from the regulator's numbered findings: the deployment, old code and flag, incomplete deployment, premarket messages, opening orders, routing/executions, incident response, positions and ultimate loss. Be precise about what the order actually establishes, its later publication date, its without-admitting-or-denying settlement language, and how its loss estimate relates to the initial company estimate. Distinguish parent orders, child orders, executions, shares and long/short positions; do not subtract gross position values to derive trading loss. Read the issuer's later financing filing to distinguish a completed infusion from a speculative need, and do not imply it was knowable in the initial call. Preserve the original conditional financing call, including the original compliant-regulatory-capital statement, without pretending a hindsight verdict was certain. Do not make unsourced allegations about individuals or their motives. reveal.json records dated events, sourced quantities, the original as-of capital bridge, financing and remaining unknowns. reveal-check.py independently parses both sets of originals, not importing or running model.py, computes and validates decision-critical values against JSON and tagged HTML, checks source URLs, and rejects the later mechanism being labeled known on August 3. Its --self-test demonstrates three distinct in-memory corrupted outputs are rejected without touching inputs. Run python3 reveal-check.py and python3 reveal-check.py --self-test and report exit codes. Make reveal.html a complete self-contained offline page with two visibly dated columns or stacked phone cards: knowable August 3 versus established by the August 6 filing and October 16, 2013 SEC order. Include one source-linked inline SVG timeline or zero-baseline comparison where quantities have comparable units; keep all decision-critical dates, dollar amounts, order and execution quantities in ordinary semantic HTML with JavaScript off. Render all explanatory monetary amounts in full dollars with comma separators, including chart labels, comparison prose, interactive evidence panels, and original bridge; do not abbreviate as $M, $B, or "$440 million". The published HTML must not contain the consecutive words `in` and `thousands` anywhere; describe the source columns as scaled by a factor of 1,000 and ensure reveal-check.py rejects this prohibited phrase outside source packets. Historical source wording must remain intact inside clearly attributed quotation elements and raw source fields, even when it says "$440 million" or "$460 million". Neither copy raw scaled balance-sheet amounts into explanations nor rewrite authentic quotes to satisfy the display rule. Let a real reader activate a clear control to inspect one order-flow transition or later source excerpt and visibly change a substantive evidence panel, with a static equivalent and reset; pointer and keyboard both work. The timeline should make the parent-to-child routing mechanism inspectable without implying every child order executed. No external assets or network dependencies. At 390 pixels wide, the page fits the screen; no word or amount ever breaks mid-word; narrow tables become stacked cards. Respect reduced motion, preserve complete source links and distinguish approximate from exact quantities.
Open what OMP made, full screen
Human check: KC4 is August 6, 2012, after the blind cutoff: the filing records $400,000,000 completed preferred financing, 400,000 preferred shares, roughly 73% as-converted ownership, and its finance committee's stated financial-viability exception. KC5 is SEC order October 16, 2013, Release 70694: 212 parent orders, 4,000,000 executions in 154 stocks, MORE THAN 397,000,000 shares over approximately 45 minutes, and approximately $460,000,000 eventual loss, versus KC2's August 2 approximately $440,000,000 initial estimate. The order describes one of eight servers missed in deployment, repurposed flag invoking Power Peg, 97 premarket messages, and removal of new code from seven servers worsening the event. The order says settlement without admission or denial of findings except jurisdiction. Do not convert later findings into August 3 knowledge. Run reveal-check.py and its self-test and inspect interaction, dates and source links.
When we checked the recording: Clean fresh OpenAI OMP reveal exited 0 with unchanged inputs; independent reveal checker and three corruption self-tests reported exit 0. Reviewed dated knowledge boundary, 212 parent orders, >4,000,000 executions, >397,000,000 shares, $400,000,000 completed financing, SEC order $460,000,000 finding, complete-dollar explanatory HTML, preserved cited source quotation and source links. Responsive CSS and interaction wiring reviewed, browser not exercised here.