Review and publish Watchlist analysis

In AI Controls, choose Current analysis or Simple analysis, then save review controls. The selection applies to new requests only; existing drafts and published reads keep their format. Simple analysis focuses on the setup, pullbacks, where price could go next and thesis invalidation. It always requires owner approval when AI generation runs. Edit prices, ranges and text in the card, or uncheck Show this section to hide it. Saving does not publish or request another analysis.

If a breakout was omitted from a draft, its editing fields remain available. Enter your corrected price and explanation, enable Show this section if needed, then save and preview. Price-order warnings do not rewrite or block your correction. The original and your saved version remain separate.

A generation may provide a primary breakout and an optional supported backup in the same AI request. Both the price and explanation are checked before selection. Only the selected breakout appears in the draft; you can edit it before approval. If neither works, independent valid setups can remain. Candidate selection and omission reasons are kept in the audit.

Website and Discord delivery are tracked separately. If delivery needs attention, inspect its status before retrying. An uncertain delivery must not be treated as a confirmed failure or blindly resent.

Open Discord notifications from the main Watchlist Admin to switch @everyone and individual role mentions on or off. Add a role using its Discord role ID and a recognizable label, or remove it from these settings without deleting the Discord role. Save mentions applies to future notifications, including approved analysis updates when Notify users is selected. Posts already approved retain their audience on retries. These controls do not change email, push or Discord channel access. Discord role permissions and each member's notification preferences still apply.

If an approved message appeared in Discord but its delivery is awaiting confirmation, select that message part, enter its Discord message ID, and choose Verify existing message. Verification checks the existing message against the approved text and records a matching receipt; it does not send a message. After successful verification, Retry Discord delivery can finish any remaining parts without resending confirmed parts. A message that cannot be verified stays unresolved.

Publish ticker without analysis lists the ticker with its available levels, indicators and saved Trader notes. Notify users for ticker-only post starts checked; uncheck it to list silently. Any analysis draft stays private until you approve it. Approving the analysis together with a new ticker sends one listing announcement. For a ticker already listed, Approve and publish analysis adds the approved read; select Notify users to announce it, or leave it unchecked to publish silently. The checkbox starts unchecked for each new approval. Discord analysis updates include the approved images when available; push and email contain links only. Members choose push and email independently in Account Notifications, and both start off. Saving edits or retrying delivery does not create another announcement. Delivery problems do not undo your approval.

Uncheck Generate analysis beside the ticker field to add a notes-only ticker. Enter My notes, then add the ticker. It stays private until you publish it, even if global analysis generation is off. My notes on its row lets you save a draft; after listing, Publish notes updates the member card without notifications. Blank published notes remove that card. Generate analysis on the row requests an analysis later when global/session generation is enabled; review and approve it before members see it. Published notes remain alongside the analysis, and unsent notes drafts stay private. The per-ticker choice prevents automatic analysis requests after restart as well. Automatic follow-up AI requests are off by default. Turning them off preserves their settings for later and does not stop live price/data updates. Manual refresh remains available and requests a new analysis. When a required-review replacement is waiting for approval or its generation failed, another automatic request waits for owner action; live data continues.

Review before publishing applies to new activations when analysis generation is enabled for that session. If generation is off when you add the ticker, the post follows the normal publishing path without an initial AI request, even if settings or the session change while it prepares. Manual refresh remains available when generation is enabled. With review enabled, replacement analyses also wait for approval, including refreshes of older public tickers. The existing public analysis and live data remain available. Changing review settings does not silently approve a post already waiting for review.

If a request was interrupted before publication was acknowledged, restart keeps that status visible. With automatic updates off, it does not send another initial request. Manual refresh remains a separate request when generation is enabled.

Choose an Audit generation, then Inspect request. Analysis checks summarize recorded section omissions, optional objective omissions and breakout selection. New original drafts retain their validation decisions with the saved version, so those reasons remain available even if temporary diagnostics expire. Owner edits remain separate from the original decisions. Expand the validation records or original version for full reasons and prices, or inspect the input packet, AI response and saved versions. Missing diagnostics are labelled. Very large sections are shortened on screen for performance; Export audit contains the full available record. Inspecting does not generate or publish analysis.

Analysis checks also explain recorded core and must-clear price anchors and failed checks. A pullback or recovery explanation that fails a volume or session-high claim check can be omitted without removing an independent valid setup. Open its validation record for the original explanation and exact reason. These checks describe the generated analysis, not your later corrections; owner edits are saved separately.

Captured records lists how many input, response, validation, prepared-analysis and transport-error records are available. A zero does not prove that no request was sent or that the stage never occurred; it means that record is unavailable in this audit.

New input records include prompt, schema and request hashes, plus a fingerprint of the named analysis code files and the deployed commit when available. Missing code files are marked incomplete. This fingerprint covers the listed analysis modules, not the entire application.

A rejected checkpoint can also remove levels that depend on it. Separately supported independent levels can remain. Older responses without dependency details may omit the later part of that sequence. The audit preserves the omitted levels and reasons; you can still edit the draft before approval.

AI Controls shows full model versions: GPT-5.6 Luna, GPT-5.6 Terra, GPT-6 Luna and GPT-6 Sol. Choose None, Low, Medium, High, Extra high or Max reasoning effort, then Apply Model. This affects future requests only; it does not regenerate or publish an analysis. Higher effort can use more tokens and take longer. Existing selections stay unchanged until you apply another choice. Fallback model defaults to Off. You can choose a fallback model and its effort, then Apply Model. If the first request cannot produce a draft, one additional paid request is allowed; a usable draft retained for your review does not trigger fallback. Both attempts and their costs appear in usage. Existing generation switches and daily budget still apply, and approval is unchanged. New cost estimates use current standard token rates unless the operator configured an override; saved historical costs are not rewritten. API requests counts distinct recorded attempt IDs for the selected generation, not the number of status or edit rows. Estimated cost uses recorded usage and pricing, including available usage for rejected analyses. Missing historical diagnostics, usage or pricing show Unavailable rather than zero. These recorded totals are not a provider billing statement.

Export audit downloads the selected generation's available request/response diagnostics and saved review history, including original, edited and approved versions when present. Failed generations can also be selected. Use Ticker history to inspect earlier saved histories, including before a ticker was removed and added again. Historical records are read only; select a request there to export its saved analysis and revisions. Review ticker returns to the current review. Missing or unavailable diagnostics are labelled; older requests are not reconstructed. Export does not automatically share the file with Codex or anyone else.

Choose Indicator Audit directly from the Watchlist Admin menu. Open a refresh to see its ticker activation, outcome, provider attempts, fallback results, timeframe coverage and calculation reference. Moomoo is primary; Yahoo is the fallback. Provider connection status remains in the existing Market Data Connections area. Changing sections keeps the runtime forms and saved analysis controls mounted.

Request counts deduplicate recorded transport IDs across the displayed refresh records. Started, finished and shared-consumer events do not each count as another request. These are observed Indicators/shared-candle requests, not all usage on your provider account. Expand provider requests for explicit HTTP rate-limit evidence, retry timing and failures; a timeout alone is not proof of a rate limit.

Choose Check calculation inputs, then Export calculation JSON when available. The export refers to that exact retained calculation, not the ticker's latest data, and includes available candles, starting state and calculation version. Expired inputs are labelled rather than reconstructed. Exporting does not automatically share data with Codex. Audit coverage shows retention, expired records and dropped writes; buffered events may be missing after a restart.