rea/docs/web-network-captures.md

Historical web network captures

Use inspect_web_network_capture / inspect-web-network-capture to inspect retained producer evidence. This is an offline operation: recorded URLs are never fetched, and no browser or proxy listener is started.

rea inspect-web-network-capture /absolute/session.har har --json
rea inspect-web-network-capture /absolute/session.har har --record 2 --record 0 --json

{
  "name": "inspect_web_network_capture",
  "arguments": {
    "capture_path": "/absolute/session.har",
    "format": "har",
    "record_ordinals": [2, 0]
  }
}

Omit ordinals to return every record. Explicit selection preserves caller order and original zero-based producer ordinals; duplicate or out-of-range selections are input errors. Every record is validated before selection, so selecting one entry does not hide a malformed entry elsewhere.

For agent discovery over MCP, request detail: "summary". REA retains the complete selected Evidence in the session and returns identity and counts. Use normalized_result.parent_evidence_id with inspect_analysis_view:

{
  "source": {
    "kind": "retained-evidence",
    "evidence_id": "<parent_evidence_id>"
  },
  "view": { "kind": "capture", "detail": "preview", "offset": 0, "limit": 20 }
}

Preview pages select recorded HTTP headline fields, event names/times when present, original locations and byte descriptors; they omit headers, XML and payload content. Choose detail: "complete" in that view for exact record pages, including all sidecars. Offsets follow the parent's selected record order; producer ordinals remain unchanged. total_records counts the original capture while retained_records and page coverage count records in the parent. An exhausted page does not establish absence from unselected records.

Views do not reopen the file or rerun its decoder. They remain historical Evidence if the source changes or disappears; inspect the capture again to observe new bytes. The inspection tool's default detail: "complete" preserves existing inline delivery. For CLI use, save complete inspection JSON and pass it as inline Evidence to inspect-analysis-view; the same summary and capture page contracts apply. Large complete records can still exceed MCP's response budget; export the retained Evidence when a single record cannot fit.

If decoding completes before cancellation or private-root cleanup fails, the failure retains the completed capture as Evidence with the requested record selection and explicit sensitivity exclusions. Earlier decode failures remain primary. Failed cleanup stays linked to its exact owner, including through redaction, and its live status clears only after a confirmed retry.

Formats and upstream profiles

HAR uses har-schema@2.0.0 draft-06 schemas, ajv@8.20.0, ajv-formats@3.0.1, jsonc-parser@3.3.1 JSON visitor, and lossless-json@4.3.1 numeric representations, pinned in the npm lockfile. Only the HAR 1.2 profile is supported. Unknown extensions remain reported evidence. Duplicate object keys, including equal values and escaped spellings, are rejected before materialization. Prototype-named members such as __proto__ are preserved as ordinary own JSON members without prototype mutation. Original capture files remain unchanged. REA owns specific schema corrections because pinned upstream schemas omit object types and use the unrecognized min keyword. Following HAR 1.2, timings must be an object with nonnegative send, wait and receive; optional blocked, dns, connect and ssl may be absent, -1, or nonnegative. Entry time must be nonnegative; page onContentLoad and onLoad may be absent, -1, or nonnegative. cache must be an object, while its optional beforeRequest and afterRequest retain their specified nullable forms. Invalid values fail with their source JSON pointer before record selection or sensitivity projection. Measurements are not invented or repaired. Decoder provenance names these corrections as har-schema-corrections@1. The observed mitmproxy 12.2.3 SaveHar profile can emit postData.text: null for a missing request body. REA preserves that reported null, omitting the optional field only from its validation copy for this exact producer profile; it does not invent an empty string or body bytes. The observed Playwright 1.58.2 profile can report a bracketed IPv6 serverIPAddress, forwarded from Chromium's remoteIPAddress through Playwright's HAR tracer. REA preserves the original spelling in reported Evidence and removes brackets only from its validation copy, only for this declared producer and a valid IPv6 literal. Invalid addresses, bracketed IPv4, ports and zone identifiers remain validation failures. Producer declarations do not authenticate origin.

Native mitmproxy decoding requires a caller-supplied mitmdump 12.2.3 on Linux. Set its absolute executable path:

REA_MITMDUMP_COMMAND=/absolute/mitmdump rea inspect-web-network-capture /absolute/session.mitm mitmproxy --json

REA does not install mitmproxy. Its owned process uses a private configuration directory, --no-server, and the unchanged upstream mitmproxy.io.tnetstring decoder. It does not run FlowReader migrations or convert native states to HAR. Raw IDs, state versions, flow types, backup states and unknown extensions are preserved without synthesizing connections, timestamps or content. The tested upstream source is tag v12.2.3, commit 6c09d56e4c29a92f5ad01b03199977584b8ea14f; other native tool versions return an explicit unsupported-profile diagnostic. Non-Linux native decoding is not supported by this adapter.

Reading Fiddler captures (.saz)

Fiddler Everywhere saves traffic in Session Archive Zip (.saz) files. Inspect native SAZ directly, without installing Fiddler or converting to HAR:

rea inspect-web-network-capture /absolute/session.saz saz --json
rea inspect-web-network-capture /absolute/session.saz saz --record 0 --json

MCP uses the same capture_path, format: "saz", and optional record_ordinals fields as the examples above. REA pairs raw/<id>_c.txt requests, _s.txt responses, and _m.xml metadata by filename stem. Archive member names identify the source; zero-based ordinals follow each session's first member in ZIP directory order, not its numeric ID or capture time.

HTTP fields and ordered duplicate headers are retained, with stored body bytes in binary sidecars. Bodies are not dechunked or decompressed. Metadata is an XML node tree; optional _w.txt and _g.txt files retain opaque WebSocket and gRPC message bytes. Missing members remain null, distinct from empty bodies. A zero-byte _c.txt or _s.txt member, which Fiddler writes for a session without that message, is also null and adds a record limitation. See Telerik's SAZ format reference.

Protobuf and gRPC bodies

Selecting record_ordinals (CLI --record) also decodes retained bodies with recognized protobuf or gRPC content types. Whole-capture inspection does not perform this extra work. The original binary sidecar, SHA-256 and producer ordinal remain the source; decoding adds flat body, frame and field facts. Automatic binary Protobuf admission validates reserved MIME parameters: encoding may be binary, while charset and the reserved version must be absent. Illegal profiles retain the source body, add a limitation, and leave protobuf coverage partial. Ordinary non-encoding parameters remain accepted; an explicit framing choice interprets bytes independently of the MIME profile. Admission follows HTTP token and quoted-string syntax: unterminated quotes and conflicting repeated reserved parameters remain invalid. Repeated headers with implicit binary encoding and encoding=binary agree without rewriting either header. Invalid declarations remain visible even with explicit framing, and declared gRPC bodies still receive credential checks before a MIME refusal. An excluded HAR mimeType remains unknown for automatic interpretation even when a visible Content-Type header exists. That header can still identify the transport for credential checks; explicit wire framing remains independent.

Native gRPC admits only data-frame flags 0/1; gRPC-web additionally admits its terminal trailer frame. grpc-web-text supports independently padded base64 chunks. Its offsets refer to base64-decoded body bytes; other offsets refer to retained body bytes. Every field has a body-scoped field_id derived from its byte offset, a frame_index, and a parent_field_id for schema-driven nested messages or groups. Repeated field numbers remain separate occurrences. Trailer blocks are parsed as HTTP field bytes. Valid UTF-8 values are retained; when a field contains opaque non-UTF-8 bytes, that field is omitted and the trailer frame is partial, while other valid trailer fields remain available. The retained body and frame byte range preserve the original opaque value.

Without a schema, varints and fixed-width values retain exact unsigned decimal representations. Length-delimited UTF-8, nested-message and packed readings are candidates; REA does not choose one or invent field names or signedness. Nested candidate ranges remain accessible without recursively expanding them. Malformed messages retain the parsed prefix and an exact error offset. Compressed frames and encoded HTTP bodies remain unknown; they are not silently treated as empty messages.

For schema interpretation, supply a local compiled FileDescriptorSet and its fully qualified message type. All imports must be included. REA never fetches schemas or executes the selected application. Descriptor identity and parser version remain in the body facts. Scalars, repeated numeric packing and nested messages are interpreted alongside the unchanged wire facts; disagreements remain explicit. Descriptor admission rejects duplicate field numbers within each message, including nested messages and oneof fields, before schema annotation. This does not reject repeated occurrences in captured wire bytes or reuse of a number in a different message. Fully qualified message, enum, service and extension names must also identify one declaration across the descriptor set, including nested and synthetic map-entry messages. Different packages or enclosing messages may reuse a local type name. Schema-admission failures retain the partial capture Evidence and its original artifact identity. REA also requires an unambiguous descriptor file identity. Repeated exact file names are coalesced only when protobuf-es equality confirms matching field presence, declared values and retained opaque unknown/custom-option fields. Opaque fields must match in their retained order and bytes; REA does not interpret them. Conflicting definitions are refused rather than selecting a file by its position. This is REA's supplied-schema admission policy, separate from protobuf wire validity and the first-occurrence precedence in protoc v32.1. The descriptor-set byte count and SHA-256 continue to identify the complete original input, including repeated file facts. Map semantics, extensions and required-field validation are unsupported. A supplied schema does not establish the sender's actual schema. An invalid nested interpretation records schema_status: "invalid-message" and schema_error locally; later outer fields remain available. Valid string contents, including a leading U+FEFF, are preserved. Descriptor wire failures name protobuf.schema.descriptor_set_path with the wire reason and byte offset; a missing selected type names protobuf.schema.message_type. Duplicate field numbers in a message are refused before registry construction, independently of declaration order. Resource exhaustion and cancellation keep their own outcomes and previously admitted capture Evidence remains available.

# Compile your own schema with your existing protoc, including local imports:
protoc --include_imports --descriptor_set_out=/evidence/api.pb api.proto
rea inspect-web-network-capture /evidence/session.har har --record 3 \
  --protobuf-schema /evidence/api.pb --protobuf-message example.Response \
  --detail summary --evidence_output /evidence/session-evidence.json --json

MCP accepts the same choices as protobuf: {schema: {descriptor_set_path: "/evidence/api.pb", message_type: "example.Response"}}, with explicit record_ordinals. Optional body_pointers batch-select exact binary pointers; framing can explicitly select raw, grpc, grpc-web or grpc-web-text when MIME metadata is unavailable. CLI exposes --protobuf-body-pointer and --protobuf-framing. Filters never invent bytes excluded by source redaction. Decoding an extra base64 layer also checks explicit exclusions; newly recognized trailer credentials exclude the entire affected binary sidecar and its reported encoding, preserving a redaction coordinate instead of a credential digest. Credential inspection covers every selected body before descriptor reads or field expansion. Later capacity or schema failures can therefore retain the sanitized source even with explicit sensitivity options. If that source pass is interrupted, unchecked bodies are not published as partial Evidence. Framing overrides apply only to format-owned body coordinates and preserve exclusions for the transport declared in MIME metadata. Explicitly excluded or conflicting MIME metadata is not treated as absent: automatic framing leaves such a non-empty body uninterpreted, adds a record limitation and keeps protobuf_coverage partial unless framing is selected. Unknown or explicitly excluded HTTP coding metadata is not treated as absent: SAZ/native bodies remain opaque until identity coding is established. Raw header values and retained body bytes remain available in source Evidence. Skipping wire analysis still runs the independent credential check. If HTTP coding metadata is unavailable and that check cannot complete, the operation returns unavailable without publishing unchecked capture bodies as Evidence.

Use detail: "summary", then inspect_analysis_view with its retained parent:

{"kind":"protobuf","collection":"bodies","offset":0,"limit":16}
{"kind":"protobuf","collection":"frames","record_ordinals":[3],"offset":0,"limit":16}
{"kind":"protobuf","collection":"fields","record_ordinals":[3],"field_numbers":[1,4],"fields":["number","schema_name","schema_value"],"offset":0,"limit":32}

Filters intersect; an empty filter selects nothing. Field pages also accept field_ids, frame_indices and schema_names. Default field pages omit text, schema_value and enum alias inventories; select these columns when needed. Identity, source offset basis and body status survive column selection. Complete Evidence retains all decoded values, and ordinary capture previews expose counts. Page exhaustion covers retained facts only; a partial source does not establish absence.

Authenticated immutable parents reuse validation and the most recent field filter per capture. These caches live only with retained Evidence, hold references to existing facts and never reopen either file. Cached selector text is bounded by retained field cardinality; larger selections use an uncached scan. Mutable input is revalidated. The existing 96 MiB report budget includes the original decoder reply plus conservative charges for new facts, strings and descriptor expansion. The explicit protobuf_coverage reports selected and inspected record counts, preserving partial decoding through composed views. Capacity failure retains the source observation as partial Evidence; cancellation yields between bounded work batches. Real elapsed time and process-family memory still depend on the selected capture and schema.

Fiddler Everywhere MCP and public skills

For agent-driven traffic debugging, connect Everywhere's MCP server separately from REA. Run a current Everywhere release with MCP access (Trial, Pro, or higher), internet access, and a compatible MCP client. Use Settings > MCP Server > Coding agent, or Manual setup with the generated API key. The default endpoint is http://localhost:8868/mcp. See Telerik's MCP setup guide and agent-skills prerequisites.

Telerik's public skills cover installation (fiddler-download-setup), connection setup, and traffic analysis:

npx skills add telerik/fiddler-agent-tools --skill fiddler-mcp-setup
npx skills add telerik/fiddler-agent-tools --skill fiddler-traffic-debugging

After setup, reload the agent client and use the connected server's advertised tool names and schemas. Check status, list sessions, then inspect relevant session IDs. Set sessionsSource to LiveTraffic for HTTP(S) traffic or AgentCalls for enriched LLM calls; active filters affect the session list. For an existing run, skip the skill's pre-capture steps that replace process filters and clear sessions. The documented MCP session sources do not provide a saved-snapshot selector; use the SAZ workflow above for saved files. REA does not register or wrap the Fiddler MCP server.

Reading the evidence

• artifact identifies the caller-selected file and stable snapshot digest.
• records preserve producer order/ordinals and original JSON pointers, byte ranges, or SAZ member names. No live browser transaction or scenario IDs are invented.
• reported retains the producer's fields. HAR container metadata is returned separately, with log.entries replaced by null and records_pointer identifying the record collection. This is an inspection representation, not a reconstructed HAR file.
• numeric_literals preserve HAR number lexemes and native integer/float representations, including unsafe integers and non-finite native floats. Values that cannot be safely represented as JSON numbers are null in the reported view. Schema-valid HAR numbers are admitted even when they cannot be published safely as JSON numbers; their exact lexemes remain in numeric_literals. Standard numeric fields must still satisfy the pinned schema, including integer requirements judged from the numeric lexeme.
• binary_fields retain exposed bytes, base64 and SHA-256 independently of declared sizes. Valid HAR content.encoding: "base64" is decoded exactly; producer-decoded Unicode text, missing text and unknown encodings do not establish original body bytes. Response readers expose decoded text only when encoding is absent or supported base64 bytes decode as UTF-8. Other declared encodings remain opaque: original text and encoding stay in Evidence, while replay and state extraction treat the decoded body as unavailable. A redacted encoding declaration does not establish plain text. See the HAR content representation. Post-data parameters do not reconstruct a multipart body. Native byte fields also have an exact UTF-8 display view when valid; binary sidecars establish their actual representation. Native missing content and empty content remain distinct.
• redactions identify known transport authentication exclusions and explicit sensitive-value exclusions. Redacted byte fields retain neither original bytes nor their digest. Unknown extension names and token-looking query parameters do not establish sensitivity. HAR authentication fields, transport URLs and base64 content are interpreted only inside log.entries. Similarly named unknown fields in the document container remain opaque evidence; explicit sensitive-value declarations still apply to both scopes.

Mark sensitive literal string/UTF-8 byte values with MCP sensitive_values, or repeat CLI --sensitive-value. These declarations are never persisted in Evidence parameters; the count is retained. Explicit text/UTF-8 byte matches exclude the entire field with a reported null rather than substitute a text marker or rewrite payload bytes. This prevents replacement markers from reintroducing declared literals. The file digest identifies the original artifact unless it matches an explicit declaration; in that case artifact.sha256 and Evidence subject are null with an explanation. A sensitive property name excludes that entire property and its sidecars; a scope: "property-name" redaction points to its actual parent. No substitute property name or child pointer is invented. Omitted subtrees are still validated, including canonical HAR base64. Explicitly sensitive artifact paths are empty in Evidence and omitted from locations, while the original SHA-256 and size remain available unless the digest itself is explicitly excluded. Failure messages, diagnostics and cleanup resources follow the same explicit declarations without changing error types; sensitive issue pointers identify a real ancestor instead of a fabricated coordinate. Header/cookie authentication values and known transport URL userinfo are excluded structurally, including native backups. HAR redaction and API-surface projection share the same recorded-URL parser. Canonical userinfo spans are removed without rewriting the remaining source text. Credential-bearing spellings without a safe raw span, such as slashless or extra-slash HTTP URLs, are excluded as a whole transport field. Ordinary extension text and noncredential path/query values remain unchanged.

Generated free-text limitations follow the same literal declarations, including global, per-record and Evidence explanations added during projection. Contract field names, typed discriminators and exact decoder identities keep their canonical meanings; declarations do not rename formats or providers.

Historical evidence does not prove runtime attribution, execution or deployment authenticity. Mitmproxy messages are upstream-assembled messages, not recovered wire fragments. Native dictionary duplicate keys have the unchanged upstream decoder's last-value semantics; this adapter does not claim to recover discarded duplicates or the original numeric spelling of parsed native floats.

Derived API surface

Pass retained HAR Evidence to inspect_analysis_view with view: {"kind":"api-surface","offset":0,"limit":50}. Omitted offset and limit are 0 and 50. The view groups request lines by method, WHATWG URL origin and path template, sorted by origin, template and method. Query strings, fragments, credentials, headers and bodies are not copied into groups.

Select include_query_parameter_names: true in the view to include distinct observed query names per group. The default and explicit false omit the query_parameter_names field. Selected groups include an array of {name, record_ordinals} entries, sorted by Unicode code point; each name's ordinals follow the parent's selected record order and occur once per record. For /products?category=books&page=1 and /products?category=games&sort=price, names are category, page and sort. Query values and fragments are not copied into this inventory.

Names use WHATWG URLSearchParams form-query decoding: literal + becomes space, %2B becomes +, percent escapes decode as UTF-8, and empty names are retained. Empty query pairs are ignored. Original URLs stay in parent Evidence; name decoding does not normalize the recorded paths used for grouping. The inventory describes observed names, without inferring required parameters, value types or a complete API schema. Sensitive-parent refusals still apply.

Rule	Exact segment format	Placeholder
numeric	ASCII digits only	{numeric}
uuid	RFC 9562 textual UUID, any version, case-insensitive	{uuid}
hex	At least 32 hexadecimal digits	{hex}
iso-date	Valid calendar date YYYY-MM-DD	{iso-date}

Rules apply in table order to the recorded, still-encoded request path, read before WHATWG dot-segment removal, so /a/%2e%2e/admin and /admin stay distinct. Segment indexes start at zero after the leading slash. Each matched segment lists its observed literals and basis: "inferred". templated_groups counts templated groups across every page. Because paging follows the full grouping, every page is inferred Evidence when that count is nonzero, and derived Evidence only when no group was templated. Version prefixes, short opaque IDs and digit-bearing route names stay literal. Samples merge only when method, origin, and exact segment rules match. The view does not infer further cross-sample patterns.

Groups cite parent record ordinals and distinct statuses and MIME types (lowercase type/subtype without parameters). An invalid type/subtype essence is unknown; parameters do not affect this projection. Flags report single samples, mixed statuses or mixed content types. coverage pages groups; records separately reports examined, total and excluded parent records. Ordinal subsets carry a limitation. Null URLs are excluded as url-redacted; unparseable URLs, URLs without an origin, spellings without scheme://authority (such as https:host/path), and authorities containing a tab or newline as url-unparseable. Unavailable request objects or methods are explicitly excluded. Unavailable response fields, including HAR's status: 0 and an empty mimeType, remain unknown, not invented statuses or MIME types.

The view makes no handler-identity, schema or causality claim. It performs no network activity. Native mitmproxy, SAZ and qlog captures return a typed unsupported error. Parents inspected with sensitive_values, or whose Evidence does not record sensitive_value_count, return a typed unsupported error whose reason is the canonical token sensitive_values_unretained. The view would write path templates and limitation text that it cannot check against declarations the parent does not retain, so the refusal does not repeat declaration text.

One accumulator owns group identity, observed-value/status/MIME sets, ordinals, exclusions and deduplicated limitations. It scans path spans without splitting literal paths into arrays. Before schema parsing, it reserves the capture clone and validation scratch by walking the parent without cloning it. Inline requests borrow the Evidence value through CLI/MCP envelope validation; the application admits the entire portable envelope before authenticated parsing or hashing. An evidence-file source is read and authenticated first and then admitted like retained Evidence, so its strings are not charged as inline envelope text: a capture too large to project inline can be projected from its file. Before authentication REA checks that live heap headroom holds its copies of the parsed Evidence, and over MCP the read follows the request's cancellation. An invalid Evidence file returns EvidenceFileError with the selected path and content_issues relative to the file's JSON root; those locations do not refer to the request's source.path string. The envelope clone, hash scratch and capture clone stay charged during grouping. Temporary URL/path/MIME/query work and retained state must fit together within a 96 MiB projection representation budget. A whole group or merge is admitted before allocating its retained sets, map entries or array values. Fixed group storage, distinct text, exclusions and limitations count as well as matched segments. Per-record reservations are released after each record; repeated values do not accumulate another retained charge. Only the requested page is materialized as output groups. Opt-in query projection reserves parser tuples and the per-record name set before decoding, then admits distinct retained names and source ordinals before changing a group. Repeated names within a request retain that ordinal once.

The representation budget uses conservative storage and worst-case JSON-escaping allowances; it is not an enforced V8 heap or process-memory limit. It is separate from the decoder's actual reply-byte limit and MCP transport admission. Exceeding it returns resource_constraint without a partial view. Limitations on the parent envelope, capture and examined records are deduplicated onto the view. Portable captures must retain unique ordinals within their declared total; caller-selected subsets can preserve any requested ordinal order.

Environment manifest

Pass retained HAR Evidence to inspect_analysis_view with view: {"kind":"environment-manifest"} to project one application's recorded request lines into a replayable tool surface: the environment manifest that an agent environment later serves from retained Evidence. The view reuses the api-surface grouping, then keeps the groups on the primary origin, the origin with the most JSON-like records (application/json, application/*+json, text/plain, text/markdown), whose path starts with /api/ or /apis/ or whose content types are JSON-like, and the write groups answered with no body (a bodiless 204 DELETE). Empty retained response text establishes a bodiless write only when its encoding is understood and unredacted; an opaque or hidden encoding leaves the body unknown. Every other group is listed under excluded with reason other-origin, ui-page-or-asset or non-json, and retained records without a usable request line are counted in a limitation.

The parent may instead be capture_browser_scenario Evidence whose scenario selected response bodies or header values: each network transaction is then one record, in request order, and the result names parent_operation: "capture_browser_scenario" with an empty artifact, since a scenario records no file. See agent environments.

{
  "name": "inspect_analysis_view",
  "arguments": {
    "source": { "kind": "retained-evidence", "evidence_id": "ev_..." },
    "view": {
      "kind": "environment-manifest",
      "template_rules": [
        "after-uid-literal",
        "opaque-token",
        "sibling-variance"
      ],
      "include_query_parameter_names": true
    }
  }
}

origin and name identify the environment; the name is the kebab-case primary host with dots and ports replaced by hyphens, grafana-example for http://grafana.example. Each tool carries a deterministic name, method, path_template, api_surface_template (the four-rule template of its earliest record), params, query_parameter_names, body_shape, statuses, content_types and record_ordinals. records lists one pointer {ordinal, method, path, status, content_type} per record a tool cites and never a header, cookie or body. Tools sort by template then method.

Without template_rules, every path_template equals the api-surface template and params carry the four format rules with names id, uuid, hash and date; a repeated name gets a numeric suffix. Opt-in rules apply only when listed, each at most once (a repeated rule is an invalid_request), in this order:

Rule	Applies to	Parameter
after-uid-literal	The literal segment right after a literal uid, unless it is a stop word	uid
opaque-token	A lowercase alphanumeric segment of twelve or more characters mixing letters and digits that is not a v1-style version	uid
sibling-pair-variance	Groups equal in method, origin and segment count that differ in two adjacent literals right after a plural literal at index 2 or later and followed by a literal, with at least three distinct pairs and both positions varying (/repos/{owner}/{repo}/issues); stop words and v1-style versions never qualify	singular plural literal plus _namespace, then _name
sibling-variance	Groups equal in method, origin and segment count that differ in one literal at index 2 or later; the siblings whose literal is identifier-like (a digit or :, or three or more hyphenated names under a plural literal) merge, never a stop word, and the other literals stay their own tools	singular previous literal plus _id, else id
prefix-variance	A literal at a position that another tool, of any method or segment count, templates by one of the rules above after the same preceding segments, when the literal is one of that parameter's observed values, carries a digit and has the length and character classes (digits, lower case, upper case, each other character) of one, or has the single length of eight or more characters every observed value has and only their character classes; stop words and v1-style versions never qualify, and tools that now match merge	the name of the parameter it shares

With sibling-variance selected, literal siblings within its comparison scope fold into existing opt-in parameters as observed values unless they are stop words. For example, select both opaque-token and sibling-variance so /api/folders/{uid} lists general beside the recorded token. With only opaque-token, /api/folders/general stays a separate literal tool. A position templated by one of the four format rules absorbs no literal, so /api/teams/{numeric} keeps its exact-format meaning and /api/teams/search stays its own tool. Opt-in placeholders use the parameter name; format-rule placeholders keep their api-surface spelling. Tool names are <method>_<literals> with api, apis, namespaces, default, v-prefixed version tokens and a uid literal before a parameter dropped, dotted literals cut to their first label, _by_<params> appended when parameters exist, sanitized to [a-z0-9_], at most 60 characters and made unique with numeric suffixes: get_dashboards_by_uid, get_plugins_settings_by_plugin_id.

prefix-variance runs last so it can share the parameters the other rules produce: with sibling-variance, a PATCH or DELETE of the one entity a session edited, /api/v1/collections/0RZ576QZQA5KG, becomes /api/v1/collections/{collection_id} beside the reads that varied there, and so do deeper routes such as /api/v1/collections/{collection_id}/books. Its parameter is labelled prefix-variance, unless a merged tool templated the position by its own rule. When parameters of different rules share a position, a literal shares the one that observed it, else the earliest one it is spelled like.

sibling-variance, sibling-pair-variance and prefix-variance compare routes of at most 32 segments; longer routes keep their grouped templates and the manifest says so in a limitation. body_shape is derived for POST, PUT and PATCH from the tool's earliest record when it carries an application/json or application/*+json request body: top-level keys and their JSON type names, never values and never nested structure; a body on a later record adds nothing. unknowns always states three facts: the authorization model, response semantics and state transitions are not established by a capture. templated_groups counts tools and excluded groups with a templated segment; whenever it is nonzero the Evidence is inferred, otherwise derived. The view performs no network activity and replays nothing. Native mitmproxy, SAZ and qlog captures, and parents inspected with sensitive_values, return the same typed unsupported errors as the api-surface view, including the canonical token sensitive_values_unretained. A capture_browser_scenario parent is not refused for declaring secrets: the capture redacted them as it recorded, a transaction whose URL was redacted is left unrepresented, and a response body keeps the capture's redaction markers, which a manifest limitation counts.

The manifest accumulator shares the api-surface admission model: the capture clone, the inline envelope (for an inline source), record pointers, groups, every route part, observed values, query names and body-shape keys are admitted before allocation, and each recorded body is charged as temporary bytes before it is parsed. Retained state and temporary work must fit together within the 96 MiB projection representation budget; exceeding it returns resource_constraint without a partial manifest. Name a large capture with {"kind": "evidence-file", "path": "/abs/capture.json"} rather than inline: response bodies the manifest never reads are then charged once per value in the capture clone, not per character. A capture_browser_scenario parent is also charged, before validation, for checking its largest body and for the decoded text of its request bodies.

Environment state

Pass the same parent to inspect_analysis_view with view: {"kind":"environment-state"} to extract the initial state the capture's reads observed, and a render plan per read tool that reproduces those reads from it. Where the environment manifest names an application's tools, the state holds its data: entities by type and identity, plus documents for reads that are not entities. Use the template_rules of the manifest the state accompanies so tool names match; the view groups records the same way.

{
  "name": "inspect_analysis_view",
  "arguments": {
    "source": { "kind": "retained-evidence", "evidence_id": "ev_..." },
    "view": {
      "kind": "environment-state",
      "template_rules": [
        "after-uid-literal",
        "opaque-token",
        "sibling-variance"
      ]
    }
  }
}

Records on the primary origin are ordered by recorded start time, or by record order when any start time is missing (timeline), as for a capture_browser_scenario parent. A POST, PUT, PATCH or DELETE with a 2xx or 3xx status, or no status, may change state, unless it is a POST whose path has a segment such as search, query, list, find, fuzzy, count or stats, which queries state instead. A type's state comes from its successful reads that finished before the first such write under the same root: the path's API base (the literal segments every tool sharing its first segment has), any v1-style version segments, then one more segment. A read that overlaps that write in time is left out. A write whose URL a scenario redacted has an unknown target, so it ends the state of every type, and a limitation counts such writes. Values from a capture_browser_scenario parent keep the capture's redaction markers in place of declared secrets; the state is not refused for them, as the manifest is not.

Each GET tool whose template ends in a parameter is an item tool, and its type's collection_template is the template without that parameter; any other GET tool is a list tool of its own template. Interior empty segments remain meaningful: /api/books and /api//books name distinct collections. A single trailing slash names the same collection. An item tool proves its type's id_field when every initial response carries the requested path value in the same field of one entity object (id_basis: "path-parameter"). The object is the response itself or one wrapper member; failing both, it is the only element of a one-element array at the root (pointer /0) or in one member (/<member>/0), as in {"posts":[{...}]} or {"data":[{...}]}, provided one of the identity names below repeats the value, since a filtered list may also answer with one element that repeats its parent's value. Otherwise the field is the first of id, uid, uuid, _id, ID, UID, Id, guid, key, slug and name that every listed item carries as a distinct string or number; when none does, it is the first key of the first listed item, in its order, whose name ends in Id or _id (such as orgId or user_id), that every item carries as a distinct string or number, and that at least one initial response shows distinct across two or more items (id_basis: "field-name"). A single listed item is distinct from nothing, and its suffix-named fields are often foreign keys, so a list that only ever held one item is left to a document. Empty initial lists constrain no field. A list is the response array, or the one array of objects in an envelope that is not itself an entity, preferring an array whose items an identity name identifies, then an array of two or more items a suffix-named field identifies.

entities merge every initial observation of one identity within one collection path: an item view's latest values win over a list view's, a view whose latest values differ keeps them in view_fields, and a field whose value changed between one tool's observations with no recorded write between is named in the type's volatile_fields, keeping the latest value. A value that changed across a write under another root may be that write's effect, so the type's initial reads end at the earliest write within the change, reads that overlap it are left out, and the state is extracted again once; the same holds for a value one tool read before a write and another tool read after it, and for repeated reads of one full request line whose list membership changes across a write, including transitions to or from an empty list, or whose entity fields are added or removed. A type whose reads still disagree keeps every tool a recorded view with reason fields-differ. Different query strings remain distinct request lines. Sparse responses do not erase the last observation of a field supplied by another request line. Entities are ordered by type and collection, and within a collection as the first line that lists every entity any initial read shows in it and names no sort, direction or random parameter (random, rand, shuffle, seed) lists them; a collection without such a line keeps its entities' earliest sightings. That order is a list's state order. Every entity cites its record_ordinals.

views lists one plan per read tool in tool order:

Shape	Plan
item	The entity object at pointer (the root, one member, or /0 or /<member>/0), the ordered fields the view shows and the required_fields every recording had (the fields include those only reads after the type's first write show, of entities that hold its identity, when no initial entity holds them, as Vikunja adds subscription to the tasks it creates; they are never required), and envelope_ordinal: the recording whose response, with the entity replaced, every item response shares, leaving out the per_response_members each response there mints: a top-level UUID or clock that differed on every repeat of every concrete path the view answers, with no write between, and never stayed the same on a repeat elsewhere; list views name theirs too
list	The array at pointer, its fields, membership (all: every entity of the collection in state order, when each recording listed exactly that, had no query or only paging parameters (limit, offset, page, per_page, perpage, page_size, pagesize, maxsize, max_results, maxresults, size, start, _start, _end, ignoring case), no item-total member (total, count, totalCount, item_count, totalcount, itemcount, numberOfElements, but not total_pages or per_page; a number or a string of digits; top-level, or in a top-level object beside the list such as pageInfo.totalRows) differing from the number listed, and no has_more, next or next-page token member set; recorded: the entities each recording listed, except on the lines all_record_ordinals names, which each meet those conditions on their own and list every entity), counters that equal the number listed, and listed_ids: top-level members that list two or more listed entities' identity values in list order (Paperless's all), which a rendered list sets to its own unless it is a page of more matches
document	The latest initial value of a list tool whose responses hold no identified list, when its initial reads share one request line; held in documents with that path, whether or not the capture writes it, unless its value changed across any recorded write between those reads (fields-differ); its per_response_members are the members of the value each response on that path mints
recorded	A reason the tool replays its recordings instead

An item or list plan is kept only when it reproduces the latest initial recording of each request line from the state, comparing JSON values without regard to member order; a volatile field compares its recorded value. Each plan lists the checked_record_ordinals. A request line that shows the entities in another representation (an expanding or reducing query, or one asking for some fields only) is left out of those checks when a query key tells it from the others: the plan lists its latest recordings in projected_record_ordinals and the keys in projected_query_keys (name, with a value or absent: true), and serving replays calls matching them and calls asking for fields the checked lines do not show. Otherwise the tool is a recorded view with reason envelope-varies, fields-differ or list-differs; the other reasons are query-body (a query POST, or one the capture shows only reading), no-init-recording, body-unusable, no-identity (an item read without a proved identity), path-varies (another read under more than one path, as a template with a parameter before its end), query-varies (another read of one path under more than one query) and rendered-oversize (a document value whose JSON renders to more than 256 KiB). A document holds the value of its one path, so for such a template it says nothing about the template's other paths. A body is usable when its content type is JSON-like, it is UTF-8 text of at most 256 KiB, and every number survives native conversion exactly. JSON member names, including __proto__, remain ordinary own data members through extraction and serving.

A list view also learns, from its checked lines that do not list the whole collection, what each query parameter does, and names the result in query. Each such line is explained by the first combination of roles that lists exactly what it recorded from the collection's initial entities:

Role	Parameter	Effect
filter	any other, on a field every listed item holds the value in; one not named like the field (shelf, or shelf_id, for shelf) stands only when lines giving it two different values agree on it	entities whose field equals the value as text, or whose array field holds it directly or as an object's id, slug or name
search	q, query, search, searchTerm, term, keyword(s)	entities with one of fields containing the value, ignoring case; the fields are those it matched in the listed items, URLs only where nothing else matched; a term that matched no field leaves the parameter inert until another line's term does
sort	a name asking for an order in any common spelling: sort, _sort, sort[], sort[0][field], sorters[...], sortBy, sort-column, sortKey, order, orderBy, ordering	ordered by the field the value names (name, -name, name:desc, name-ASC): numbers by value, text ignoring case, null last ascending, ties in state order
direction	a direction name (dir, direction, sortDir, sortOrder, sortDirection, order_direction, json-server's _order, and order) given asc, desc, ascending or descending; given anything else, order is a sort and the others are ordinary parameters	the sort's direction
limit	limit, per_page, perPage, page_size, size, count, take, itemsPerPage, maxSize, max_results, maxResults	page size; zero or a negative value lists every match
offset	offset, start, _start, skip	matches skipped before the page
page	page	page number, counted from first (1 or 0, whichever explains the line)
inert	any that names no order or direction; never a true reverse, or a paging value that is not a safe integer	changes nothing for its recorded values; other values are not computed

No two parameters of one call take the same sort, direction or paging role, and a line with more than 12 query parameters is left unexplained. Lines that disagree on a parameter's role drop it, inert values gather and search fields unite. query.record_ordinals are the lines the merged roles reproduce, envelope and fields included, and query.totals the envelope members that count every match before paging on each of them, top-level or one level down in a paging object (/pageInfo/totalRows); a number-of-items member counts the page. The roles are inferences from the recorded lines; a line they do not reproduce keeps its recorded members.

writes lists one plan per tool whose request may change state, in tool order. Its effect is create for a POST to a type's collection_template, update for a PUT, PATCH or POST to that template plus one trailing parameter, or to that item path plus one action segment, delete for a DELETE to it, document for a PUT, PATCH or POST to a document's path, an update or delete of each entity a write at the collection names by id in its body, a create for a POST whose answer names an id (at its top level, in an object member or in the one object of a one-element array member) that a later item read of a type first shows (served only with the path values its recorded call sent, unless it creates into the type's own collection, whose path has no parameters, and a path rule stores each of its path values), and otherwise none with reason no-state-target. A plan is built from the tool's first successful call: its request body must be a usable JSON object (as response bodies are usable), and the written entity must show in that call's response or in a later read before the next write. Each field of a created entity gets the first rule that explains it: generated identity (a sequence, UUID or token, by the recorded value's shape), echo of a body member (sorted when the application stored its strings in code-point order rather than as sent), copy of a body string of 8 or more characters, a path value, a slug of a body member, a clock value within 10 minutes of the request in its recorded format, a sequence one past the largest value of its collection or type, and never below its start, a unique UUID, opaque token or integer of 10 or more digits that no entity of the type holds (where every one holds a distinct integer of the same length, as time-based ids do), a nested-clock object whose clocks the request stamped, a made-object no other entity of the type holds otherwise, whose clocks the request stamped and whose other members hold the entity's own identity (own), one past the largest value held objects show (counts, each from a start), or what every held object shows, a template around a generated value, a reference to an entity, or a composed string; other fields keep the observed value. When the first read of a created entity before the next write shows fields the create's answer lacks, its clocks to another precision, or values the answer had not loaded (an empty, zero or false value, Go's zero time, or an object whose members are each loaded so), the observed entity takes them (observation.read, whose unloaded points to the values the answer had not loaded). An item read that agrees with the answer otherwise lends its fields and clock spellings and no longer confirms the create; one that disagrees or omits an answer field its item view renders still checks it. Clock agreement uses the coarser recorded precision, including microseconds. A list read only adds fields the answer lacks and still checks it. Serving answers a create in its answer's shape and clock precision, with empty lists and maps the call did not send spelled (null, [] or {}) as its answer spelled them and the values it had not loaded as it answered them (unless the call sent one the plan stores as sent), leaving out members the read added inside an object the answer showed, and stores the fields of both, as the item read-back shows them, with its clock precision. Updates answer empty lists and maps they did not send as their answers spelled them too; a member a call sent is answered as sent. An update or delete that answers with the entity leaves out a field only later reads showed, which no initial entity holds, when the entity does not hold it, as Vikunja's reads of tasks it did not create show no subscription. A reference is an object holding fields of the entity a sent member names, as the state holds it (a parent beside parentId): the member is accepted with any value, serving looks the entity up and refuses a call naming one the environment does not hold or holds in more than one collection, and the field is absent, from the entity and the create's answer, when the call does not send the member. An id naming entities of one type in several collections teaches no reference. Each later create widens its plan when, explained on its own, it agrees with it: both calls show the same fields but for a reference whose member a call without it did not send (one an earlier create added stays), every field both rule they rule alike (an echo one showed sorted stays sorted, unless a call stored its strings out of order as sent), and a field neither rules holds one value, unless the later call's rule for it spells the earlier calls' value too: a composed or template string, or a sequence started at that value (start), as Vikunja numbers a project's tasks past those deleted before the capture. Two made-object rules agree in their clocks and own members, and a count only the later call shows starts at the first object's value; a later create's made object must hold clocks its own request stamped, in the rule's format. Members either call sent are accepted, only those every call sent are required, and an echoed member a call left out takes the value that call stored (body.defaults). A member a create sent empty ("", null, false, [] or {}), as a form sends a field left as it was, is not required: a call that leaves it out is applied as sending that value (body.blank), unless a call that left it out showed a default, which it then keeps. A call may send it as any recorded call sent it empty, even after a default is learned, and one leaving it out is applied as the earliest did, as when one create sent false and another null. A clock rule widens to trimmed only when no held entity or earlier call showed that clock at a fixed width, and a slug to spelling & as the word (ampersand: "and") only when no call the plan accepted nor entity the type held dropped one. A later create that disagrees leaves the plan as it was. A field the first create left unexplained, such as an address whose two parts were the same text, takes the composed rule a later create learns for it when that rule spells the first create's value too, and the first create's body contract is learned again with it. A composed string or template a later create's answer shows otherwise or leaves out is unexplained-change. A create whose observed reference differs from the entity the plan rebuilds, whether or not it sent the reference member, is unexplained-change. So is a create that left out an echoed member yet shows a value different from the learned default, or that stored a field no rule writes otherwise than the plan keeps it. A composed string is literal text and parts read from the fields other rules write: a field's value, its slug, or a clock's calendar digits (YYYYMMDDhhmmss, YYYYMMDD, YYYY-MM-DD or YYYY/MM/DD, in the clock's zone). A slug drops & unless the application spells it as the word, as Go's gosimple/slug and JavaScript's slugify do (ampersand: "and", on a slug rule or part). When a create slugs no &, the entities the type holds show which, and no slug rule is learned when they spell it both ways; a create whose slug spells & otherwise than they do, or alongside one whose slug neither spelling gives, has no slug rule; such a held slug also keeps a later create from widening the spelling. An update learns the spelling the same way: a later update whose slug spells & as the word refines a rule an earlier update learned from a name without &, unless an accepted call or a held entity dropped it, and every earlier update must still replay. An update whose slug spells no & takes the spelling the entity's earlier value and the type's other entities show, as a create does, and learns no slug rule when they spell it both ways. A create composes a string only when at least one other entity of the type spells it with the same parts and none spells it otherwise. Where one text could come from two sources serving could tell apart, such as an id that equals the title's slug, the entities the type holds must leave exactly one reading. A composed string that reads a field besides the generated value a template would rebuild it around replaces the template. A clock format is trimmed when the field's values differ only in how many fraction digits they show and none ends a fraction in a zero, as Go spells nanoseconds: its fraction is the most it shows, and serving drops trailing zeros and the point with them. Other clocks the plan stamps take that spelling too unless their own value shows a fixed width. A body that wraps one object in a one-element array is that object, wrapped so on every call (body.envelope). An update's members echo when stored as sent. A clock sent back as the entity held it and stored as a fresh clock near the call is a clock, and a member stored as the slug of another member stored as sent is a slug, both accepted with any value, as a whole-object write that sends what it read shows. A save that echoes an already-derived slug or holds a clock in its own second remains compatible with that derivation in either call order; earlier saves must also reproduce a newly learned slug. Any other member stored differently is accepted only with its recorded value (body.fixed) and stores the value the call that showed it recorded (the value of its constant rule), and a field it changed without sending (an empty list or map one answer spells null and another [] or {} is unchanged) must be a clock, a nested clock, an increment, a slug, a fresh unique value, a stamp of the acting user, a value derived from the one member it stores nowhere, a reference to the entity a sent member names (kept by a call that names the entity the field already repeats), a composed string whose parts read members it stored as sent or fields it left as they were and spell the value before the call too, or for an item action a set of a scalar to the value it showed, or the plan is unexplained-change; an increment, slug, stamp, derived value or set must hold for every call of the tool, and an action is served only once a call found what it sets already set. A write is served only with the query its first call sent, which serving applies to a call that leaves a parameter out and the catalog advertises as required. Every member the first call sent is required, except by PATCH and POST updates and members a create sent empty.

A read on a line query answers is checked against the query evaluated over the replayed state. It covers the collection only when every parameter sorts or pages and the answer holds every match; otherwise it covers the entities it lists, and those the line or the read listed that the state no longer holds.

The view then replays the timeline on a copy of the initial state. Modelled writes apply their effect; only a 2xx write is modelled or an example, and any other write (a redirect, an unanswered or unmodelled one) leaves the entity below whose item path it wrote unknown until a read shows it again (a whole collection when the path names no held entity), and taints what it wrote at or above. A read the served environment would answer from the state is compared with its recording as the read plans are, as the whole response it would serve: the entity or list rendered into the envelope of the item view's or the line's base recording, with the members the state still holds. A 400, 404 or 410, a null body, or an object marked deleted (deleted, isDeleted or is_deleted true, as EspoCRM answers a read of a deleted record) agrees with an entity the copy does not hold; such a read drops a held entity it names and adopts none, while an initial read keeps the record it shows held, as its item read answers it. The item view records the flag its reads set as deleted_flag, a field it shows even when only a query read showed it, which serving answers by: later reads of a served-deleted entity show it marked deleted, and a served delete answers with the flag as its recorded answer showed it. A match confirms the writes that changed its entity or collection since it was last read, less those whose entity a read since showed absent, which no longer count for the collection; a mismatch contradicts them, or, when an unmodelled write came after them, is checked only on the members they sent; a mismatch no write explains marks the related writes since as side-effect. A plan is served when a read confirmed it and none contradicted it; otherwise its reason is the first of its build reason (no-call, body-unusable, no-observation, unexplained-change), contradicted (with contradicted_by), side-effect, unconfirmed and incomplete-entities (an update or delete that answers with the entity, when no initial entity holds every field its recorded answer shows but those only later reads showed, counting for an update the fields its rules write, so it could answer only for entities the environment created). An otherwise served plan whose learned rules generate clock values is withheld with unknown-clock when no global clock start is established; its learned rules and confirming reads remain retained. checked_record_ordinals lists the confirming reads, insert places created entities (sorted by a field the collection's recorded order shows, else last), and write_clock is the end of the last timed record, to the next second, when all request completions are known; otherwise it is null. An update that writes the field its type's creates are sorted by moves its entity to where the new value sorts (its own insert); a bulk update moves each entity it writes in turn, as replay applies them. When a read listing the whole collection in its own order, not one its query sorts, fails after an update and the type's creates are not sorted by a field that update writes, the view tries the order serving gives when each write since a list in the collection's own order showed its entity moves it in turn by such a field, against the others' values before their own writes, when the entities no write has changed since a list in the collection's own order last showed them already follow it, as serving moves only the entities a write changes, and unless an earlier read listing the whole collection in its own order showed a write that changed such a field where moving it by that order would not have put it, a create was placed by a sort that excludes it, or a read listing the whole collection in its own order, not one its query sorts, that placed a create did not follow it; on a match, the type's creates and updates take that order, as Grafana keeps folders by title though no earlier read could tell that from creation order, or from id order where both ascend. An update moves its entity by its creates' sort only when a list read showed a create placed, in replay as in serving, so no read confirms an update by a sort a create only guessed from the initial order.

Values in the state are recorded data, so the view refuses the same parents as the manifest, including those inspected with sensitive_values, and a capture_browser_scenario parent must have selected response bodies. Types and identity are inferences, so the Evidence is always inferred. unknowns states that the state establishes no authorization model, no way the application computes responses or derived values, and write effects only as modelled from recorded calls. The state accumulator charges the manifest's admission and grouping, every record line, each body as temporary bytes before it is decoded and parsed, and every value it keeps, within the same 96 MiB budget. Documents hold whole recorded response values whose recorded text and rendered JSON are each at most 256 KiB, so a capture with many read-only paths yields a correspondingly larger state result. The view performs no network activity and serves nothing; rea environment-serve derives it from the capture at startup, serves read tools and the served write plans from it and loads replacement states through the AgentEnv data plane, as serve from the state describes.

Environment coverage

view: {"kind":"environment-coverage"} with the same template_rules reports what the state an environment derives from the parent would serve, without the state's entities: reads.state counts read tools answered from the state, reads.recorded those that replay their recordings with reads.recorded_reasons per view reason, writes.served lists the write tools whose effect later reads confirmed, and writes.withheld lists every other write tool with its reason, the record its plan was built from (exemplar_ordinal), the read that disagreed (contradicted_by) and a hint about what to record next. Hints are advice; they never change what is served. The view projects the environment-state view and summarizes it, so it refuses every parent that view refuses, with the same errors, and it is always inferred. Call it after each recording to see which writes an environment built from it would serve. It does not apply the serve-time admission checks: an environment whose served catalog exceeds the discovery budget, or whose state exceeds what data/add loads, serves replay, and rea environment-build then reports coverage: null; otherwise the build prints this summary as its coverage. See Build a bundle in one step for the hint per reason.

Resource and lifecycle boundaries

Complete evidence has a 32 MiB input, 96 MiB decoder reply and 64-level nesting budget. Those budgets have no runtime override. capture_timeout_ms does not raise them. Exceeding a budget returns no partial success. Oversized selected captures return an input error identifying capture_path and the byte budget; owned reply/output failures remain output or provider errors. An explicit complete-reply refusal names the 100663296-byte allowance and its lack of a runtime override. Record selection, paging, delivery mode and worker memory do not increase that allowance; a smaller producer export is a narrower observation, and the refusal leaves no completed capture Evidence to retrieve. Cancellation during reply-file finalization retains validated completed capture facts as partial observations. Malformed or unvalidated replies are not promoted. HAR, SAZ and qlog run in an owned Node old-generation heap with one V8 worker. REA_CAPTURE_DECODER_OLD_GENERATION_MIB is a decimal integer of mebibytes from 8 to 16384. Unset uses 192, the previous constant. Invalid values fail configuration and name the setting; they do not fall back to 192. The setting seeds the launch value. In MCP, set capture.decoder_old_generation_mib through update_analysis_settings for decoders belonging to newly accepted operations, without restarting REA. Changing the launch environment still requires a new process. A memory failure names the effective mebibytes, known source, default, floor, and maximum. The decoder passes that value as --max-old-space-size and removes inherited NODE_OPTIONS, so additional host memory and NODE_OPTIONS do not raise the heap. The 8 MiB semi-space stays fixed; a smaller old generation aborts Node during startup instead of enforcing a limit. Mitmproxy decoding applies Linux limits of 768 MiB address space, 30 seconds CPU and 96 MiB file output; both commands use capture_timeout_ms (default 30000) and independently supervised process cleanup. Owned decoder diagnostics retain at most 1 MiB across stdout and stderr; the collector drops overflow as it arrives, reports the limit failure and verifies cleanup. Heap limits alone do not establish aggregate RSS limits. SAZ and qlog refuse decoding when neither a Node heap limit nor a Windows Job Object bound can be enforced. Existing HAR availability is preserved. When REA runs on Bun, the HAR/SAZ/qlog decoder runs on Bun, which ignores these V8 heap flags; its limitations and resource failures then omit the V8 heap sizes. On Windows, the decoder's owned Job Object instead limits its committed memory to the effective REA_CAPTURE_DECODER_OLD_GENERATION_MIB value plus a 512 MiB reserve for Bun itself. The default 192 MiB setting therefore yields 704 MiB. The limitations state that limit, and reaching it returns resource_constraint with process_memory_limit_bytes, whatever exit status Bun reports. Other Bun hosts enforce no decoder heap limit, so raising the setting does not bound them.

Observed decoder memory exhaustion returns resource_constraint, with the selected capture, applicable limits and retained process diagnostics. Raising host memory or NODE_OPTIONS does not raise the old-generation heap. Update capture.decoder_old_generation_mib through MCP, or set REA_CAPTURE_DECODER_OLD_GENERATION_MIB before starting a new REA process. Retry after owned cleanup completes; the update does not release resources. See the settings design. Record selection happens after complete decoding and does not reduce that workload. Use a smaller capture exported by its producer or another decoder with sufficient capacity, retaining the original for provenance. Subsets do not establish complete-capture coverage. Repeating the same workload or running provider health diagnostics does not resolve the heap constraint.

Stable regular-file reads reject symlinks, replacement and concurrent changes. Private snapshots, declarations and tool configuration are removed before the result returns. Cleanup uncertainty retains the selected capture, resource and original failure information. The original capture is never changed.

Excluded coordinates and read failures

Explicitly marked values also exclude matching source coordinates and their sidecars. A hidden record coordinate is reported as location.kind: "unknown"; REA does not invent a replacement pointer. Original record ordinals and artifact digests remain available.

Host filesystem read denials return code: "access_denied", category: "unavailable", the actual EACCES/EPERM code and read-access guidance. Missing files remain input failures. Cleanup failures retain the prior structured error, including its input constraints or provider diagnostics, with explicitly marked text excluded.

An observed replacement or in-place change during stable artifact acquisition returns code: "artifact_changed", category: "integrity_mismatch" and retryable: true, retaining the selected path and reader's reason. Wait until the capture file is stable before retrying. No unstable snapshot is decoded. Command cleanup failures retain the original status and both diagnostic streams; the decoder's ownership preparation receives the same cancellation signal.

MCP advertises the same strict named input contract used by the application. Historical inspection delegates argument validation to that shared application boundary so accepted explicit-sensitive declarations also cover invalid argument names and other correction details. The SDK receives raw arguments through its Standard Schema interface; malformed inputs still fail before decoder or filesystem effects and return the normal structured REA error. Protocol-envelope validation remains owned by MCP.

Retained aioquic qlog

Select format: "qlog" through inspect_web_network_capture, or:

rea inspect-web-network-capture /absolute/origin.qlog qlog --json
rea inspect-web-network-capture /absolute/origin.qlog qlog --record 95 --record 0 --json

The initial profile is aioquic 1.5.0, JSON qlog 0.3: one JSON object with qlog_format: "JSON", qlog_version: "0.3", traces containing object events with name, finite numeric time and object data. JSON-SEQ, JSONL, columnar array events and other versions are unsupported. Duplicate JSON members, malformed UTF-8, invalid interpreted QUIC integers and excessive nesting fail without partial success, including in events outside the selected ordinals. Well-formed event subsets and missing frame metadata remain inspectable. Unknown extensions are retained; this is not validation of every qlog schema.

Choosing the format selects the interpretation profile; the file does not necessarily report or authenticate an aioquic version. The profile's clock uses Unix-epoch milliseconds, verified in aioquic 1.5.0's QuicLoggerTrace log_event/encode_time. The qlog.clock.basis and Evidence analysis profile state that assumption explicitly. Numeric lexemes survive independently of safe JSON-number projections; decimal spelling does not establish physical clock precision or monotonicity.

Every event has a global ordinal from trace/event source traversal and its original /traces/T/events/E pointer. This order is not a global time sort. Trace index and the selected artifact identify a connection observation; reported ODCID is preserved without claiming global uniqueness. Streams are scoped by trace, stream ID and sent/received direction. Control/QPACK streams are retained, so counting streams does not count HTTP requests.

qlog is a typed facet required exactly for this format. Its scope is all-recorded-events-before-selection: selection changes returned event records while facts cite the original ordinals across the entire recorded file. The container retains trace metadata with each events array replaced by null; there is no single events pointer shared by all traces.

Fact	Basis and limits
Original event and loss declaration	Observed endpoint report; recovery:packet_lost does not prove actual network loss. Missing packet number/type remains unknown.
STREAM ranges and byte totals	Derived per trace and direction from reported offsets and lengths using exact integer arithmetic. unique_bytes unions intervals; frame_length_sum includes overlaps; repeated_bytes is their difference. Overlap alone does not prove retransmission or equal contents.
FIN final sizes	Reported offset plus length where FIN is true; multiple conflicting sizes remain visible. A final size does not establish complete coverage or a received body.
Incomplete fact records	Ordinals with omitted/redacted interpretable frame metadata, including an unknown STREAM FIN; usable fields such as known ranges in that record survive. An empty list does not establish complete logging or capture.
HTTP bodies and causality	body_bytes: "not-reported"; lengths and headers are metadata, with no reconstructed body, hash, timing causality or live process attribution.

QUIC integer fields interpreted here accept exact nonnegative JSON integer values through 2^62 - 1, including equivalent exponent notation. Offsets plus lengths must fit that range. Derived sums are decimal strings and are not rounded to JavaScript numbers. A missing offset is unknown, never defaulted to zero. Ranges can start above zero or contain gaps. Packet headers, HTTP events, raw lengths and unknown extensions remain producer evidence rather than being normalized into invented HAR requests.

The adapter excludes authorization/proxy authorization/cookie/set-cookie values in known HTTP frame events, reset tokens in transport parameters and NEW_CONNECTION_ID frames, and NEW_TOKEN frame tokens. Known URL header userinfo and explicit sensitive values follow the shared exclusions. Ordinary extension fields and token-looking query text remain retained. Explicit declarations also cover normalized numeric projections, derived totals and trace pointers; facts that cannot be safely shown are omitted with incomplete-record coordinates. No keylog, PCAP, live URL or recorded target is opened by the qlog decoder.

To repeat the real-provider lane with local caller-supplied evidence from the educational aioquic 1.5.0 experiment (one origin trace, known 32,803-byte stream coverage and one 1,169-byte repeated range in the loss case):

npm run verify:web:qlog -- /absolute/control/origin.qlog /absolute/loss/origin.qlog /absolute/new-report.json

This lane checks built CLI/stdio MCP parity, SDK-converted JSON Schemas, artifact digests, original event count, selection scope, retained parent lookup, credential exclusions and the known control/loss facts. The report is created exclusively and contains no keys or captured credential values. It is a fixture-specific verification, not a general loss detector. CI uses synthetic format inputs for malformed cases and limits; those tests do not establish producer provenance. PCAP parsing and correlation remain outside this stage.

Focused qlog views

Use MCP inspect_web_network_capture with format: "qlog", detail: "summary" for identity, profile, scope and retained-fact counts. Then pass its normalized_result.parent_evidence_id to inspect_analysis_view as a retained-evidence source.

{"kind":"qlog","facet":"traces","offset":0,"limit":10}
{"kind":"qlog","facet":"streams","trace_index":0,"offset":0,"limit":10}
{"kind":"qlog","facet":"contributions","trace_index":0,"stream_id":"0","direction":"sent","offset":0,"limit":20}

Trace pages omit nested lists and give their counts. Stream pages preserve exact byte totals but omit contributions, ranges and final-size lists. For those lists select contributions, ranges or final-sizes with all three stream selectors. loss-declarations and incomplete-records need only trace_index. Offsets count rows in that retained facet; defaults are 0 and 64. Missing or ambiguous selectors fail rather than choosing a connection.

These views page already interpreted facts without decoding again or opening the source, and work after its deletion. Facts still cover all recorded events before parent record selection. Their exhaustion does not prove logging or network completeness. Event preview/complete pages remain available through kind capture, with original numeric lexemes and producer ordinals preserved.